[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M795ZSMD85PWAF56Z7PX7W7Y":3,"skill-same:01M795ZSMD85PWAF56Z7PX7W7Y":173},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":43,"commitSha":44,"skillDirectory":45,"forksCount":14,"repositoryHtmlUrl":46,"repositoryHomepageUrl":6,"ownerHtmlUrl":47,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":48,"install":49,"resources":62},"01M795ZSMD85PWAF56Z7PX7W7Y","ecl-harness-engineer",null,"Create or audit ECL Agent Harness infrastructure: AGENTS.md, change tracking, repository guidance, lint checks, CI gates, and agent handoff docs.","Elogic360\u002Fsoftware-factory","Elogic360","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F156535519?v=4","01M74NH3F0RB13PWB7JZDE2V80","elogic360","01M74NH3FPC65082MDZVMFHDM3",0,"MIT","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: ecl-harness-engineer\ndescription: \"Create or audit ECL Agent Harness infrastructure: AGENTS.md, change tracking, repository guidance, lint checks, CI gates, and agent handoff docs.\"\ncategory: development\nrisk: safe\nsource: community\nsource_repo: qinghui316\u002Fecl-harness-engineer\nsource_type: community\ndate_added: \"2026-06-13\"\nauthor: qinghui316\ntags: [codex, agent-harness, ecl, workflow, ci]\ntools: [codex, claude, cursor, gemini, antigravity]\nlicense: MIT\nlicense_source: \"https:\u002F\u002Fgithub.com\u002Fqinghui316\u002Fecl-harness-engineer\u002Fblob\u002Fmain\u002FLICENSE\"\n---\n\n# ECL Harness Engineer\nDesign and create Harness Engineering infrastructure so AI agents can work reliably in a codebase.\n\n> **Core Philosophy**: \"Intelligence without infrastructure is just a demo.\" The Agent Harness is the Operating System — the LLM is just the CPU. The repository becomes the single source of truth — if an agent can't see it in context, it doesn't exist.\n\n## When to Use This Skill\n\n- Use when a repository needs AI-agent collaboration infrastructure such as `AGENTS.md`, `docs\u002FECL.md`, `docs\u002FSTATUS.md`, harness change tracking, or mechanical validation gates.\n- Use when auditing an existing Agent Harness for missing ECL lifecycle docs, change templates, lint checks, environment contracts, or CI integration.\n- Use when converting repeated agent workflow failures into repository-local documentation, tests, lint rules, or lightweight auto-evolution checks.\n- Do not use for ordinary business feature implementation unless the requested work is specifically about creating or improving the repository harness.\n\n## Limitations\n\n- This skill creates or audits harness infrastructure; it does not replace product requirements, implementation planning, code review, or release approval for the target project.\n- The generated ECL docs, linters, scripts, and CI examples must be adapted to the repository's actual stack, security model, and existing contributor workflow before enforcement.\n- Auto-evolve recommendations are guidance only. Apply harness changes through normal review, validation, and rollback discipline instead of accepting them as autonomous policy changes.\n\n## Unified Workflow\n\nThis skill follows a single unified workflow regardless of project state (empty, existing code, or existing harness). The core idea: **detect the gap between current state and target state, then fill it**.\n\nDefault to a **core ECL harness**. Core includes lightweight auto-evolve threshold checking:\nclosed changes are counted, a pending evolution note is generated when the threshold is reached,\nand Codex applies harness improvements only through evidence, validation, scoring, and rollback.\nAdvanced agent-platform capabilities such as eval datasets, execution traces, durable state,\ncheckpoints, long-term memory, and metrics remain optional profiles only when the user explicitly\nasks for agent evaluation, observability, resumable execution, or long-term memory.\n\nThis skill improves the target repository's agent harness. It does **not** implement ordinary\nbusiness features, replace the coding agent's plan mode, or create a separate requirements product.\nPlan mode is useful for live discussion; ECL artifacts are the repository record that later agents,\nlinters, CI, and archive history can inspect.\n\n1. **Quick Detection + Intent Confirmation** — what exists, what already passes, and what the user wants.\n2. **Analysis** — architecture, harness state, environment, and project identity.\n3. **Intake Review + Delta Synthesis** — classify small vs structured work, support requirement-first\n   and plan-first inputs, and compute exactly what to create or update.\n4. **Creation\u002FUpdate** — docs, status handoff, linters, ECL\u002Fchange scripts, environment config, and CI.\n5. **Verification + Handoff** — run checks, attribute failures, update STATUS.md, trigger auto-evolve checks, and summarize results.\n\n---\n\n## Phase 1: Quick Detection + Intent Confirmation\n\n**Goal**: In under 5 minutes, understand project state and user intent.\n\n### 1.1 Project State Detection\n\nRun this quick scan:\n\n```bash\n# Count files\nfile_count=$(find . -type f ! -path '.\u002F.git\u002F*' ! -path '.\u002Fnode_modules\u002F*' ! -path '.\u002Fvendor\u002F*' 2>\u002Fdev\u002Fnull | wc -l)\ncode_files=$(find . -type f \\( -name \"*.go\" -o -name \"*.ts\" -o -name \"*.js\" -o -name \"*.py\" -o -name \"*.rs\" \\) ! -path '.\u002F.git\u002F*' ! -path '.\u002Fnode_modules\u002F*' ! -path '.\u002Fvendor\u002F*' 2>\u002Fdev\u002Fnull | wc -l)\n\n# Check harness components\nhas_agents_md=$(test -f AGENTS.md && echo \"yes\" || echo \"no\")\nhas_architecture=$(test -f docs\u002FARCHITECTURE.md && echo \"yes\" || echo \"no\")\nhas_linters=$(ls scripts\u002Flint-* 2>\u002Fdev\u002Fnull | wc -l)\nhas_harness_dir=$(test -d harness && echo \"yes\" || echo \"no\")\nhas_ecl_doc=$(test -f docs\u002FECL.md && echo \"yes\" || echo \"no\")\nhas_changes_dir=$(test -d harness\u002Fchanges && echo \"yes\" || echo \"no\")\nhas_change_templates=$(test -d harness\u002Ftemplates\u002Fchange && echo \"yes\" || echo \"no\")\nhas_change_script=$(ls scripts\u002Fharness-change.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_evolve_script=$(ls scripts\u002Fharness-evolve.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_ecl_lint=$(ls scripts\u002Flint-ecl.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_encoding_lint=$(ls scripts\u002Flint-encoding.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_makefile=$(test -f Makefile && echo \"yes\" || echo \"no\")\nhas_package_json=$(test -f package.json && echo \"yes\" || echo \"no\")\n\n# Detect tech stack\nif test -f go.mod; then TECH=\"Go\"\nelif test -f package.json; then TECH=\"TypeScript\u002FNode.js\"\nelif test -f requirements.txt || test -f pyproject.toml; then TECH=\"Python\"\nelse TECH=\"Unknown\"\nfi\n```\n\n### 1.2 Classify Project State\n\nBased on detection:\n\n| State | Criteria | Action |\n|-------|----------|--------|\n| **Empty** | file_count \u003C 5 AND code_files = 0 | Guide user through project choices first |\n| **Code Only** | code_files > 0 AND has_agents_md = \"no\" | Full analysis + core harness creation |\n| **Partial Harness** | has_agents_md = \"yes\" AND (has_linters = 0 OR has_harness_dir = \"no\") | Gap analysis + fill gaps |\n| **Harness Present** | Core harness components exist | Audit + improvement suggestions |\n\nAlso classify ECL readiness:\n\n| ECL State | Criteria | Action |\n|-----------|----------|--------|\n| **ECL Missing** | has_ecl_doc = \"no\" OR has_changes_dir = \"no\" | Create ECL docs, change templates, and scripts |\n| **ECL Partial** | ECL doc exists but scripts\u002Ftemplates missing | Fill ECL automation gaps |\n| **ECL Ready** | docs\u002FECL.md, harness\u002Fchanges, templates, harness-change, harness-evolve, lint-ecl, lint-encoding exist | Audit index freshness and workflow quality |\n\n### 1.3 Baseline Verification Snapshot\n\nFor existing projects, capture a best-effort baseline before creating or updating harness files.\nThe baseline is for attribution only: it distinguishes pre-existing project failures from\nfailures introduced by harness work. It must not be used to weaken default CI.\n\nRun only commands that already exist in the project:\n\n| Ecosystem | Baseline commands |\n|-----------|-------------------|\n| TypeScript\u002FNode.js | package scripts such as `lint`, `typecheck`, `test`, `build`; include nested package build scripts when detected |\n| Go | `go test .\u002F...`, `go build .\u002F...`, existing `make lint` or `make test` |\n| Python | existing test\u002Flint scripts, `python -m compileall .` |\n\nRecord each command as `pass`, `fail`, or `missing`, with the short failure reason. If a command\nfails before harness creation, report it later as **pre-existing project debt**, not as harness\nfailure. Default CI remains strict and should still include normal business gates unless the user\nexplicitly asks for a temporary staged rollout.\n\n### 1.4 Intent Confirmation\n\nBefore planning changes, classify requested scope:\n\n| Scope | Default? | Includes |\n|-------|----------|----------|\n| **Core harness** | Yes | AGENTS.md, docs\u002FECL.md, docs\u002FSTATUS.md, docs, ECL changes, lightweight auto-evolve, linters, environment contract, CI |\n| **Advanced harness** | No | Core harness plus explicitly requested eval, trace, state, checkpoints, memory, or metrics |\n| **Documentation only** | No | AGENTS.md and docs without linters, scripts, or CI |\n\nWhen a user-confirmation tool is available, confirm scope. In Codex, use `request_user_input`.\nOn other platforms, use the equivalent user-choice tool. If no such tool is available, use the\ndetected context and record assumptions.\n\n```json\n{\n  \"question\": \"What's your priority for this harness setup?\",\n  \"header\": \"Scope\",\n  \"multiSelect\": false,\n  \"options\": [\n    {\n      \"label\": \"Core harness (Recommended)\",\n      \"description\": \"Project-first AGENTS.md, ECL changes, STATUS handoff, auto-evolve threshold checks, linters, environment contract, and strict CI\"\n    },\n    {\n      \"label\": \"Advanced harness\",\n      \"description\": \"Core harness plus explicitly requested eval, trace, memory, checkpoint, or metrics infrastructure\"\n    },\n    {\n      \"label\": \"Documentation only\",\n      \"description\": \"AGENTS.md and project docs only; skip linters, scripts, and CI for now\"\n    }\n  ]\n}\n```\n\n**If Empty project**, also ask for basics:\n\n```json\n{\n  \"question\": \"What tech stack for this project?