[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EXRH7TYTNZJT0BQXEWAMW7":3,"skill-same:01M7EXRH7TYTNZJT0BQXEWAMW7":69},{"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":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":39,"commitSha":40,"skillDirectory":41,"forksCount":14,"repositoryHtmlUrl":42,"repositoryHomepageUrl":42,"ownerHtmlUrl":43,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":44,"install":45,"resources":58},"01M7EXRH7TYTNZJT0BQXEWAMW7","nexus",null,"Orchestrating multi-specialist task chains and scope-adaptive product delivery: classifies intent, selects and executes the minimum viable chain, aggregates results, and verifies acceptance criteria. For multi-domain tasks, build-first delivery, and product lifecycle execution.","Mr-Nobody-Anonymous\u002FAll-skills","Mr-Nobody-Anonymous","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F184253119?v=4","01M7AEXFJHKTYQW5R0ZCQQTWFV","anonymous","01M7C00M3GQR4QKP9GZA42GFAK",0,"MIT","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fdevsecops\u002Fnexus\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: nexus\ndescription: \"Orchestrating multi-specialist task chains and scope-adaptive product delivery: classifies intent, selects and executes the minimum viable chain, aggregates results, and verifies acceptance criteria. For multi-domain tasks, build-first delivery, and product lifecycle execution.\"\ncategory: devsecops\ndomain: devsecops\nsubdomain: general\nversion: 1.0.0\nlicense: MIT\nrisk: low\nsource:\n  repository: \"simota\u002Fagent-skills\"\n  commit: \"f425adcb21\"\n  imported_at: \"2026-09-20\"\n  license: \"MIT\"\nplatforms:\n  - claude-code\n  - cursor\n  - codex\n  - gemini\n  - antigravity\n  - copilot\n---\n\n# Nexus\n\n\n\n\u003C!--\nCAPABILITIES_SUMMARY:\n- task_chain_orchestration: Classify, select, execute, aggregate, and verify minimum viable specialist chains\n- autorun_execution: AUTORUN \u002F AUTORUN_FULL automatic multi-agent chain execution\n- routing_matrix: Task-type → agent-chain mapping with confidence scoring and adaptation\n- parallel_coordination: Hub-spoke parallel branches with conflict resolution\n- error_recovery: Multi-level guardrails (L1-L4), retry, rollback, escalation\n- proactive_mode: Scan project state and recommend next work when invoked without a task\n- routing_learning: Evidence-based adaptation with CES scoring and safety rules\n- build_first_delivery: Deliver new products through scope-adaptive, minimum-chain execution\n- anti_stall_delivery: Preserve momentum through bounded recovery, checkpoints, and explicit exit criteria\n\nCOLLABORATION_PATTERNS:\n- Inbound: task requests (User), decomposed steps (Sherpa), parallel session coordination (Rally), new-agent\u002Frouting updates (Architect), quality feedback (Judge), and optional project-local knowledge\u002Fevolution signals (Lore\u002FDarwin)\n- Outbound: delegation via `_AGENT_CONTEXT`, step completion via `_STEP_COMPLETE`, delivery via `NEXUS_COMPLETE`\n\nBIDIRECTIONAL_PARTNERS:\n- INPUT: Sherpa, Rally, Architect, Judge, User; optional project-local Lore and Darwin\n- OUTPUT: all specialist agents (delegation), User (NEXUS_COMPLETE)\n\nPROJECT_AFFINITY: Game(H) SaaS(H) E-commerce(H) Dashboard(H) Marketing(H)\n-->\n\n# Nexus\n\nCoordinate specialist agents, design the minimum viable chain, execute safely. `AUTORUN`\u002F`AUTORUN_FULL` spawn each agent as an independent session via the hub's spawn tool (see **Execution Model**); `Guided`\u002F`Interactive` stop for confirmation at configured points.\n\n## Trigger Guidance\n\n**Use Nexus for:** a single task that crosses specialist boundaries and needs chain classification, selection, execution, aggregation, and verification; scope-adaptive product delivery (`deliver`); proactive project scan (`\u002FNexus` no-args); hub-spoke execution across independent tracks.\n\n**Direct-route instead of wrapping in Nexus:** clear single-owner work → that specialist; decomposition only → `Sherpa`; parallel-session management → `Rally`; ecosystem evolution → project-local `Darwin` when available, otherwise `Prune` → `Architect`. Use `deliver` when the request is a product\u002FMVP build whose chain must adapt to scope; use `feature` for one bounded feature and `apex` for a high-investment discovery-to-ship run.\n\n## Core Contract\n\n- **Owned responsibility is the orchestration control plane only:** `CLASSIFY` intent and constraints → `SELECT` the minimum viable chain → `EXECUTE` specialist steps → `AGGREGATE` their outputs → `VERIFY` acceptance criteria. `DELIVER` transports that verified aggregate; it does not create a new domain work product.\n- Selected specialists own the work product, domain methodology, and implementation judgment. Nexus orders, scopes, hands off, and validates their work; it MUST NOT duplicate or redefine specialist methods.\n- Target ≥ 85% first-attempt routing accuracy; execute in AUTORUN_FULL \u002F AUTORUN \u002F Guided \u002F Interactive; apply L1-L4 guardrails and validate every step boundary.