\",\n  \"header\": \"Tech Stack\",\n  \"multiSelect\": false,\n  \"options\": [\n    {\"label\": \"Go\", \"description\": \"CLI tools, high-performance services, system programming\"},\n    {\"label\": \"TypeScript\u002FNode.js\", \"description\": \"Web APIs, full-stack apps, rapid prototyping\"},\n    {\"label\": \"Python\", \"description\": \"Data processing, ML\u002FAI, scripting\"}\n  ]\n}\n```\n\nIf no user-confirmation tool is available, use detected values and document assumptions:\n\n```markdown\n## Auto-Detected Context\n\n| Field | Value | Confidence | Evidence |\n|-------|-------|------------|----------|\n| Tech Stack | {TECH} | High | Found {config file} |\n| Project State | {state} | High | {criteria matched} |\n| Scope | Core harness | Default | No user preference specified |\n\nProceeding with these assumptions. Tell me if any need adjustment.\n```\n\n### 1.5 ECL Work Intake Rules\n\nWhen generating ECL guidance for a target project, keep the process small enough to use:\n\n| Intake type | Criteria | Required ECL handling |\n|-------------|----------|-----------------------|\n| **Small Change** | Local, low-risk edits such as copy, comments, style-only tweaks, or single-file bug fixes with no interface, data, permission, architecture, or release impact | Active change optional; still record the verification command in the final response or existing task notes |\n| **Structured Change** | Cross-file\u002Fmodule behavior, APIs, data model, permissions, architecture, validation chain, unclear requirements, or work likely to exceed 20 minutes | Use active change files and require intake\u002Fspec\u002Fplan review before implementation |\n\nDecision tree:\n\n1. If an active change already exists, keep using it; do not create a second active context.\n2. If the change is copy, comments, README text, formatting, or an obviously local single-file fix\n   with no runtime, API, data, permission, architecture, or validation-chain impact, treat it as\n   Small Change.\n3. If the change touches APIs, data, permissions, architecture, multiple modules, release\u002Fruntime\n   behavior, or unclear requirements, treat it as Structured Change.\n4. If impact is unclear, do read-only investigation first. If uncertainty remains after inspection,\n   ask one high-impact question or upgrade to Structured Change; do not assume Small Change.\n\nFor structured changes, support both common entry points:\n\n- **Requirement-first input**: extract target users\u002Fscenarios, evidence, success criteria,\n  acceptance criteria, non-goals, constraints, assumptions, and risks into `spec.md`.\n- **Plan-first input**: treat the user's plan as a draft, split WHAT\u002FWHY into `spec.md` and HOW into\n  `plan.md`, then ask only about high-impact gaps that affect implementation direction or acceptance.\n  If the plan is complete and does not conflict with repository evidence, do not repeat a full\n  interview. If it conflicts with code, docs, commands, or existing harness constraints, record the\n  conflict and return to Intake Review.\n\nQuestions are allowed and expected, but must be bounded: ask at most three high-impact questions per\nround. Low-risk unknowns become assumptions; high-impact unknowns become\n`[NEEDS CLARIFICATION: ...]` and block implementation until resolved.\n\nFor complex structured changes, use a lightweight iteration loop rather than treating the first\nspec as final:\n\n```text\nDraft Spec -> Draft Plan -> Review Gaps -> Revise Spec\u002FPlan -> Gate -> Tasks\n```\n\nDefault to at most two loops. If key gaps remain, continue up to five loops; after that, record a\nblocker instead of implementing from guesses. `plan.md` must include any planning-discovered spec\ngaps, because plans often expose missing acceptance, boundary, permission, data, or validation\nrequirements.\n\n---\n\n## Phase 2: Analysis\n\n**Goal**: Deeply understand codebase architecture, harness state, and environment requirements.\n\n### 2.1 Execution Mode\n\nUse subagents only when the user authorized delegation and the environment supports it. Otherwise, execute the same responsibilities inline.\n\nIf using subagents, assign:\n\n- Code architecture analysis: follow `agents\u002Fanalyzer.md`; output `harness\u002F.analysis\u002Farchitecture.json`.\n- Harness state audit: follow `agents\u002Fauditor.md`; output `harness\u002F.analysis\u002Faudit.json`.\n- Environment analysis: follow `references\u002Fenvironment-detection-guide.md`; output `harness\u002F.analysis\u002Fenvironment.json`.\n\nIf working inline, produce the same three analysis artifacts or equivalent in-memory summaries before Phase 3.\n\n### 2.2 Project Identity Extraction\n\nFor existing projects, extract target-project meaning before writing docs:\n- One-sentence project identity: what it does and for whom.\n- Core workflow or domain model: user\u002Fsystem flow, key entities, API resources, jobs, or commands.\n- Primary source entrypoints and where common changes belong.\n\nUse `README.md`, manifests, entrypoints, routes\u002Fcontrollers, schemas\u002Fmodels, and key source\ndirectories. Harness files are not sufficient evidence for project identity.\n\n### 2.3 Adapter Selection\n\nAfter detecting the tech stack, load the matching adapter before creating linters, scripts, CI,\nor environment config. Adapter guidance overrides generic templates for language-specific details.\n\n| Detected stack | Required adapter |\n|----------------|------------------|\n| TypeScript\u002FNode.js | `references\u002Fadapters\u002Ftypescript.md` |\n| Go | `references\u002Fadapters\u002Fgo.md` |\n| Python | `references\u002Fadapters\u002Fpython.md` |\n| Rust | `references\u002Fadapters\u002Frust.md` |\n| Java | `references\u002Fadapters\u002Fjava.md` |\n| Unknown\u002Fmixed | `references\u002Fadapters\u002Fgeneric.md` plus any detected language adapters |\n\nFor TypeScript\u002FNode.js projects, prefer Node\u002FTS-native outputs: `scripts\u002Flint-deps.mjs` or\nequivalent, `scripts\u002Flint-quality.mjs`, npm\u002Fpackage-manager scripts, and Node\u002FTS GitHub Actions.\nDo not adapt Go linter or Makefile-only patterns to TypeScript unless the project is actually Go\nor already uses Makefile as the primary command surface.\n\n### 2.4 Command Surface Selection\n\nBefore creating ECL scripts, select the target project's command surface. Do not assume\nPowerShell is the only Windows option. This selection is normally automatic; do not ask the user to\nchoose a script format unless project evidence conflicts or the user has already expressed a hard\nconstraint.\n\nPriority:\n\n1. Existing project entrypoints: package-manager scripts, Makefile targets, README commands,\n   or CI shell conventions.\n2. Explicit user\u002Fproject constraints. If the project rejects `.ps1`, do not generate PowerShell\n   as the only harness entrypoint.\n3. Bash profile when allowed. For Windows projects that accept Bash, generate `.sh` scripts and\n   document the prerequisite: Git Bash, WSL, MSYS2, or a CI Linux runner.\n4. PowerShell profile when the project accepts Windows-native PowerShell. Keep it compatible with\n   Windows PowerShell 5.1 and PowerShell 7.\n5. Node or Python profiles when those runtimes are already first-class project dependencies.\n\nDefault when evidence is sparse: for TypeScript\u002FNode projects choose Node\u002Fpackage-manager scripts;\nfor Windows projects that allow Bash choose Bash profile and document Git Bash\u002FWSL\u002FMSYS2; otherwise\nchoose the adapter's native lightweight scripting profile.\n\nAll profiles must implement the same ECL invariants and command set. `harness-change`,\n`harness-evolve`, `lint-ecl`, and `lint-encoding` may be implemented as `.ps1`, `.sh`, `.mjs`,\nor `.py`, but docs, CI, Makefile\u002Fpackage scripts, and verification commands must use the chosen\nentrypoint consistently.\n\n### 2.5 Wait for Analysis Completion\n\nWhen subagents are running, wait for their final reports. While waiting, you can:\n- Review any existing documentation\n- Prepare templates for Phase 4\n\n### 2.5 For Empty Projects\n\nSkip Phase 2 analysis agents. Instead:\n- Use templates from `references\u002Fgreenfield-templates.md`\n- Base decisions on user's tech stack choice\n- Design a standard 3-layer architecture\n\n---\n\n## Phase 3: Delta Synthesis\n\n**Goal**: Merge analysis results and compute exactly what needs to be created\u002Fupdated.\n\n### 3.1 Read Analysis Results\n\n```bash\ncat harness\u002F.analysis\u002Farchitecture.json\ncat harness\u002F.analysis\u002Faudit.json\ncat harness\u002F.analysis\u002Fenvironment.json\n```\n\n### 3.2 Compute Delta\n\nCreate a delta list:\n\n```markdown\n## Delta: What Needs to Be Done\n\n### Core To Create (doesn't exist)\n- [ ] AGENTS.md\n- [ ] docs\u002FECL.md\n- [ ] docs\u002FSTATUS.md\n- [ ] docs\u002FARCHITECTURE.md\n- [ ] scripts\u002Flint-deps.go\n- [ ] scripts\u002Fharness-change.{ps1|sh|mjs|py}\n- [ ] scripts\u002Fharness-evolve.{ps1|sh|mjs|py}\n- [ ] scripts\u002Flint-ecl.{ps1|sh|mjs|py}\n- [ ] scripts\u002Flint-encoding.{ps1|sh|mjs|py}\n- [ ] harness\u002Fchanges\u002F{active,parking,archive}\n- [ ] harness\u002Ftemplates\u002Fchange\u002F\n- [ ] harness\u002Fconfig\u002Fenvironment.json\n- [ ] harness\u002Fevolution\u002F{state.json,results.tsv,proposals\u002F} (`pending.md` is generated later only when the archive threshold is reached)\n\n### Optional Advanced (only if explicitly requested)\n- [ ] harness\u002Feval\u002F — agent evaluation datasets and runner inputs\n- [ ] harness\u002Ftrace\u002F — execution traces for agent runs\n- [ ] harness\u002Fstate\u002F — executor runtime state\n- [ ] harness\u002Fcheckpoints\u002F — resumable execution checkpoints\n- [ ] harness\u002Fmemory\u002F — long-term agent memory experiments\n- [ ] harness\u002Fmetrics\u002F — execution and quality metrics\n\n### To Update (exists but has gaps)\n- [ ] docs\u002FDEVELOPMENT.md — missing build commands\n- [ ] scripts\u002Flint-quality.py — missing 3 packages in layer map\n\n### Already Good (no changes needed)\n- [x] Makefile — has all required targets\n- [x] .github\u002Fworkflows\u002Fci.yml — properly configured\n```\n\n### 3.3 Confirm with User (if confirmation tool is available)\n\nFor significant changes:\n\n```json\n{\n  \"question\": \"I've analyzed the codebase. Ready to proceed with these changes?\",\n  \"header\": \"Confirm\",\n  \"multiSelect\": false,\n  \"options\": [\n    {\"label\": \"Yes, proceed with all\", \"description\": \"Create\u002Fupdate all identified items\"},\n    {\"label\": \"Show me the details first\", \"description\": \"I'll explain what each change involves\"},\n    {\"label\": \"Only critical items\", \"description\": \"Just P0\u002FP1 items, skip P2\u002FP3 for now\"}\n  ]\n}\n```\n\n---\n\n## Phase 4: Creation\u002FUpdate\n\n**Goal**: Create or update all harness files from the delta.\n\n### 4.1 Execution Mode\n\nUse subagents only when authorized and available. Otherwise, perform the same work inline. Keep write scopes disjoint if using parallel workers.\n\nCreation responsibilities:\n\n- Documentation: follow `agents\u002Fcreator-docs.md`; create\u002Fupdate AGENTS.md, docs\u002FECL.md, docs\u002FSTATUS.md, docs\u002FARCHITECTURE.md, docs\u002FDEVELOPMENT.md, and design docs. AGENTS.md is the target project's entry map, not a harness creation record. Keep the first screen project-first, but preserve ECL\u002Fcurrent-change priority in context loading: `AGENTS.md` -> `docs\u002FECL.md` -> active change if present -> auto-evolve pending if present -> otherwise `docs\u002FSTATUS.md` -> task-specific project docs.\n- Linters: follow `agents\u002Fcreator-linters.md`; create\u002Fupdate dependency, quality, ECL, and encoding checks.\n- Config and scripts: follow `agents\u002Fcreator-config.md`; create\u002Fupdate environment contract, harness scripts, changes directories\u002Ftemplates, lightweight evolution state, harness-change, harness-evolve, Makefile targets, and CI. Create advanced directories only when the confirmed scope requires them.\n\nECL change templates must include `summary.md`, `spec.md`, `plan.md`, `tasks.md`, and\n`reviews\u002Freview.md`. `spec.md` captures WHAT\u002FWHY, `plan.md` captures HOW and planning-discovered\nspec gaps, and `tasks.md` is generated only after the spec\u002Fplan gate is ready enough for\nimplementation. Do not require old archived changes to contain `plan.md`; compatibility applies to\nhistory.\n\nImportant: do not create static verification config such as `harness\u002Fconfig\u002Fverify.json`. Verification plans are generated at runtime by the executor from `environment.json` and the task context.\n\nStrict CI rule: default CI must include normal business quality gates (`lint`, `typecheck`, `test`,\n`build`, and backend\u002Fpackage-specific equivalents when available) plus harness checks. Do not remove\nor skip business gates because the baseline is red. If the baseline was already red, explain that CI\nwill be red until the pre-existing project issues are fixed. Generate staged or relaxed CI only when\nthe user explicitly asks for it.\n\nCommand surface rule: create ECL scripts for the selected profile, not a hardcoded shell. If Bash is\nselected on Windows, document Git Bash, WSL, MSYS2, or CI Linux shell requirements in the generated\nenvironment\u002Fdevelopment docs. If PowerShell is selected, detect whether `pwsh` is available; if not,\nuse `powershell -NoProfile -ExecutionPolicy Bypass`. PowerShell templates must be compatible with\nWindows PowerShell 5.1: avoid ambiguous overloads such as `TrimStart(\".\\\")`, and avoid non-ASCII\nmojibake marker string literals in `.ps1`; represent markers by Unicode codepoint or another\nPS5-safe construction.\n\n### 4.2 For Empty Projects: Also Create Business Code Plan\n\nFor empty projects, add one more agent:\n\n```\nAgent(\"create-exec-plan\", prompt=\"\"\"\nCreate execution plan for business code (harness-executor will implement this):\n\nTech stack: {TECH}\nProject type: {from user choice}\nArchitecture: 3-layer (Types → Core → Entry Points)\n\nCreate: docs\u002Fexec-plans\u002Factive\u002Fbootstrap-code.md\n\nContents:\n- Full source code for initial project structure\n- main.go\u002Findex.ts\u002Fmain.py entry point\n- Basic types and core logic\n- Test files\n\nThis is for harness-executor to implement — not ecl-harness-engineer's responsibility.\n\"\"\")\n```\n\n### 4.3 Wait for Creation Completion\n\nAgents will notify when done. Collect any issues they encountered.\n\n---\n\n## Phase 5: Verification + Handoff\n\n**Goal**: Ensure everything works, then hand off or present results.\n\n### 5.1 Run Verification\n\n```bash\n# 0. Compare against the baseline snapshot\n# Re-run the same existing lint\u002Ftypecheck\u002Ftest\u002Fbuild commands captured in Phase 1.\n\n# 1. Harness checks pass\nmake verify-harness || npm run lint:harness || {generated_harness_lint_command}\n\n# 2. Architecture linters pass\nmake lint-arch || npm run lint:arch\n\n# 3. Business build\u002Ftest gates run\ngo build .\u002F... || npm run build || python -m compileall .\n\n# 4. AGENTS.md size check\nwc -l AGENTS.md  # Should be 80-120 lines\n\n# 4b. AGENTS.md content gate\n# Confirm it explains project identity, core workflow\u002Fdomain model, source entrypoints,\n# task-based verification, active-change-before-STATUS loading, and contains no\n# ECL Harness Engineer internal boundary language.\n\n# 5. All expected files exist\ntest -f AGENTS.md && echo \"✓ AGENTS.md\"\ntest -f docs\u002FARCHITECTURE.md && echo \"✓ ARCHITECTURE.md\"\ntest -f docs\u002FECL.md && echo \"✓ ECL.md\"\ntest -f docs\u002FSTATUS.md && echo \"✓ STATUS.md\"\ntest -f scripts\u002Flint-deps* && echo \"✓ lint-deps\"\ntest -f scripts\u002Fharness-change.* && echo \"✓ harness-change\"\ntest -f scripts\u002Flint-ecl.* && echo \"✓ lint-ecl\"\ntest -f scripts\u002Fharness-evolve.* && echo \"✓ harness-evolve\"\ntest -d harness\u002F && echo \"✓ harness\u002F\"\ntest -d harness\u002Fchanges && echo \"✓ harness\u002Fchanges\"\ntest -f harness\u002Fevolution\u002Fstate.json && echo \"✓ evolution state\"\n\n# 6. Design docs exist (not just index)\nfind docs\u002Fdesign-docs -name \"*.md\" ! -name \"index.md\" | wc -l\n```\n\nClassify every verification result:\n\n| Classification | Meaning |\n|----------------|---------|\n| Harness pass | Harness-created checks\u002Ffiles\u002Fscripts work |\n| Pre-existing project failure | The same command failed in the Phase 1 baseline |\n| New regression | The command passed in Phase 1 and fails after harness creation |\n| Not available | The command\u002Fscript does not exist in this project |\n\nAGENTS.md content gate:\n- A new agent can tell what the project does within 30 seconds.\n- The core product\u002Fsystem workflow or domain model is visible.\n- Main source entrypoints and task-to-directory mapping are visible.\n- Verification guidance maps to task type.\n- Context loading reads `docs\u002FECL.md` first, then active change when present.\n- If no active change exists and `harness\u002Fevolution\u002Fpending.md` exists, read it before\n  `docs\u002FSTATUS.md`, mention it as pending maintenance, and ask whether to handle it now unless the\n  user already prioritized the current task. Reading or asking does not start auto-evolve and must\n  not block ordinary user work.\n- If no active change exists and no pending evolution exists, context loading reads `docs\u002FSTATUS.md` before task-specific project docs.\n- For structured work, `docs\u002FECL.md` explains Small Change vs Structured Change, bounded Intake\n  Review, plan-first input handling, and the spec\u002Fplan review gate.\n- Archive history is loaded selectively through `docs\u002FSTATUS.md` paths or `harness\u002Fchanges\u002FINDEX.json`, starting with historical `summary.md` only.\n- No skill-internal boundary leaks, such as sections or sentences that describe this skill's own scope limits as target-project rules.\n\n### 5.2 STATUS.md Handoff Update\n\nWhen a target project uses ECL changes, maintain `docs\u002FSTATUS.md` as a lightweight handoff file.\nIt is not the authority while an active change exists, but it becomes the default recent-history\nentry point after the active change is closed.\n\nClose-change handoff protocol:\n\n1. Before running `harness-change close`, read the active change `summary.md`, `spec.md`,\n   `plan.md`, `tasks.md`, and relevant `reviews\u002F`; update `docs\u002FSTATUS.md` with completed work,\n   verification results, residual risks, and the next recommended resume point.\n2. Run the close command so the active change moves to `harness\u002Fchanges\u002Farchive\u002F...` and\n   `harness\u002Fchanges\u002FINDEX.json` is rebuilt.\n3. After close, update `docs\u002FSTATUS.md` again with the final archive path, normally pointing to\n   the archived `summary.md`.\n4. Run the harness lint command (`npm run lint:harness`, `make verify-harness`, or the generated\n   ECL lint command) to confirm STATUS, ECL structure, and INDEX state are consistent.\n\nHooks and CI may validate `docs\u002FSTATUS.md`, but must not auto-write it or move changes.\n\n### 5.3 Auto-Evolve Check\n\nCore harnesses include lightweight auto-evolve by default. The script layer only detects when\nenough new archive evidence exists and writes `harness\u002Fevolution\u002Fpending.md`; Codex performs the\nsemantic improvement pass.\n\nTrigger model: `harness-change close` and `reindex` run `harness-evolve check`; `new` only reminds\nwhen pending exists. Hooks and CI may warn, but must not modify docs, scripts, STATUS, or changes.\nGenerated scripts do not call subagents. They only count archive evidence and create pending\ncontext. When no active change exists and Codex notices pending maintenance, it should ask the user\nwhether to handle it now unless the user already prioritized the current task. Asking does not start\npending evolution.\n\n`harness\u002Fevolution\u002Fpending.md` is a maintenance reminder, not a hard lock. Reading it for context\ndoes not start pending evolution. Pending evolution starts only when Codex creates or uses an\n`auto-evolve-harness-*` change, writes an evolution proposal\u002Fresult, or edits Harness files based\non the pending evidence. Once started, finish with a proposal, one `harness\u002Fevolution\u002Fresults.tsv`\nrow, and `harness-evolve mark-complete`; otherwise park or close blocked, not completed.\n\nApply only the smallest evidence-backed delta that passes review. No independent scorer =\nno auto-apply: user approval to handle pending implies permission to request an independent\nauditor\u002Fsubagent when the environment supports it. If the environment still requires explicit\nauthorization, ask once. If scoring is unavailable, declined, or still unauthorized after asking,\nrecord `noop` with `eval_mode=dry_run`, keep the proposal, run `mark-complete`, and stop.