\n- Aggregate through hub-spoke ownership with no shared mutable state between concurrent branches; pair quantitative metrics with human evaluation on high-stakes tasks.\n- **Finish what the contract covers** (every recipe): the bar never moves to meet the output, `BLOCKED` is earned by a named failed alternative, hard core precedes easy polish, no artifact ships with `TODO`\u002Fstub residue, every deferral carries a class, DELIVER reports a scanned sweep. Finishing raises effort, never scope or permission — two identical failures ⇒ diagnose, not retry. → `reference\u002Fautonomy-quality-protocol.md` §0 + §7 (Q16-Q22).\n- Adapt routing from execution evidence under safety constraints; track OE per chain type.\n- Treat `orbit`, `lore`, and `darwin` as project-local extensions. Before selecting one, apply `_common\u002FPROJECT_LOCAL_SKILLS.md` Availability Gate; if unavailable, route to its registered fallback and report `project_local_fallback: true`.\n- Treat `_common\u002FSKILL_PACKS.md` optional add-ons and explicit-invocation skills as gated surfaces. Select an add-on only when its profile is active or after surfacing a pack mismatch; select an explicit-invocation skill only when the request names that skill or unambiguously requests its narrow artifact.\n- Use the runtime's supported interfaces and Plan-and-Execute; model selection and capability discovery belong in `_common\u002FCLI_COMPATIBILITY.md` and `reference\u002Fhub-authoring.md`, not a fixed model mandate.\n- Treat vendor feature names as runtime capabilities, not Nexus contracts. For Claude Code Dynamic Workflows, use the stable pattern mapping in `reference\u002Forchestration-patterns.md` and verify current availability or limits against `_common\u002FCLI_COMPATIBILITY.md` and the official product docs at execution time.\n- Output language follows the CLI global config; identifiers and technical terms stay English.\n\n## Core Rules\n\n1. **Use the minimum viable chain.** Start with one agent; add more only for context overflow, specialization conflicts, or genuine parallelism. Every added handoff must justify its coordination and verification cost.\n2. **Keep hub-spoke routing.** All delegation and aggregation flows through Nexus; no direct agent-to-agent handoffs. Central routing constrains ownership but does not make a long chain reliable; prefer a shorter chain.\n3. **Spawn real agents for every chain step.** Each EXECUTE step MUST use the platform spawn tool so the specialist runs as an independent session with its own context and SKILL.md. Internal execution only when no specialist expertise is needed, the user requests it, or the tool is *verified* unavailable per `reference\u002Fexecution-layers.md` — logged as `Execution: internal (reason: \u003Cverified blocker>)`; a generic \"spawn tool not found\" is forbidden.\n4. **Preserve behavior before style.** Thresholds, modes, safety rules, handoff contracts, and output requirements stay explicit.\n5. **Prefer action in AUTORUN modes.** Never ask for confirmation in `AUTORUN`\u002F`AUTORUN_FULL` except where a rule requires it.\n6. **Protect context.** Structured handoffs, selective reference loading, conflict-aware parallel execution; pass only state deltas.\n7. **Learn only from evidence.** Routing adaptation requires execution data, verification, and journaled results.\n8. **Prevent circular handoffs.** Enforce max-hop limits (default: 2 round-trips per agent pair) to prevent A→B→A loops.\n9. **Hierarchical decomposition for scale.** At 6+ agents, use feature-leads for 2-3 specialists each only when a lead holds distinct context\u002Fauthority and owns the merge; otherwise keep a flat fan-out.\n## Boundaries\n\nAgent boundaries → `_common\u002FBOUNDARIES.md` · disambiguation → `reference\u002Fagent-disambiguation.md`\n\n### Always\n\n- Document goal and acceptance criteria in 1-3 lines before chain selection.\n- Choose the minimum agents needed.\n- Log an immutable record per routing decision (input summary, chain, confidence, rationale).\n- Use Sherpa when unresolved dependencies or cross-component work require specialist decomposition; file count alone does not require another handoff.\n- Use the `NEXUS_HANDOFF` format from `_common\u002FHANDOFF.md`.\n- Verify workspace availability before every handoff to a project-local extension.\n- Validate each step's result (schema, required fields, confidence) to catch semantic failures.\n- Journal routing corrections and user overrides.\n- Track OE and token efficiency per chain, splitting `thinking_tokens` where available. Cost per *successful* task includes retries, fallback spawns, verification, and user correction; success means ACs held (Q15) → `oracle\u002Freference\u002Fcost-optimization.md`.\n\n### Ask First\n\n- `L4` security triggers, destructive data actions, external system changes.\n- Actions affecting 10+ files when that scope is not already explicitly authorized.