\nMachinery repair\n(`harness-evolve`, pending templates, lint) does not complete pending evolution by itself; after\nrepair, still evaluate candidate archives or leave the work parked\u002Fblocked.\n\nDetailed proposal format, scoring weights, status values, and complexity budget live in\n`references\u002Fecl-harness.md`.\n\n### 5.4 Present Summary\n\n```markdown\n## Harness Infrastructure Complete\n\n**Project**: {project-name}\n**Tech Stack**: {TECH}\n**Files Created\u002FUpdated**: {count}\n\n### Created Files\n- AGENTS.md ({N} lines)\n- docs\u002FARCHITECTURE.md\n- docs\u002FECL.md\n- docs\u002FSTATUS.md\n- docs\u002FDEVELOPMENT.md\n- docs\u002Fdesign-docs\u002F{component}.md\n- scripts\u002Flint-deps.{ext}\n- scripts\u002Flint-quality.{ext}\n- scripts\u002Fharness-change.{ps1|sh|mjs|py}\n- scripts\u002Flint-ecl.{ps1|sh|mjs|py}\n- scripts\u002Flint-encoding.{ps1|sh|mjs|py}\n- scripts\u002Fharness-evolve.{ps1|sh|mjs|py}\n- harness\u002Fconfig\u002Fenvironment.json\n- harness\u002Fchanges\u002F\n- harness\u002Fevolution\u002F\n- harness\u002Ftemplates\u002Fchange\u002F\n- Makefile\n\n### Verification Results\n- Harness checks: ✓\n- Architecture checks: ✓\n- Business gates: ✓ or pre-existing failures listed below\n- AGENTS.md size: ✓ ({N} lines)\n\n### Pre-existing Project Failures\n- {List baseline-red commands and short reasons, or \"None observed.\"}\n\n### New Regressions Introduced By Harness\n- {List commands that passed before and failed after, or \"None observed.\"}\n\n### Next Steps\n{For empty projects: \"Run harness-executor to implement business code from docs\u002Fexec-plans\u002Factive\u002Fbootstrap-code.md\"}\n{For existing projects: \"The harness is ready. AI agents can now use AGENTS.md as their entry point.\"}\n```\n\n### 5.5 Automatic Handoff (for Empty Projects)\n\nIf this was an empty project with a bootstrap exec-plan, invoke harness-executor:\n\n```\nSkill(skill=\"harness-executor\")\n```\n\nWith context: \"Implement the bootstrap exec-plan at docs\u002Fexec-plans\u002Factive\u002Fbootstrap-code.md\"\n\n---\n\n## Core Principles\n\n### 1. Repository as Single Source of Truth\n\nAgents cannot access Slack, Google Docs, or tribal knowledge. If it's not in the repository, it doesn't exist for the agent.\n\n### 2. AGENTS.md is a Map, Not a Manual\n\nKeep it 80-120 lines. Link to detailed docs, don't embed them.\n\n### 3. Enforce Invariants Mechanically\n\nLinter errors must be agent-actionable:\n```\n✗ BAD: \"Forbidden import in core\u002Ftypes\u002Fuser.go\"\n\n✓ GOOD: \"core\u002Ftypes\u002Fuser.go:15 imports core\u002Fconfig (layer 0 → layer 2).\n         Layer 0 packages must have NO internal dependencies.\n\n         Fix options:\n         1. Move config-dependent logic to a higher layer\n         2. Pass the config value as a parameter\n         3. Use dependency injection via an interface\"\n```\n\n### 4. Build to Delete\n\nEvery component should be replaceable. Capabilities that required complex pipelines yesterday may be single prompts tomorrow.\n\n### 5. Start Simple\n\nAtomic, well-documented tools > complex agent choreography. Don't over-engineer.\n\n### 6. Change State Is Explicit\n\nUse a single `harness\u002Fchanges\u002Factive\u002F` task for personal development. Move paused work to `parking\u002F` and closed work to `archive\u002F` with the generated `scripts\u002Fharness-change.*` command. Maintain `docs\u002FSTATUS.md` as the soft handoff summary after active work is closed. Never hand-edit `harness\u002Fchanges\u002FINDEX.json`; it is a generated index rebuilt by `park`, `close`, `resume`, and `reindex`. Structured changes use `spec.md` for WHAT\u002FWHY, `plan.md` for HOW, and `tasks.md` for executable work.\n\n### 7. Harness Evolves From Evidence\n\nEvery few closed changes, the generated `scripts\u002Fharness-evolve.* check` command may create\n`harness\u002Fevolution\u002Fpending.md`. Treat it as a maintenance reminder to improve harness rules from\nreal archived evidence, not as a hard blocker for unrelated user work. If you start acting on the\npending evidence, first refresh `harness\u002Fchanges\u002FINDEX.json` and use the current eligible archive\nwindow; the Candidate Archives in an old pending file are a trigger snapshot, not the only evidence.\nThen finish with proposal + results.tsv + `mark-complete`, or park\u002Fblock the work.\nDo not turn one-off business bugs into permanent process. Keep only changes that improve the audit\nscore and pass validation.\n\n---\n\n## Reference Files\n\n| File | When to Read | Contents |\n|------|-------------|----------|\n| `references\u002Fgreenfield-templates.md` | Empty projects (Phase 2.5) | Complete Go\u002FTS\u002FPython scaffolding |\n| `references\u002Fdocumentation-templates.md` | Phase 4 doc creation | Doc templates with numbered sections |\n| `references\u002Flinter-templates.md` | Phase 4 linter creation | Linter code templates per language |\n| `references\u002Fecl-harness.md` | ECL-aware harness creation | docs\u002FECL.md, docs\u002FSTATUS.md, change lifecycle, INDEX.json, PowerShell script templates |\n| `references\u002Fdarwin-eval-prompts.md` | Skill quality evaluation | Dry-run prompts for darwin-skill review |\n| `references\u002Fenvironment-detection-guide.md` | Phase 2 env analysis | Environment ecosystem detection |\n| `references\u002Fenvironment-config-guide.md` | Phase 4 config creation | Startup, services, env vars, user-confirmation templates |\n| `references\u002Fadapters\u002Ftypescript.md` | TypeScript\u002FNode.js projects | npm scripts, Node linters, package-manager detection, CI defaults |\n| `references\u002Fadapters\u002F{go,python,rust,java,generic}.md` | Matching detected stacks | Language-specific commands and conventions |\n\nAgent prompts for Phase 2 and Phase 4 subagents are in `agents\u002F`.\n\nFor small projects (\u003C 20 files) or when subagents aren't available, execute phases inline instead of spawning agents.\n","# ECL Harness Engineer\nDesign and create Harness Engineering infrastructure so AI agents can work reliably in a codebase.\n\n> **Core Philosophy**: \"Intelligence without infrastructure is just a demo.\" The Agent Harness is the Operating System — the LLM is just the CPU. The repository becomes the single source of truth — if an agent can't see it in context, it doesn't exist.\n\n## When to Use This Skill\n\n- Use when a repository needs AI-agent collaboration infrastructure such as `AGENTS.md`, `docs\u002FECL.md`, `docs\u002FSTATUS.md`, harness change tracking, or mechanical validation gates.\n- Use when auditing an existing Agent Harness for missing ECL lifecycle docs, change templates, lint checks, environment contracts, or CI integration.\n- Use when converting repeated agent workflow failures into repository-local documentation, tests, lint rules, or lightweight auto-evolution checks.\n- Do not use for ordinary business feature implementation unless the requested work is specifically about creating or improving the repository harness.\n\n## Limitations\n\n- This skill creates or audits harness infrastructure; it does not replace product requirements, implementation planning, code review, or release approval for the target project.\n- The generated ECL docs, linters, scripts, and CI examples must be adapted to the repository's actual stack, security model, and existing contributor workflow before enforcement.\n- Auto-evolve recommendations are guidance only. Apply harness changes through normal review, validation, and rollback discipline instead of accepting them as autonomous policy changes.\n\n## Unified Workflow\n\nThis skill follows a single unified workflow regardless of project state (empty, existing code, or existing harness). The core idea: **detect the gap between current state and target state, then fill it**.\n\nDefault to a **core ECL harness**. Core includes lightweight auto-evolve threshold checking:\nclosed changes are counted, a pending evolution note is generated when the threshold is reached,\nand Codex applies harness improvements only through evidence, validation, scoring, and rollback.\nAdvanced agent-platform capabilities such as eval datasets, execution traces, durable state,\ncheckpoints, long-term memory, and metrics remain optional profiles only when the user explicitly\nasks for agent evaluation, observability, resumable execution, or long-term memory.\n\nThis skill improves the target repository's agent harness. It does **not** implement ordinary\nbusiness features, replace the coding agent's plan mode, or create a separate requirements product.\nPlan mode is useful for live discussion; ECL artifacts are the repository record that later agents,\nlinters, CI, and archive history can inspect.\n\n1. **Quick Detection + Intent Confirmation** — what exists, what already passes, and what the user wants.\n2. **Analysis** — architecture, harness state, environment, and project identity.\n3. **Intake Review + Delta Synthesis** — classify small vs structured work, support requirement-first\n   and plan-first inputs, and compute exactly what to create or update.\n4. **Creation\u002FUpdate** — docs, status handoff, linters, ECL\u002Fchange scripts, environment config, and CI.\n5. **Verification + Handoff** — run checks, attribute failures, update STATUS.md, trigger auto-evolve checks, and summarize results.\n\n---\n\n## Phase 1: Quick Detection + Intent Confirmation\n\n**Goal**: In under 5 minutes, understand project state and user intent.\n\n### 1.1 Project State Detection\n\nRun this quick scan:\n\n```bash\n# Count files\nfile_count=$(find . -type f ! -path '.\u002F.git\u002F*' ! -path '.\u002Fnode_modules\u002F*' ! -path '.\u002Fvendor\u002F*' 2>\u002Fdev\u002Fnull | wc -l)\ncode_files=$(find . -type f \\( -name \"*.go\" -o -name \"*.ts\" -o -name \"*.js\" -o -name \"*.py\" -o -name \"*.rs\" \\) ! -path '.\u002F.git\u002F*' ! -path '.\u002Fnode_modules\u002F*' ! -path '.\u002Fvendor\u002F*' 2>\u002Fdev\u002Fnull | wc -l)\n\n# Check harness components\nhas_agents_md=$(test -f AGENTS.md && echo \"yes\" || echo \"no\")\nhas_architecture=$(test -f docs\u002FARCHITECTURE.md && echo \"yes\" || echo \"no\")\nhas_linters=$(ls scripts\u002Flint-* 2>\u002Fdev\u002Fnull | wc -l)\nhas_harness_dir=$(test -d harness && echo \"yes\" || echo \"no\")\nhas_ecl_doc=$(test -f docs\u002FECL.md && echo \"yes\" || echo \"no\")\nhas_changes_dir=$(test -d harness\u002Fchanges && echo \"yes\" || echo \"no\")\nhas_change_templates=$(test -d harness\u002Ftemplates\u002Fchange && echo \"yes\" || echo \"no\")\nhas_change_script=$(ls scripts\u002Fharness-change.