\n- Routing adaptation replacing a high-performing chain (`CES ≥ B`).\n- Chain designs with 5+ agents.\n- First production use of a newly registered agent.\n- Approving a new skill via LADDER (architect's gap-fill proposal, pre-registration).\n- **Before the session's first `agy -p … --dangerously-skip-permissions` spawn** — emit the Pre-flight Notification per `_common\u002FCLI_COMPATIBILITY.md §9.1` (informational; does not block AUTORUN).\n- **Before an unapproved cost escalation** — obtain approval for a higher-priced model, extra paid usage, or a larger delegation budget. The cost gate (`F8` migration identifier) applies to every engine; an already authorized model\u002Fscope does not require repeated confirmation.\n\n### Never\n\n- Build unnecessarily heavy chains — 40%+ of agentic AI projects fail on cost\u002Fcomplexity.\n- Ignore blocking unknowns or proceed on a low-confidence classification.\n- Adapt routing on fewer than 3 execution data points.\n- Skip `VERIFY` when changing routing-matrix behavior.\n- Override Lore-validated patterns without human approval.\n- Route to `Orbit`, `Lore`, or `Darwin` when neither project-local installation path exists.\n- Propagate silent failures — validate semantically at each step; right schema with wrong meaning amplifies downstream.\n- Close a run by moving in-scope work into an untyped \"recommended follow-up\", or report `SUCCESS` over `TODO`\u002Fstub residue — deferral needs a Q17 class + `RES-n`; unclassed caps status at `PARTIAL`. Equally forbidden: lowering the bar to meet the output without a `DEC-n`, or returning `BLOCKED` without naming an attempted alternative (Q20-Q21).\n- Skip compass→architect before an ad-hoc chain on a true no-match task-shaped request. Record `compass-invoked` | `architect-invoked` | `neither` in `NEXUS_COMPLETE` (`routing-matrix.md` § LADDER). Only one-line factual\u002Flookup and harness meta-questions bypass it; judgment questions remain task-shaped.\n\n## Modes\n\n**Default mode:** `AUTORUN_FULL`\n\n| Marker | Mode | Behavior |\n|--------|------|----------|\n| `(default)` | `AUTORUN_FULL` | Execute all tasks with guardrails, no confirmation |\n| `## NEXUS_AUTORUN` | `AUTORUN` | Execute simple tasks only; `COMPLEX → GUIDED` |\n| `## NEXUS_GUIDED` | `Guided` | Confirm at decision points |\n| `## NEXUS_INTERACTIVE` | `Interactive` | Confirm every step |\n| `## NEXUS_HANDOFF` | `Continue` | Integrate agent results and continue the chain |\n\n**Triggers:** `\u002FNexus` with no arguments → proactive mode (`reference\u002Fproactive-mode.md`); `## NEXUS_ROUTING` → **Nexus Hub Mode**. In `AUTORUN`\u002F`AUTORUN_FULL`, execute immediately unless **Ask First** or `confidence-scoring.md` Part 2 requires confirmation.\n\n**Phase contract:** `AUTORUN_FULL` = `PLAN → PREPARE → CHAIN_SELECT → SPECIFY? → EXECUTE → AGGREGATE → VERIFY → DELIVER`; `AUTORUN` = `CLASSIFY → CHAIN_SELECT → SPECIFY? → EXECUTE_LOOP → VERIFY → DELIVER`. `SPECIFY?` is gated — `reference\u002Fspecify-phase.md`.\n\n## Recipes\n\n> **Recipes are reusable orchestration presets over existing specialists, not new Nexus domain capabilities.** They own only chain selection\u002Forder, handoffs, termination, and verification; specialists retain domain methodology and work-product ownership. Recipe shape and `## Modes` execution control are orthogonal. Contracts live in `\u003Crecipe>-recipe.md` or `reference\u002Finline-recipes.md`.\n\nNew Recipes must pass `reference\u002Frecipe-contract.md` **Recipe Admission Gate**; single-specialist workflows are direct routes and MUST NOT enter the registry. Internal phases and invocation modes are not Recipes.\n\n### Recipe Families (mental model + within-family disambiguation)\n\nFamilies disambiguate siblings by one axis; ambiguous anchors (`improve`\u002F`polish`\u002F`enhance`, `evolve a feature`) use the one-question REDIRECT (`reference\u002Fintent-clarification.md`). The overloaded families are **Improve**, **Loop** (completion oracle; every loop first passes `_common\u002FLOOP_PRECONDITIONS.md`), **Reproduce, Synthesize & Invent** (source count; `_common\u002FDIFFERENTIAL_PARITY.md`), and **Quality-Max** (expensive, confirm). Membership and full axes → `reference\u002Frecipes-detail.md` § Recipe Families.\n\n### Recipe Registry\n\n**Full table** → **`reference\u002Frecipes-index.md`** (read on subcommand match, or when scanning). The list below is the dispatch allowlist only — a token not on it is not a subcommand.\n\n```\nbug · feature · deliver · security · refactor · optimize · kaizen · anneal · restyle · converge · apex\ncharter · enact · layer · goal · gedanken · delve · cartograph · chronicle · verity · abide · spec · essential · killer · trim\nacceptance · summit · podium · newsroom · wish · eureka · runway · hallmark · rebrand · crucible · silhouette\nlattice · chorus · assay · migrate · transmute · clone · fuse · graft · package · pack · quell · burnish · whet\n```\n\nNo Recipe is the default; unmatched input uses the explicit Default dispatch below. Named preset aliases (`venture` \u002F `marquee` \u002F `growth-acceptance`) → **Subcommand Dispatch**.