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_evolve_script=$(ls scripts\u002Fharness-evolve.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_ecl_lint=$(ls scripts\u002Flint-ecl.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_encoding_lint=$(ls scripts\u002Flint-encoding.* 2>\u002Fdev\u002Fnull | wc -l)\nhas_makefile=$(test -f Makefile && echo \"yes\" || echo \"no\")\nhas_package_json=$(test -f package.json && echo \"yes\" || echo \"no\")\n\n# Detect tech stack\nif test -f go.mod; then TECH=\"Go\"\nelif test -f package.json; then TECH=\"TypeScript\u002FNode.js\"\nelif test -f requirements.txt || test -f pyproject.toml; then TECH=\"Python\"\nelse TECH=\"Unknown\"\nfi\n```\n\n### 1.2 Classify Project State\n\nBased on detection:\n\n| State | Criteria | Action |\n|-------|----------|--------|\n| **Empty** | file_count \u003C 5 AND code_files = 0 | Guide user through project choices first |\n| **Code Only** | code_files > 0 AND has_agents_md = \"no\" | Full analysis + core harness creation |\n| **Partial Harness** | has_agents_md = \"yes\" AND (has_linters = 0 OR has_harness_dir = \"no\") | Gap analysis + fill gaps |\n| **Harness Present** | Core harness components exist | Audit + improvement suggestions |\n\nAlso classify ECL readiness:\n\n| ECL State | Criteria | Action |\n|-----------|----------|--------|\n| **ECL Missing** | has_ecl_doc = \"no\" OR has_changes_dir = \"no\" | Create ECL docs, change templates, and scripts |\n| **ECL Partial** | ECL doc exists but scripts\u002Ftemplates missing | Fill ECL automation gaps |\n| **ECL Ready** | docs\u002FECL.md, harness\u002Fchanges, templates, harness-change, harness-evolve, lint-ecl, lint-encoding exist | Audit index freshness and workflow quality |\n\n### 1.3 Baseline Verification Snapshot\n\nFor existing projects, capture a best-effort baseline before creating or updating harness files.\nThe baseline is for attribution only: it distinguishes pre-existing project failures from\nfailures introduced by harness work. It must not be used to weaken default CI.\n\nRun only commands that already exist in the project:\n\n| Ecosystem | Baseline commands |\n|-----------|-------------------|\n| TypeScript\u002FNode.js | package scripts such as `lint`, `typecheck`, `test`, `build`; include nested package build scripts when detected |\n| Go | `go test .\u002F...`, `go build .\u002F...`, existing `make lint` or `make test` |\n| Python | existing test\u002Flint scripts, `python -m compileall .` |\n\nRecord each command as `pass`, `fail`, or `missing`, with the short failure reason. If a command\nfails before harness creation, report it later as **pre-existing project debt**, not as harness\nfailure. Default CI remains strict and should still include normal business gates unless the user\nexplicitly asks for a temporary staged rollout.\n\n### 1.4 Intent Confirmation\n\nBefore planning changes, classify requested scope:\n\n| Scope | Default? | Includes |\n|-------|----------|----------|\n| **Core harness** | Yes | AGENTS.md, docs\u002FECL.md, docs\u002FSTATUS.md, docs, ECL changes, lightweight auto-evolve, linters, environment contract, CI |\n| **Advanced harness** | No | Core harness plus explicitly requested eval, trace, state, checkpoints, memory, or metrics |\n| **Documentation only** | No | AGENTS.md and docs without linters, scripts, or CI |\n\nWhen a user-confirmation tool is available, confirm scope. In Codex, use `request_user_input`.\nOn other platforms, use the equivalent user-choice tool. If no such tool is available, use the\ndetected context and record assumptions.\n\n```json\n{\n  \"question\": \"What's your priority for this harness setup?\",\n  \"header\": \"Scope\",\n  \"multiSelect\": false,\n  \"options\": [\n    {\n      \"label\": \"Core harness (Recommended)\",\n      \"description\": \"Project-first AGENTS.md, ECL changes, STATUS handoff, auto-evolve threshold checks, linters, environment contract, and strict CI\"\n    },\n    {\n      \"label\": \"Advanced harness\",\n      \"description\": \"Core harness plus explicitly requested eval, trace, memory, checkpoint, or metrics infrastructure\"\n    },\n    {\n      \"label\": \"Documentation only\",\n      \"description\": \"AGENTS.md and project docs only; skip linters, scripts, and CI for now\"\n    }\n  ]\n}\n```\n\n**If Empty project**, also ask for basics:\n\n```json\n{\n  \"question\": \"What tech stack for this project?\",\n  \"header\": \"Tech Stack\",\n  \"multiSelect\": false,\n  \"options\": [\n    {\"label\": \"Go\", \"description\": \"CLI tools, high-performance services, system programming\"},\n    {\"label\": \"TypeScript\u002FNode.js\", \"description\": \"Web APIs, full-stack apps, rapid prototyping\"},\n    {\"label\": \"Python\", \"description\": \"Data processing, ML\u002FAI, scripting\"}\n  ]\n}\n```\n\nIf no user-confirmation tool is available, use detected values and document assumptions:\n\n```markdown\n## Auto-Detected Context\n\n| Field | Value | Confidence | Evidence |\n|-------|-------|------------|----------|\n| Tech Stack | {TECH} | High | Found {config file} |\n| Project State | {state} | High | {criteria matched} |\n| Scope | Core harness | Default | No user preference specified |\n\nProceeding with these assumptions. Tell me if any need adjustment.\n```\n\n### 1.5 ECL Work Intake Rules\n\nWhen generating ECL guidance for a target project, keep the process small enough to use:\n\n| Intake type | Criteria | Required ECL handling |\n|-------------|----------|-----------------------|\n| **Small Change** | Local, low-risk edits such as copy, comments, style-only tweaks, or single-file bug fixes with no interface, data, permission, architecture, or release impact | Active change optional; still record the verification command in the final response or existing task notes |\n| **Structured Change** | Cross-file\u002Fmodule behavior, APIs, data model, permissions, architecture, validation chain, unclear requirements, or work likely to exceed 20 minutes | Use active change files and require intake\u002Fspec\u002Fplan review before implementation |\n\nDecision tree:\n\n1. If an active change already exists, keep using it; do not create a second active context.\n2. If the change is copy, comments, README text, formatting, or an obviously local single-file fix\n   with no runtime, API, data, permission, architecture, or validation-chain impact, treat it as\n   Small Change.\n3. If the change touches APIs, data, permissions, architecture, multiple modules, release\u002Fruntime\n   behavior, or unclear requirements, treat it as Structured Change.\n4. If impact is unclear, do read-only investigation first. If uncertainty remains after inspection,\n   ask one high-impact question or upgrade to Structured Change; do not assume Small Change.\n\nFor structured changes, support both common entry points:\n\n- **Requirement-first input**: extract target users\u002Fscenarios, evidence, success criteria,\n  acceptance criteria, non-goals, constraints, assumptions, and risks into `spec.md`.\n- **Plan-first input**: treat the user's plan as a draft, split WHAT\u002FWHY into `spec.md` and HOW into\n  `plan.md`, then ask only about high-impact gaps that affect implementation direction or acceptance.\n  If the plan is complete and does not conflict with repository evidence, do not repeat a full\n  interview. If it conflicts with code, docs, commands, or existing harness constraints, record the\n  conflict and return to Intake Review.\n\nQuestions are allowed and expected, but must be bounded: ask at most three high-impact questions per\nround. Low-risk unknowns become assumptions; high-impact unknowns become\n`[NEEDS CLARIFICATION: ...]` and block implementation until resolved.\n\nFor complex structured changes, use a lightweight iteration loop rather than treating the first\nspec as final:\n\n```text\nDraft Spec -> Draft Plan -> Review Gaps -> Revise Spec\u002FPlan -> Gate -> Tasks\n```\n\nDefault to at most two loops. If key gaps remain, continue up to five loops; after that, record a\nblocker instead of implementing from guesses. `plan.md` must include any planning-discovered spec\ngaps, because plans often expose missing acceptance, boundary, permission, data, or validation\nrequirements.\n\n---\n\n## Phase 2: Analysis\n\n**Goal**: Deeply understand codebase architecture, harness state, and environment requirements.\n\n### 2.1 Execution Mode\n\nUse subagents only when the user authorized delegation and the environment supports it. Otherwise, execute the same responsibilities inline.\n\nIf using subagents, assign:\n\n- Code architecture analysis: follow `agents\u002Fanalyzer.md`; output `harness\u002F.analysis\u002Farchitecture.json`.\n- Harness state audit: follow `agents\u002Fauditor.md`; output `harness\u002F.analysis\u002Faudit.json`.\n- Environment analysis: follow `references\u002Fenvironment-detection-guide.md`; output `harness\u002F.analysis\u002Fenvironment.json`.\n\nIf working inline, produce the same three analysis artifacts or equivalent in-memory summaries before Phase 3.\n\n### 2.2 Project Identity Extraction\n\nFor existing projects, extract target-project meaning before writing docs:\n- One-sentence project identity: what it does and for whom.\n- Core workflow or domain model: user\u002Fsystem flow, key entities, API resources, jobs, or commands.\n- Primary source entrypoints and where common changes belong.\n\nUse `README.md`, manifests, entrypoints, routes\u002Fcontrollers, schemas\u002Fmodels, and key source\ndirectories. Harness files are not sufficient evidence for project identity.\n\n### 2.3 Adapter Selection\n\nAfter detecting the tech stack, load the matching adapter before creating linters, scripts, CI,\nor environment config. Adapter guidance overrides generic templates for language-specific details.\n\n| Detected stack | Required adapter |\n|----------------|------------------|\n| TypeScript\u002FNode.js | `references\u002Fadapters\u002Ftypescript.md` |\n| Go | `references\u002Fadapters\u002Fgo.md` |\n| Python | `references\u002Fadapters\u002Fpython.md` |\n| Rust | `references\u002Fadapters\u002Frust.md` |\n| Java | `references\u002Fadapters\u002Fjava.md` |\n| Unknown\u002Fmixed | `references\u002Fadapters\u002Fgeneric.md` plus any detected language adapters |\n\nFor TypeScript\u002FNode.js projects, prefer Node\u002FTS-native outputs: `scripts\u002Flint-deps.mjs` or\nequivalent, `scripts\u002Flint-quality.mjs`, npm\u002Fpackage-manager scripts, and Node\u002FTS GitHub Actions.