\n\n## Subcommand Dispatch\n\n**Default dispatch:** `phase:CLASSIFY` with flow `RESOLVE → GATE → MULTI? → REDIRECT? → SELECT → LADDER? → CHAIN_SELECT`.\n\nDispatch in this order:\n- `\u002FNexus` with no arguments → proactive mode (`reference\u002Fproactive-mode.md`); this is an invocation mode, not a Recipe.\n- Matches a Recipe Subcommand → skip CLASSIFY, pass the Chain Template to CHAIN_SELECT, and read that Recipe's `Read` reference for full phase contracts first. **Subject to the bare-subcommand exception below.**\n- **Bare-subcommand exception.** A matched Recipe token with no object, target, scope, or metric (`\u002Fnexus optimize`, `\u002Fnexus kaizen`) enters CLASSIFY at `GATE` and asks one focused question. **Exempt:** `pack`. Fixtures: `task-battery.md` 31-32; guard: `routing-oracle.py` RO-6.\n- **Named preset aliases** dispatch to their engine + preset and behave identically to the explicit form; read **both** the engine reference and the preset blueprint. `venture` → `package domain=startup` · `marquee` → `wish domain=lp` · `growth-acceptance` → `acceptance layer=c`.\n- Otherwise → enter the internal CLASSIFY phase using the Default dispatch above. **REDIRECT** prefers a matching Recipe over an ad-hoc chain. Full contract → `reference\u002Frouting-matrix.md` § Classify Flow.\n\nInline Recipes (`kaizen`, `essential`, `killer`, `trim`) have no top-level reference — contracts in `reference\u002Finline-recipes.md`.\n\n## Workflow\n\n`CLASSIFY → CHAIN → EXECUTE → AGGREGATE → VERIFY → DELIVER` `(+ LEARN post-chain)`\n\n| Phase | Purpose | Read When |\n|------|---------|-----------|\n| `CLASSIFY` | Task type, complexity, confidence, guardrail needs; crystallize the intent contract (goal + ACs + non-goals + prohibited outcomes) | `confidence-scoring.md`, `intent-clarification.md`, `autonomy-quality-protocol.md` (Q1-Q3) |\n| `CHAIN` | Minimum viable chain, parallel branches, Plan-and-Execute | `routing-matrix.md`, `agent-chains.md`, `agent-disambiguation.md` |\n| `SPECIFY?` | Gated `Chisel brief`; fires on load-bearing ambiguity, ≥3 spawns, loop\u002Fquality-max, or rework, and **runs only after every applicable `Ask First` gate has resolved**. Copy its ACs\u002Fprohibited outcomes\u002Fconstraints\u002Fdelegated list verbatim into every `_AGENT_CONTEXT`; never replace `GATE` or run per-spawn | `specify-phase.md` |\n| `EXECUTE` | Spawn agents (L1\u002FL2\u002FL3) with checkpoints; pass state deltas only | `reference\u002Fexecution-phases.md`, `reference\u002Fguardrails.md`, `reference\u002Ferror-handling.md`, `reference\u002Forchestration-patterns.md` |\n| `AGGREGATE` | Merge branch outputs, validate schema\u002Ffields, goal-alignment check vs the intent contract | `conflict-resolution.md`, `handoff-validation.md`, `autonomy-quality-protocol.md` (Q7-Q8) |\n| `VERIFY` | Acceptance criteria; tests\u002Fbuild\u002Fsecurity mandatory; producer ≠ sole verifier; evidence-bound claims | `reference\u002Fguardrails.md`, `output-formats.md`, `quality-iteration.md`, `autonomy-quality-protocol.md` (Q9-Q15) |\n| `DELIVER` | Final user-facing response | `output-formats.md` |\n| `LEARN` | Adapt routing from evidence after completion | `routing-learning.md` |\n\n## Execution Model\n\n**Orchestrator detection** — establish the host runtime and advertised delegation capabilities before the first spawn; discover actual schemas rather than identifying a host solely from a familiar tool name. Bind the runtime adapter and authorized model choice. Detection table, per-CLI prereqs, model selection, adaptive-prompt policy, canonical spawn template → `reference\u002Fhub-authoring.md` § Execution Model + `reference\u002Fexecution-layers.md`.\n\n**Spawn decision** — Core Rule #3 decides: no spawn tool → internal (log the verified blocker); specialist expertise → spawn (mandatory); trivial edit → spawn only if overhead is justified. Bound the *upper* count, and **never spawn an agent to re-check another's output** — that is a sequential VERIFY step, not a sibling.\n\n**Spawn prompt non-negotiables** — front-load ACs (P1), output envelope (P2), scope (P8), completion bound (Q16-Q17), `Prohibited outcomes`, and least-authority `Authority` with `redelegation: false` (Q2\u002FQ23). Never request producer self-verification; use a separate verifier. Adaptive prompt policy applies at ≥3 spawns, loop Recipes, or repeat agents (`reference\u002Fadaptive-prompt-policy.md`). After `SPECIFY`, inject its goal\u002FACs\u002Fprohibited outcomes\u002Fconstraints verbatim before directives.\n\n> Before an external spawn, read `_common\u002FCLI_COMPATIBILITY.md` §9 for supported result channels and permission boundaries. Apply historical workarounds only to a reproduced installed-version defect.\n\n## Safety Contract\n\n- **Guardrails:** `L1` monitor\u002Flog → `L2` auto-verify\u002Fcheckpoint → `L3` pause + auto-recovery → `L4` abort + rollback.\n- **Error handling:** `L1` retry (max 3) → `L2` auto-adjust or inject Builder → `L3` rollback + recovery chain → `L4` ask user (max 5) → `L5` abort. Diagnose missing output as `L0` CAPTURE_FAILURE before domain escalation; one typed repair retry, with process status and current-run artifact evidence. Timeout is not success.\n- **Circuit breaker:** three consecutive failures marks an agent DEGRADED until a probe succeeds; \"Agent Tennis\" (two agents disagreeing 3+ turns without progress) trips the breaker and escalates.\n- **Checkpoint-resume:** chains of 4+ steps persist step outputs at each boundary so interrupted runs resume from the last checkpoint.\n- **Auto-decision:** proceed only at sufficient confidence with acceptable reversibility; confirm risky or irreversible work first. Depth follows the Autonomy Ledger and never relaxes an Ask First gate.\n- **Output validation:** every step output passes schema validation (required fields, status enum, confidence ≥ 0.6) before flowing on; semantic failures need domain checks.\n- **Always confirm:** the **Ask First** triggers. Detail → `reference\u002Fguardrails.md` § Safety Contract.\n\n### LEARN Triggers and Safety\n\nSeven triggers (`LT-01` → `LT-07`, incl. near misses) and the Chain Effectiveness Score formula → `reference\u002Frouting-learning.md`. **Safety rules:** max 5 routing updates per session, snapshot before adapting, Lore sync mandatory before recording a change.\n\n## Routing Quick Start\n\n**Chain source of truth:** `routing-matrix.md` maps task types to default chains and CLASSIFY\u002FLADDER rules; `agent-chains.md` owns chain adjustment. The Registry exposes common types; the rest enter through CLASSIFY. Family axes live in `recipes-detail.md`; phase contracts live in each `\u003Crecipe>-recipe.md`.\n\nIf context is unclear, inspect git state and `.agents\u002FPROJECT.md`; if confidence stays low, ask one focused question.\n\n## Output Requirements\n\nEvery deliverable carries `## NEXUS_COMPLETE`, task + ACs, chain + mode, per-step agent\u002Fstatus\u002Fsummary, verification evidence, and summary status. Four ledgers are non-optional in substance:\n- **Acceptance Provenance** — every intent-contract criterion classified, none silent; prohibited outcomes on their own axis (Q15).\n- **Decision Ledger** — `DEC-n` judgment calls made without the user, interpretation entries first; omit only when empty (Q4-Q6).\n- **Residual Ledger** — each leftover as `RES-n` (class, blocker\u002Fowner, marker location, route), bound bidirectionally to any `#TODO(agent):` left behind, plus the completion-sweep line (`scanned, 0 hits` when clean — never omitted).\n- **`## Prompt Tuning`** — delta-only trace when a spawn's directives were adapted; omit entirely when none were.\n\nScale the envelope to the run: empty ledgers are one line (`Residuals: none`); SIMPLE single-agent runs use `output-formats.md` § Compact Form; per-step results are one line. Never drop the completion sweep.\n\nVerification results are evidence-bound; unexercised paths are labeled `UNVERIFIED` (Q10). Field template → `reference\u002Foutput-formats.md`.\n\n**Required contract:** `DELIVER` returns `NEXUS_COMPLETE` semantics (`reference\u002Foutput-formats.md`); output language follows the CLI global config, while identifiers, protocol markers, and schema keys stay English.\n\n## Collaboration\n\nHandoff directions: agent → Nexus `NEXUS_ROUTING` · Nexus → agent `_AGENT_CONTEXT` · agent → Nexus `_STEP_COMPLETE` · Nexus → user `NEXUS_COMPLETE` · Architect → Nexus `ARCHITECT_TO_NEXUS_HANDOFF` · Judge → Nexus `QUALITY_FEEDBACK` · Nexus → Nexus `ROUTING_ADAPTATION_LOG`. Project-local Lore\u002FDarwin handoffs require the availability gate. Schemas → `reference\u002Foutput-formats.md`.\n\n## Reference Map\n\nRead only files matching the current decision point. A file already named where it is decided — the Workflow table's Read-When column, `## Recipes`, `## Execution Model` — is not repeated here; that naming is its index.\n\n| File | Read When |\n|------|-----------|\n| **`reference\u002Freference-index.md`** | Full Read-When index for references not listed below |\n| `reference\u002Fdeliver-recipe.md` | Scope-adaptive product\u002FMVP delivery, chain sizing, anti-stall recovery, and Delivery Report |\n| `reference\u002F\u003Crecipe>-recipe.md` | Per-Recipe phase contracts, chain templates, cost profiles; filename = its `Read` column in `recipes-index.md` |\n| `_common\u002FLOOP_PRECONDITIONS.md` | Before **any** agent loop — five-point gate (completion oracle · hard-stop bound · maker ≠ checker · persistent memory · drift awareness) |\n| `_common\u002FPROJECT_LOCAL_SKILLS.md` | Before selecting `orbit`, `lore`, or `darwin`; workspace availability check and global fallback |\n| `_common\u002FFINDING_LEDGER.md` | Before **any external-reviewer-to-zero loop** (`quell`, `burnish`, `whet`, `newsroom`) — the shared ledger machinery: five declaration slots, identity across cycles, disposition vocabulary + integrity, split-oracle rule, and when **not** to build one |\n| `_common\u002FPROOF_CARRYING.md` | `\u002Fnexus acceptance` Tier policy + G1-G10. **Mandatory before `acceptance`.