\nDo not adapt Go linter or Makefile-only patterns to TypeScript unless the project is actually Go\nor already uses Makefile as the primary command surface.\n\n### 2.4 Command Surface Selection\n\nBefore creating ECL scripts, select the target project's command surface. Do not assume\nPowerShell is the only Windows option. This selection is normally automatic; do not ask the user to\nchoose a script format unless project evidence conflicts or the user has already expressed a hard\nconstraint.\n\nPriority:\n\n1. Existing project entrypoints: package-manager scripts, Makefile targets, README commands,\n   or CI shell conventions.\n2. Explicit user\u002Fproject constraints. If the project rejects `.ps1`, do not generate PowerShell\n   as the only harness entrypoint.\n3. Bash profile when allowed. For Windows projects that accept Bash, generate `.sh` scripts and\n   document the prerequisite: Git Bash, WSL, MSYS2, or a CI Linux runner.\n4. PowerShell profile when the project accepts Windows-native PowerShell. Keep it compatible with\n   Windows PowerShell 5.1 and PowerShell 7.\n5. Node or Python profiles when those runtimes are already first-class project dependencies.\n\nDefault when evidence is sparse: for TypeScript\u002FNode projects choose Node\u002Fpackage-manager scripts;\nfor Windows projects that allow Bash choose Bash profile and document Git Bash\u002FWSL\u002FMSYS2; otherwise\nchoose the adapter's native lightweight scripting profile.\n\nAll profiles must implement the same ECL invariants and command set. `harness-change`,\n`harness-evolve`, `lint-ecl`, and `lint-encoding` may be implemented as `.ps1`, `.sh`, `.mjs`,\nor `.py`, but docs, CI, Makefile\u002Fpackage scripts, and verification commands must use the chosen\nentrypoint consistently.\n\n### 2.5 Wait for Analysis Completion\n\nWhen subagents are running, wait for their final reports. While waiting, you can:\n- Review any existing documentation\n- Prepare templates for Phase 4\n\n### 2.5 For Empty Projects\n\nSkip Phase 2 analysis agents. Instead:\n- Use templates from `references\u002Fgreenfield-templates.md`\n- Base decisions on user's tech stack choice\n- Design a standard 3-layer architecture\n\n---\n\n## Phase 3: Delta Synthesis\n\n**Goal**: Merge analysis results and compute exactly what needs to be created\u002Fupdated.\n\n### 3.1 Read Analysis Results\n\n```bash\ncat harness\u002F.analysis\u002Farchitecture.json\ncat harness\u002F.analysis\u002Faudit.json\ncat harness\u002F.analysis\u002Fenvironment.json\n```\n\n### 3.2 Compute Delta\n\nCreate a delta list:\n\n```markdown\n## Delta: What Needs to Be Done\n\n### Core To Create (doesn't exist)\n- [ ] AGENTS.md\n- [ ] docs\u002FECL.md\n- [ ] docs\u002FSTATUS.md\n- [ ] docs\u002FARCHITECTURE.md\n- [ ] scripts\u002Flint-deps.go\n- [ ] scripts\u002Fharness-change.{ps1|sh|mjs|py}\n- [ ] scripts\u002Fharness-evolve.{ps1|sh|mjs|py}\n- [ ] scripts\u002Flint-ecl.{ps1|sh|mjs|py}\n- [ ] scripts\u002Flint-encoding.{ps1|sh|mjs|py}\n- [ ] harness\u002Fchanges\u002F{active,parking,archive}\n- [ ] harness\u002Ftemplates\u002Fchange\u002F\n- [ ] harness\u002Fconfig\u002Fenvironment.json\n- [ ] harness\u002Fevolution\u002F{state.json,results.tsv,proposals\u002F} (`pending.md` is generated later only when the archive threshold is reached)\n\n### Optional Advanced (only if explicitly requested)\n- [ ] harness\u002Feval\u002F — agent evaluation datasets and runner inputs\n- [ ] harness\u002Ftrace\u002F — execution traces for agent runs\n- [ ] harness\u002Fstate\u002F — executor runtime state\n- [ ] harness\u002Fcheckpoints\u002F — resumable execution checkpoints\n- [ ] harness\u002Fmemory\u002F — long-term agent memory experiments\n- [ ] harness\u002Fmetrics\u002F — execution and quality metrics\n\n### To Update (exists but has gaps)\n- [ ] docs\u002FDEVELOPMENT.md — missing build commands\n- [ ] scripts\u002Flint-quality.py — missing 3 packages in layer map\n\n### Already Good (no changes needed)\n- [x] Makefile — has all required targets\n- [x] .github\u002Fworkflows\u002Fci.yml — properly configured\n```\n\n### 3.3 Confirm with User (if confirmation tool is available)\n\nFor significant changes:\n\n```json\n{\n  \"question\": \"I've analyzed the codebase. Ready to proceed with these changes?\",\n  \"header\": \"Confirm\",\n  \"multiSelect\": false,\n  \"options\": [\n    {\"label\": \"Yes, proceed with all\", \"description\": \"Create\u002Fupdate all identified items\"},\n    {\"label\": \"Show me the details first\", \"description\": \"I'll explain what each change involves\"},\n    {\"label\": \"Only critical items\", \"description\": \"Just P0\u002FP1 items, skip P2\u002FP3 for now\"}\n  ]\n}\n```\n\n---\n\n## Phase 4: Creation\u002FUpdate\n\n**Goal**: Create or update all harness files from the delta.\n\n### 4.1 Execution Mode\n\nUse subagents only when authorized and available. Otherwise, perform the same work inline. Keep write scopes disjoint if using parallel workers.\n\nCreation responsibilities:\n\n- Documentation: follow `agents\u002Fcreator-docs.md`; create\u002Fupdate AGENTS.md, docs\u002FECL.md, docs\u002FSTATUS.md, docs\u002FARCHITECTURE.md, docs\u002FDEVELOPMENT.md, and design docs. AGENTS.md is the target project's entry map, not a harness creation record. Keep the first screen project-first, but preserve ECL\u002Fcurrent-change priority in context loading: `AGENTS.md` -> `docs\u002FECL.md` -> active change if present -> auto-evolve pending if present -> otherwise `docs\u002FSTATUS.md` -> task-specific project docs.\n- Linters: follow `agents\u002Fcreator-linters.md`; create\u002Fupdate dependency, quality, ECL, and encoding checks.\n- Config and scripts: follow `agents\u002Fcreator-config.md`; create\u002Fupdate environment contract, harness scripts, changes directories\u002Ftemplates, lightweight evolution state, harness-change, harness-evolve, Makefile targets, and CI. Create advanced directories only when the confirmed scope requires them.\n\nECL change templates must include `summary.md`, `spec.md`, `plan.md`, `tasks.md`, and\n`reviews\u002Freview.md`. `spec.md` captures WHAT\u002FWHY, `plan.md` captures HOW and planning-discovered\nspec gaps, and `tasks.md` is generated only after the spec\u002Fplan gate is ready enough for\nimplementation. Do not require old archived changes to contain `plan.md`; compatibility applies to\nhistory.\n\nImportant: do not create static verification config such as `harness\u002Fconfig\u002Fverify.json`. Verification plans are generated at runtime by the executor from `environment.json` and the task context.\n\nStrict CI rule: default CI must include normal business quality gates (`lint`, `typecheck`, `test`,\n`build`, and backend\u002Fpackage-specific equivalents when available) plus harness checks. Do not remove\nor skip business gates because the baseline is red. If the baseline was already red, explain that CI\nwill be red until the pre-existing project issues are fixed. Generate staged or relaxed CI only when\nthe user explicitly asks for it.\n\nCommand surface rule: create ECL scripts for the selected profile, not a hardcoded shell. If Bash is\nselected on Windows, document Git Bash, WSL, MSYS2, or CI Linux shell requirements in the generated\nenvironment\u002Fdevelopment docs. If PowerShell is selected, detect whether `pwsh` is available; if not,\nuse `powershell -NoProfile -ExecutionPolicy Bypass`. PowerShell templates must be compatible with\nWindows PowerShell 5.1: avoid ambiguous overloads such as `TrimStart(\".\\\")`, and avoid non-ASCII\nmojibake marker string literals in `.ps1`; represent markers by Unicode codepoint or another\nPS5-safe construction.\n\n### 4.2 For Empty Projects: Also Create Business Code Plan\n\nFor empty projects, add one more agent:\n\n```\nAgent(\"create-exec-plan\", prompt=\"\"\"\nCreate execution plan for business code (harness-executor will implement this):\n\nTech stack: {TECH}\nProject type: {from user choice}\nArchitecture: 3-layer (Types → Core → Entry Points)\n\nCreate: docs\u002Fexec-plans\u002Factive\u002Fbootstrap-code.md\n\nContents:\n- Full source code for initial project structure\n- main.go\u002Findex.ts\u002Fmain.py entry point\n- Basic types and core logic\n- Test files\n\nThis is for harness-executor to implement — not ecl-harness-engineer's responsibility.\n\"\"\")\n```\n\n### 4.3 Wait for Creation Completion\n\nAgents will notify when done. Collect any issues they encountered.\n\n---\n\n## Phase 5: Verification + Handoff\n\n**Goal**: Ensure everything works, then hand off or present results.\n\n### 5.1 Run Verification\n\n```bash\n# 0. Compare against the baseline snapshot\n# Re-run the same existing lint\u002Ftypecheck\u002Ftest\u002Fbuild commands captured in Phase 1.\n\n# 1. Harness checks pass\nmake verify-harness || npm run lint:harness || {generated_harness_lint_command}\n\n# 2. Architecture linters pass\nmake lint-arch || npm run lint:arch\n\n# 3. Business build\u002Ftest gates run\ngo build .\u002F... || npm run build || python -m compileall .\n\n# 4. AGENTS.md size check\nwc -l AGENTS.md  # Should be 80-120 lines\n\n# 4b. AGENTS.md content gate\n# Confirm it explains project identity, core workflow\u002Fdomain model, source entrypoints,\n# task-based verification, active-change-before-STATUS loading, and contains no\n# ECL Harness Engineer internal boundary language.\n\n# 5. All expected files exist\ntest -f AGENTS.md && echo \"✓ AGENTS.md\"\ntest -f docs\u002FARCHITECTURE.md && echo \"✓ ARCHITECTURE.md\"\ntest -f docs\u002FECL.md && echo \"✓ ECL.md\"\ntest -f docs\u002FSTATUS.md && echo \"✓ STATUS.md\"\ntest -f scripts\u002Flint-deps* && echo \"✓ lint-deps\"\ntest -f scripts\u002Fharness-change.* && echo \"✓ harness-change\"\ntest -f scripts\u002Flint-ecl.* && echo \"✓ lint-ecl\"\ntest -f scripts\u002Fharness-evolve.