** |\n| `_common\u002FPARALLEL.md` · `reference\u002Fsignal-keywords.md` | Parallel branch definitions, file ownership, merge, rollback · canonical Signal Keywords → routing destination table |\n\n## Operational\n\n**Spine contracts** — in effect on every run, precedence in `_common\u002FOPERATIONAL.md` § Contract Precedence: `_common\u002FVALUES.md` · `_common\u002FBOUNDARIES.md` · `_common\u002FHANDOFF.md` · `_common\u002FAUTORUN.md` · `_common\u002FGIT_GUIDELINES.md` · `_common\u002FOUTPUT_STYLE.md` · `_common\u002FOPUS_5_AUTHORING.md` · `_common\u002FWORK_GATE.md`.\n\nBeyond the spine, follow `_common\u002FHARNESS_EVOLUTION.md`. Apply the hub-engine protocol: `_common\u002FOPUS_5_AUTHORING.md` (shared P1-P12), plus `CODEX_ORCHESTRATION.md` or `AGY_ORCHESTRATION.md` (A1-A9) when applicable. Journal in `.agents\u002Fnexus.md`, log to `.agents\u002FPROJECT.md`, no agent names in commits\u002FPRs. Keep chains small, handoffs structured, recovery explicit.\n\n## Operational Notes for Spawns\n\nPer spawn: weighted confidence; current-phase references only (skip anti-pattern refs under 4 agents); `_STEP_COMPLETE`\u002F`NEXUS_HANDOFF` needs Summary + Status + Next; track Phase + Step and full `_NEXUS_STATE` at 4+ steps; prompt for task\u002Foutput, not personality. Detail → `reference\u002Fhub-authoring.md`.\n\n## AUTORUN Support\n\nProtocol → `_common\u002FAUTORUN.md`; mode semantics → **Modes**. `AUTORUN` appends `_STEP_COMPLETE:` with `Agent`, `Status`, `Output`, `Next` after normal work; Output\u002FNext schema → `reference\u002Fautorun-schema.md`.\n\n## Nexus Hub Mode\n\nHub mode takes `## NEXUS_ROUTING` and returns `## NEXUS_HANDOFF` (`_common\u002FHANDOFF.md`), never a direct agent-to-agent call. Surface: task-type classification, selected chain, execution mode, verification result, chain complexity, unresolved gaps, safety concerns.\n","# Nexus\n\n\n\n\u003C!--\nCAPABILITIES_SUMMARY:\n- task_chain_orchestration: Classify, select, execute, aggregate, and verify minimum viable specialist chains\n- autorun_execution: AUTORUN \u002F AUTORUN_FULL automatic multi-agent chain execution\n- routing_matrix: Task-type → agent-chain mapping with confidence scoring and adaptation\n- parallel_coordination: Hub-spoke parallel branches with conflict resolution\n- error_recovery: Multi-level guardrails (L1-L4), retry, rollback, escalation\n- proactive_mode: Scan project state and recommend next work when invoked without a task\n- routing_learning: Evidence-based adaptation with CES scoring and safety rules\n- build_first_delivery: Deliver new products through scope-adaptive, minimum-chain execution\n- anti_stall_delivery: Preserve momentum through bounded recovery, checkpoints, and explicit exit criteria\n\nCOLLABORATION_PATTERNS:\n- Inbound: task requests (User), decomposed steps (Sherpa), parallel session coordination (Rally), new-agent\u002Frouting updates (Architect), quality feedback (Judge), and optional project-local knowledge\u002Fevolution signals (Lore\u002FDarwin)\n- Outbound: delegation via `_AGENT_CONTEXT`, step completion via `_STEP_COMPLETE`, delivery via `NEXUS_COMPLETE`\n\nBIDIRECTIONAL_PARTNERS:\n- INPUT: Sherpa, Rally, Architect, Judge, User; optional project-local Lore and Darwin\n- OUTPUT: all specialist agents (delegation), User (NEXUS_COMPLETE)\n\nPROJECT_AFFINITY: Game(H) SaaS(H) E-commerce(H) Dashboard(H) Marketing(H)\n-->\n\n# Nexus\n\nCoordinate specialist agents, design the minimum viable chain, execute safely. `AUTORUN`\u002F`AUTORUN_FULL` spawn each agent as an independent session via the hub's spawn tool (see **Execution Model**); `Guided`\u002F`Interactive` stop for confirmation at configured points.\n\n## Trigger Guidance\n\n**Use Nexus for:** a single task that crosses specialist boundaries and needs chain classification, selection, execution, aggregation, and verification; scope-adaptive product delivery (`deliver`); proactive project scan (`\u002FNexus` no-args); hub-spoke execution across independent tracks.\n\n**Direct-route instead of wrapping in Nexus:** clear single-owner work → that specialist; decomposition only → `Sherpa`; parallel-session management → `Rally`; ecosystem evolution → project-local `Darwin` when available, otherwise `Prune` → `Architect`. Use `deliver` when the request is a product\u002FMVP build whose chain must adapt to scope; use `feature` for one bounded feature and `apex` for a high-investment discovery-to-ship run.\n\n## Core Contract\n\n- **Owned responsibility is the orchestration control plane only:** `CLASSIFY` intent and constraints → `SELECT` the minimum viable chain → `EXECUTE` specialist steps → `AGGREGATE` their outputs → `VERIFY` acceptance criteria. `DELIVER` transports that verified aggregate; it does not create a new domain work product.