* && echo \"✓ harness-evolve\"\ntest -d harness\u002F && echo \"✓ harness\u002F\"\ntest -d harness\u002Fchanges && echo \"✓ harness\u002Fchanges\"\ntest -f harness\u002Fevolution\u002Fstate.json && echo \"✓ evolution state\"\n\n# 6. Design docs exist (not just index)\nfind docs\u002Fdesign-docs -name \"*.md\" ! -name \"index.md\" | wc -l\n```\n\nClassify every verification result:\n\n| Classification | Meaning |\n|----------------|---------|\n| Harness pass | Harness-created checks\u002Ffiles\u002Fscripts work |\n| Pre-existing project failure | The same command failed in the Phase 1 baseline |\n| New regression | The command passed in Phase 1 and fails after harness creation |\n| Not available | The command\u002Fscript does not exist in this project |\n\nAGENTS.md content gate:\n- A new agent can tell what the project does within 30 seconds.\n- The core product\u002Fsystem workflow or domain model is visible.\n- Main source entrypoints and task-to-directory mapping are visible.\n- Verification guidance maps to task type.\n- Context loading reads `docs\u002FECL.md` first, then active change when present.\n- If no active change exists and `harness\u002Fevolution\u002Fpending.md` exists, read it before\n  `docs\u002FSTATUS.md`, mention it as pending maintenance, and ask whether to handle it now unless the\n  user already prioritized the current task. Reading or asking does not start auto-evolve and must\n  not block ordinary user work.\n- If no active change exists and no pending evolution exists, context loading reads `docs\u002FSTATUS.md` before task-specific project docs.\n- For structured work, `docs\u002FECL.md` explains Small Change vs Structured Change, bounded Intake\n  Review, plan-first input handling, and the spec\u002Fplan review gate.\n- Archive history is loaded selectively through `docs\u002FSTATUS.md` paths or `harness\u002Fchanges\u002FINDEX.json`, starting with historical `summary.md` only.\n- No skill-internal boundary leaks, such as sections or sentences that describe this skill's own scope limits as target-project rules.\n\n### 5.2 STATUS.md Handoff Update\n\nWhen a target project uses ECL changes, maintain `docs\u002FSTATUS.md` as a lightweight handoff file.\nIt is not the authority while an active change exists, but it becomes the default recent-history\nentry point after the active change is closed.\n\nClose-change handoff protocol:\n\n1. Before running `harness-change close`, read the active change `summary.md`, `spec.md`,\n   `plan.md`, `tasks.md`, and relevant `reviews\u002F`; update `docs\u002FSTATUS.md` with completed work,\n   verification results, residual risks, and the next recommended resume point.\n2. Run the close command so the active change moves to `harness\u002Fchanges\u002Farchive\u002F...` and\n   `harness\u002Fchanges\u002FINDEX.json` is rebuilt.\n3. After close, update `docs\u002FSTATUS.md` again with the final archive path, normally pointing to\n   the archived `summary.md`.\n4. Run the harness lint command (`npm run lint:harness`, `make verify-harness`, or the generated\n   ECL lint command) to confirm STATUS, ECL structure, and INDEX state are consistent.\n\nHooks and CI may validate `docs\u002FSTATUS.md`, but must not auto-write it or move changes.\n\n### 5.3 Auto-Evolve Check\n\nCore harnesses include lightweight auto-evolve by default. The script layer only detects when\nenough new archive evidence exists and writes `harness\u002Fevolution\u002Fpending.md`; Codex performs the\nsemantic improvement pass.\n\nTrigger model: `harness-change close` and `reindex` run `harness-evolve check`; `new` only reminds\nwhen pending exists. Hooks and CI may warn, but must not modify docs, scripts, STATUS, or changes.\nGenerated scripts do not call subagents. They only count archive evidence and create pending\ncontext. When no active change exists and Codex notices pending maintenance, it should ask the user\nwhether to handle it now unless the user already prioritized the current task. Asking does not start\npending evolution.\n\n`harness\u002Fevolution\u002Fpending.md` is a maintenance reminder, not a hard lock. Reading it for context\ndoes not start pending evolution. Pending evolution starts only when Codex creates or uses an\n`auto-evolve-harness-*` change, writes an evolution proposal\u002Fresult, or edits Harness files based\non the pending evidence. Once started, finish with a proposal, one `harness\u002Fevolution\u002Fresults.tsv`\nrow, and `harness-evolve mark-complete`; otherwise park or close blocked, not completed.\n\nApply only the smallest evidence-backed delta that passes review. No independent scorer =\nno auto-apply: user approval to handle pending implies permission to request an independent\nauditor\u002Fsubagent when the environment supports it. If the environment still requires explicit\nauthorization, ask once. If scoring is unavailable, declined, or still unauthorized after asking,\nrecord `noop` with `eval_mode=dry_run`, keep the proposal, run `mark-complete`, and stop.\nMachinery repair\n(`harness-evolve`, pending templates, lint) does not complete pending evolution by itself; after\nrepair, still evaluate candidate archives or leave the work parked\u002Fblocked.\n\nDetailed proposal format, scoring weights, status values, and complexity budget live in\n`references\u002Fecl-harness.md`.\n\n### 5.4 Present Summary\n\n```markdown\n## Harness Infrastructure Complete\n\n**Project**: {project-name}\n**Tech Stack**: {TECH}\n**Files Created\u002FUpdated**: {count}\n\n### Created Files\n- AGENTS.md ({N} lines)\n- docs\u002FARCHITECTURE.md\n- docs\u002FECL.md\n- docs\u002FSTATUS.md\n- docs\u002FDEVELOPMENT.md\n- docs\u002Fdesign-docs\u002F{component}.md\n- scripts\u002Flint-deps.{ext}\n- scripts\u002Flint-quality.{ext}\n- scripts\u002Fharness-change.{ps1|sh|mjs|py}\n- scripts\u002Flint-ecl.{ps1|sh|mjs|py}\n- scripts\u002Flint-encoding.{ps1|sh|mjs|py}\n- scripts\u002Fharness-evolve.{ps1|sh|mjs|py}\n- harness\u002Fconfig\u002Fenvironment.json\n- harness\u002Fchanges\u002F\n- harness\u002Fevolution\u002F\n- harness\u002Ftemplates\u002Fchange\u002F\n- Makefile\n\n### Verification Results\n- Harness checks: ✓\n- Architecture checks: ✓\n- Business gates: ✓ or pre-existing failures listed below\n- AGENTS.md size: ✓ ({N} lines)\n\n### Pre-existing Project Failures\n- {List baseline-red commands and short reasons, or \"None observed.\"}\n\n### New Regressions Introduced By Harness\n- {List commands that passed before and failed after, or \"None observed.\"}\n\n### Next Steps\n{For empty projects: \"Run harness-executor to implement business code from docs\u002Fexec-plans\u002Factive\u002Fbootstrap-code.md\"}\n{For existing projects: \"The harness is ready. AI agents can now use AGENTS.md as their entry point.\"}\n```\n\n### 5.5 Automatic Handoff (for Empty Projects)\n\nIf this was an empty project with a bootstrap exec-plan, invoke harness-executor:\n\n```\nSkill(skill=\"harness-executor\")\n```\n\nWith context: \"Implement the bootstrap exec-plan at docs\u002Fexec-plans\u002Factive\u002Fbootstrap-code.md\"\n\n---\n\n## Core Principles\n\n### 1. Repository as Single Source of Truth\n\nAgents cannot access Slack, Google Docs, or tribal knowledge. If it's not in the repository, it doesn't exist for the agent.\n\n### 2. AGENTS.md is a Map, Not a Manual\n\nKeep it 80-120 lines. Link to detailed docs, don't embed them.\n\n### 3. Enforce Invariants Mechanically\n\nLinter errors must be agent-actionable:\n```\n✗ BAD: \"Forbidden import in core\u002Ftypes\u002Fuser.go\"\n\n✓ GOOD: \"core\u002Ftypes\u002Fuser.go:15 imports core\u002Fconfig (layer 0 → layer 2).\n         Layer 0 packages must have NO internal dependencies.\n\n         Fix options:\n         1. Move config-dependent logic to a higher layer\n         2. Pass the config value as a parameter\n         3. Use dependency injection via an interface\"\n```\n\n### 4. Build to Delete\n\nEvery component should be replaceable. Capabilities that required complex pipelines yesterday may be single prompts tomorrow.\n\n### 5. Start Simple\n\nAtomic, well-documented tools > complex agent choreography. Don't over-engineer.\n\n### 6. Change State Is Explicit\n\nUse a single `harness\u002Fchanges\u002Factive\u002F` task for personal development. Move paused work to `parking\u002F` and closed work to `archive\u002F` with the generated `scripts\u002Fharness-change.*` command. Maintain `docs\u002FSTATUS.md` as the soft handoff summary after active work is closed. Never hand-edit `harness\u002Fchanges\u002FINDEX.json`; it is a generated index rebuilt by `park`, `close`, `resume`, and `reindex`. Structured changes use `spec.md` for WHAT\u002FWHY, `plan.md` for HOW, and `tasks.md` for executable work.\n\n### 7. Harness Evolves From Evidence\n\nEvery few closed changes, the generated `scripts\u002Fharness-evolve.* check` command may create\n`harness\u002Fevolution\u002Fpending.md`. Treat it as a maintenance reminder to improve harness rules from\nreal archived evidence, not as a hard blocker for unrelated user work. If you start acting on the\npending evidence, first refresh `harness\u002Fchanges\u002FINDEX.json` and use the current eligible archive\nwindow; the Candidate Archives in an old pending file are a trigger snapshot, not the only evidence.\nThen finish with proposal + results.tsv + `mark-complete`, or park\u002Fblock the work.\nDo not turn one-off business bugs into permanent process. Keep only changes that improve the audit\nscore and pass validation.\n\n---\n\n## Reference Files\n\n| File | When to Read | Contents |\n|------|-------------|----------|\n| `references\u002Fgreenfield-templates.md` | Empty projects (Phase 2.5) | Complete Go\u002FTS\u002FPython scaffolding |\n| `references\u002Fdocumentation-templates.md` | Phase 4 doc creation | Doc templates with numbered sections |\n| `references\u002Flinter-templates.md` | Phase 4 linter creation | Linter code templates per language |\n| `references\u002Fecl-harness.md` | ECL-aware harness creation | docs\u002FECL.md, docs\u002FSTATUS.md, change lifecycle, INDEX.json, PowerShell script templates |\n| `references\u002Fdarwin-eval-prompts.md` | Skill quality evaluation | Dry-run prompts for darwin-skill review |\n| `references\u002Fenvironment-detection-guide.md` | Phase 2 env analysis | Environment ecosystem detection |\n| `references\u002Fenvironment-config-guide.md` | Phase 4 config creation | Startup, services, env vars, user-confirmation templates |\n| `references\u002Fadapters\u002Ftypescript.md` | TypeScript\u002FNode.js projects | npm scripts, Node linters, package-manager detection, CI defaults |\n| `references\u002Fadapters\u002F{go,python,rust,java,generic}.md` | Matching detected stacks | Language-specific commands and conventions |\n\nAgent prompts for Phase 2 and Phase 4 subagents are in `agents\u002F`.