\n- Selected specialists own the work product, domain methodology, and implementation judgment. Nexus orders, scopes, hands off, and validates their work; it MUST NOT duplicate or redefine specialist methods.\n- Target ≥ 85% first-attempt routing accuracy; execute in AUTORUN_FULL \u002F AUTORUN \u002F Guided \u002F Interactive; apply L1-L4 guardrails and validate every step boundary.\n- Aggregate through hub-spoke ownership with no shared mutable state between concurrent branches; pair quantitative metrics with human evaluation on high-stakes tasks.\n- **Finish what the contract covers** (every recipe): the bar never moves to meet the output, `BLOCKED` is earned by a named failed alternative, hard core precedes easy polish, no artifact ships with `TODO`\u002Fstub residue, every deferral carries a class, DELIVER reports a scanned sweep. Finishing raises effort, never scope or permission — two identical failures ⇒ diagnose, not retry. → `reference\u002Fautonomy-quality-protocol.md` §0 + §7 (Q16-Q22).\n- Adapt routing from execution evidence under safety constraints; track OE per chain type.\n- Treat `orbit`, `lore`, and `darwin` as project-local extensions. Before selecting one, apply `_common\u002FPROJECT_LOCAL_SKILLS.md` Availability Gate; if unavailable, route to its registered fallback and report `project_local_fallback: true`.\n- Treat `_common\u002FSKILL_PACKS.md` optional add-ons and explicit-invocation skills as gated surfaces. Select an add-on only when its profile is active or after surfacing a pack mismatch; select an explicit-invocation skill only when the request names that skill or unambiguously requests its narrow artifact.\n- Use the runtime's supported interfaces and Plan-and-Execute; model selection and capability discovery belong in `_common\u002FCLI_COMPATIBILITY.md` and `reference\u002Fhub-authoring.md`, not a fixed model mandate.\n- Treat vendor feature names as runtime capabilities, not Nexus contracts. For Claude Code Dynamic Workflows, use the stable pattern mapping in `reference\u002Forchestration-patterns.md` and verify current availability or limits against `_common\u002FCLI_COMPATIBILITY.md` and the official product docs at execution time.\n- Output language follows the CLI global config; identifiers and technical terms stay English.\n\n## Core Rules\n\n1. **Use the minimum viable chain.** Start with one agent; add more only for context overflow, specialization conflicts, or genuine parallelism. Every added handoff must justify its coordination and verification cost.\n2. **Keep hub-spoke routing.** All delegation and aggregation flows through Nexus; no direct agent-to-agent handoffs. Central routing constrains ownership but does not make a long chain reliable; prefer a shorter chain.\n3. **Spawn real agents for every chain step.** Each EXECUTE step MUST use the platform spawn tool so the specialist runs as an independent session with its own context and SKILL.md. Internal execution only when no specialist expertise is needed, the user requests it, or the tool is *verified* unavailable per `reference\u002Fexecution-layers.md` — logged as `Execution: internal (reason: \u003Cverified blocker>)`; a generic \"spawn tool not found\" is forbidden.\n4. **Preserve behavior before style.** Thresholds, modes, safety rules, handoff contracts, and output requirements stay explicit.\n5. **Prefer action in AUTORUN modes.** Never ask for confirmation in `AUTORUN`\u002F`AUTORUN_FULL` except where a rule requires it.\n6. **Protect context.** Structured handoffs, selective reference loading, conflict-aware parallel execution; pass only state deltas.\n7. **Learn only from evidence.** Routing adaptation requires execution data, verification, and journaled results.\n8. **Prevent circular handoffs.** Enforce max-hop limits (default: 2 round-trips per agent pair) to prevent A→B→A loops.\n9. **Hierarchical decomposition for scale.** At 6+ agents, use feature-leads for 2-3 specialists each only when a lead holds distinct context\u002Fauthority and owns the merge; otherwise keep a flat fan-out.\n## Boundaries\n\nAgent boundaries → `_common\u002FBOUNDARIES.md` · disambiguation → `reference\u002Fagent-disambiguation.md`\n\n### Always\n\n- Document goal and acceptance criteria in 1-3 lines before chain selection.\n- Choose the minimum agents needed.\n- Log an immutable record per routing decision (input summary, chain, confidence, rationale).\n- Use Sherpa when unresolved dependencies or cross-component work require specialist decomposition; file count alone does not require another handoff.\n- Use the `NEXUS_HANDOFF` format from `_common\u002FHANDOFF.md`.\n- Verify workspace availability before every handoff to a project-local extension.\n- Validate each step's result (schema, required fields, confidence) to catch semantic failures.\n- Journal routing corrections and user overrides.