\n\nFor small projects (\u003C 20 files) or when subagents aren't available, execute phases inline instead of spawning agents.\n",{"name":5,"risk":25,"tags":26,"tools":32,"author":37,"source":38,"license":15,"category":39,"date_added":40,"description":7,"source_repo":41,"source_type":38,"license_source":42},"safe",[27,28,29,30,31],"codex","agent-harness","ecl","workflow","ci",[27,33,34,35,36],"claude","cursor","gemini","antigravity","qinghui316","community","development","2026-06-13","qinghui316\u002Fecl-harness-engineer","https:\u002F\u002Fgithub.com\u002Fqinghui316\u002Fecl-harness-engineer\u002Fblob\u002Fmain\u002FLICENSE","main","c475fc64e94e803b50d14a791dfee8853fbb2c5c","integrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory","https:\u002F\u002Fgithub.com\u002FElogic360","2026-11-08 22:21:22",{"source":8,"skillName":5,"commands":50,"prompt":60,"zipAvailable":61},[51,54,57],{"tool":52,"command":53},"npx","npx skills add Elogic360\u002Fsoftware-factory --skill ecl-harness-engineer",{"tool":55,"command":56},"bunx","bunx skills add Elogic360\u002Fsoftware-factory --skill ecl-harness-engineer",{"tool":58,"command":59},"pnpm","pnpm dlx skills add Elogic360\u002Fsoftware-factory --skill ecl-harness-engineer","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002FSKILL.md",true,[63,69,73,77,81,85,89,93,97,101,105,109,113,117,121,125,129,133,137,141,145,149,153,157,161,165,169],{"path":64,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":67,"sourceHtmlUrl":68},"agents\u002Fanalyzer.md","REFERENCE","参考文档",3675,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Fagents\u002Fanalyzer.md",{"path":70,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":71,"sourceHtmlUrl":72},"agents\u002Fauditor.md",10989,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Fagents\u002Fauditor.md",{"path":74,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":75,"sourceHtmlUrl":76},"agents\u002Fcreator-config.md",11346,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Fagents\u002Fcreator-config.md",{"path":78,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":79,"sourceHtmlUrl":80},"agents\u002Fcreator-docs.md",10613,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Fagents\u002Fcreator-docs.md",{"path":82,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":83,"sourceHtmlUrl":84},"agents\u002Fcreator-linters.md",5877,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Fagents\u002Fcreator-linters.md",{"path":86,"resourceType":86,"resourceTypeDesc":87,"sizeBytes":88,"sourceHtmlUrl":16},"LICENSE","许可证",1093,{"path":90,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":91,"sourceHtmlUrl":92},"references\u002Fadapters\u002Fadapter-schema.md",9531,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Fadapter-schema.md",{"path":94,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":95,"sourceHtmlUrl":96},"references\u002Fadapters\u002Fgeneric.md",4505,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Fgeneric.md",{"path":98,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":99,"sourceHtmlUrl":100},"references\u002Fadapters\u002Fgo.md",7164,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Fgo.md",{"path":102,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":103,"sourceHtmlUrl":104},"references\u002Fadapters\u002Fjava.md",7067,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Fjava.md",{"path":106,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":107,"sourceHtmlUrl":108},"references\u002Fadapters\u002Fpython.md",7484,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Fpython.md",{"path":110,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":111,"sourceHtmlUrl":112},"references\u002Fadapters\u002Frust.md",6682,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Frust.md",{"path":114,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":115,"sourceHtmlUrl":116},"references\u002Fadapters\u002Ftypescript.md",8596,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fadapters\u002Ftypescript.md",{"path":118,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":119,"sourceHtmlUrl":120},"references\u002Farchitecture-diagrams.md",11607,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Farchitecture-diagrams.md",{"path":122,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":123,"sourceHtmlUrl":124},"references\u002Faudit-templates.md",18110,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Faudit-templates.md",{"path":126,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":127,"sourceHtmlUrl":128},"references\u002Fcapability-registry.md",13281,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fcapability-registry.md",{"path":130,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":131,"sourceHtmlUrl":132},"references\u002Fdarwin-eval-prompts.md",15938,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fdarwin-eval-prompts.md",{"path":134,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":135,"sourceHtmlUrl":136},"references\u002Fdocumentation-templates.md",21329,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fdocumentation-templates.md",{"path":138,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":139,"sourceHtmlUrl":140},"references\u002Fdurability-patterns.md",14720,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fdurability-patterns.md",{"path":142,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":143,"sourceHtmlUrl":144},"references\u002Fecl-harness.md",49632,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fecl-harness.md",{"path":146,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":147,"sourceHtmlUrl":148},"references\u002Fenvironment-config-guide.md",16959,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fenvironment-config-guide.md",{"path":150,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":151,"sourceHtmlUrl":152},"references\u002Fenvironment-detection-guide.md",21785,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fenvironment-detection-guide.md",{"path":154,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":155,"sourceHtmlUrl":156},"references\u002Feval-templates.md",10193,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Feval-templates.md",{"path":158,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":159,"sourceHtmlUrl":160},"references\u002Fgc-templates.md",25790,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fgc-templates.md",{"path":162,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":163,"sourceHtmlUrl":164},"references\u002Fgreenfield-templates.md",30510,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fgreenfield-templates.md",{"path":166,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":167,"sourceHtmlUrl":168},"references\u002Flinter-templates.md",10086,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Flinter-templates.md",{"path":170,"resourceType":65,"resourceTypeDesc":66,"sizeBytes":171,"sourceHtmlUrl":172},"references\u002Fobservability-templates.md",8241,"https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fintegrations\u002Fantigravity-skills\u002Fskills\u002Fecl-harness-engineer\u002Freferences\u002Fobservability-templates.md",[174,182,189,196,203,210],{"skillNo":175,"name":176,"nameZh":6,"description":177,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":178,"licenseFileUrl":6,"sourceHtmlUrl":179,"categories":180,"keywords":181,"publishedAt":20},"01M79G206MT3J6S0GY3C45P2KV","vss-summarize-video","Use to summarize a recorded video via the LVS summarization microservice (HITL-gated) with a VLM fallback. Not for report generation or live RTSP captioning.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fvss-summarize-video\u002FSKILL.md",[],[],{"skillNo":183,"name":184,"nameZh":6,"description":185,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":178,"licenseFileUrl":6,"sourceHtmlUrl":186,"categories":187,"keywords":188,"publishedAt":20},"01M79G1WN63PGRE8CKD47WEBQ5","vss-setup-video-analytics-api","Use to deploy the vss-video-analytics-api REST service standalone (config-source, data-log bind, Elasticsearch, optional Kafka). Not for full warehouse deploy.","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fvss-setup-video-analytics-api\u002FSKILL.md",[],[],{"skillNo":190,"name":191,"nameZh":6,"description":192,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":178,"licenseFileUrl":6,"sourceHtmlUrl":193,"categories":194,"keywords":195,"publishedAt":20},"01M79G1S9F7VY3EF6H53PRP5AN","vss-setup-behavior-analytics","Use to deploy the vss-behavior-analytics service standalone (entrypoint, config-source, optional calibration). Not for the full warehouse deploy.","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fvss-setup-behavior-analytics\u002FSKILL.md",[],[],{"skillNo":197,"name":198,"nameZh":6,"description":199,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":178,"licenseFileUrl":6,"sourceHtmlUrl":200,"categories":201,"keywords":202,"publishedAt":20},"01M79G1N1S1Y27PFZPYMNK1MTQ","vss-search-archive","Use this skill to run top-level VSS fusion search on archived video, or to ingest video files \u002F RTSP streams for search. Do NOT use for ad-hoc visual Q&A (use vss-ask-video), live captioning (use vss-deploy-dense-captioning), or video summarization and reports (use vss-summarize-video).","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fvss-search-archive\u002FSKILL.md",[],[],{"skillNo":204,"name":205,"nameZh":6,"description":206,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":178,"licenseFileUrl":6,"sourceHtmlUrl":207,"categories":208,"keywords":209,"publishedAt":20},"01M79G1H6WN0Q6C0VCAXCZX6B7","vss-query-analytics","Use this skill when reading video-analytics metrics, incidents, alerts, and sensor data via the VA-MCP server (port 9901). Not for live VLM or incident-range narrative reports.","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fvss-query-analytics\u002FSKILL.md",[],[],{"skillNo":211,"name":212,"nameZh":6,"description":213,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":178,"licenseFileUrl":6,"sourceHtmlUrl":214,"categories":215,"keywords":216,"publishedAt":20},"01M79G1DYPNXQS7QHZ0AAF5FX5","vss-manage-video-io-storage","Use to call the VIOS REST API (sensor list, timelines, clip extraction, snapshots, add\u002Fdelete sensors and streams). Not for VLM inference or search.","https:\u002F\u002Fgithub.com\u002FElogic360\u002Fsoftware-factory\u002Fblob\u002Fmain\u002Fskills\u002Fnvidia\u002Fvss-manage-video-io-storage\u002FSKILL.md",[],[]]