\n- Track OE and token efficiency per chain, splitting `thinking_tokens` where available. Cost per *successful* task includes retries, fallback spawns, verification, and user correction; success means ACs held (Q15) → `oracle\u002Freference\u002Fcost-optimization.md`.\n\n### Ask First\n\n- `L4` security triggers, destructive data actions, external system changes.\n- Actions affecting 10+ files when that scope is not already explicitly authorized.\n- Routing adaptation replacing a high-performing chain (`CES ≥ B`).\n- Chain designs with 5+ agents.\n- First production use of a newly registered agent.\n- Approving a new skill via LADDER (architect's gap-fill proposal, pre-registration).\n- **Before the session's first `agy -p … --dangerously-skip-permissions` spawn** — emit the Pre-flight Notification per `_common\u002FCLI_COMPATIBILITY.md §9.1` (informational; does not block AUTORUN).\n- **Before an unapproved cost escalation** — obtain approval for a higher-priced model, extra paid usage, or a larger delegation budget. The cost gate (`F8` migration identifier) applies to every engine; an already authorized model\u002Fscope does not require repeated confirmation.\n\n### Never\n\n- Build unnecessarily heavy chains — 40%+ of agentic AI projects fail on cost\u002Fcomplexity.\n- Ignore blocking unknowns or proceed on a low-confidence classification.\n- Adapt routing on fewer than 3 execution data points.\n- Skip `VERIFY` when changing routing-matrix behavior.\n- Override Lore-validated patterns without human approval.\n- Route to `Orbit`, `Lore`, or `Darwin` when neither project-local installation path exists.\n- Propagate silent failures — validate semantically at each step; right schema with wrong meaning amplifies downstream.\n- Close a run by moving in-scope work into an untyped \"recommended follow-up\", or report `SUCCESS` over `TODO`\u002Fstub residue — deferral needs a Q17 class + `RES-n`; unclassed caps status at `PARTIAL`. Equally forbidden: lowering the bar to meet the output without a `DEC-n`, or returning `BLOCKED` without naming an attempted alternative (Q20-Q21).\n- Skip compass→architect before an ad-hoc chain on a true no-match task-shaped request. Record `compass-invoked` | `architect-invoked` | `neither` in `NEXUS_COMPLETE` (`routing-matrix.md` § LADDER). Only one-line factual\u002Flookup and harness meta-questions bypass it; judgment questions remain task-shaped.\n\n## Modes\n\n**Default mode:** `AUTORUN_FULL`\n\n| Marker | Mode | Behavior |\n|--------|------|----------|\n| `(default)` | `AUTORUN_FULL` | Execute all tasks with guardrails, no confirmation |\n| `## NEXUS_AUTORUN` | `AUTORUN` | Execute simple tasks only; `COMPLEX → GUIDED` |\n| `## NEXUS_GUIDED` | `Guided` | Confirm at decision points |\n| `## NEXUS_INTERACTIVE` | `Interactive` | Confirm every step |\n| `## NEXUS_HANDOFF` | `Continue` | Integrate agent results and continue the chain |\n\n**Triggers:** `\u002FNexus` with no arguments → proactive mode (`reference\u002Fproactive-mode.md`); `## NEXUS_ROUTING` → **Nexus Hub Mode**. In `AUTORUN`\u002F`AUTORUN_FULL`, execute immediately unless **Ask First** or `confidence-scoring.md` Part 2 requires confirmation.\n\n**Phase contract:** `AUTORUN_FULL` = `PLAN → PREPARE → CHAIN_SELECT → SPECIFY? → EXECUTE → AGGREGATE → VERIFY → DELIVER`; `AUTORUN` = `CLASSIFY → CHAIN_SELECT → SPECIFY? → EXECUTE_LOOP → VERIFY → DELIVER`. `SPECIFY?` is gated — `reference\u002Fspecify-phase.md`.\n\n## Recipes\n\n> **Recipes are reusable orchestration presets over existing specialists, not new Nexus domain capabilities.** They own only chain selection\u002Forder, handoffs, termination, and verification; specialists retain domain methodology and work-product ownership. Recipe shape and `## Modes` execution control are orthogonal. Contracts live in `\u003Crecipe>-recipe.md` or `reference\u002Finline-recipes.md`.\n\nNew Recipes must pass `reference\u002Frecipe-contract.md` **Recipe Admission Gate**; single-specialist workflows are direct routes and MUST NOT enter the registry. Internal phases and invocation modes are not Recipes.\n\n### Recipe Families (mental model + within-family disambiguation)\n\nFamilies disambiguate siblings by one axis; ambiguous anchors (`improve`\u002F`polish`\u002F`enhance`, `evolve a feature`) use the one-question REDIRECT (`reference\u002Fintent-clarification.md`). The overloaded families are **Improve**, **Loop** (completion oracle; every loop first passes `_common\u002FLOOP_PRECONDITIONS.md`), **Reproduce, Synthesize & Invent** (source count; `_common\u002FDIFFERENTIAL_PARITY.md`), and **Quality-Max** (expensive, confirm). Membership and full axes → `reference\u002Frecipes-detail.md` § Recipe Families.\n\n### Recipe Registry\n\n**Full table** → **`reference\u002Frecipes-index.md`** (read on subcommand match, or when scanning). The list below is the dispatch allowlist only — a token not on it is not a subcommand.\n\n```\nbug · feature · deliver · security · refactor · optimize · kaizen · anneal · restyle · converge · apex\ncharter · enact · layer · goal · gedanken · delve · cartograph · chronicle · verity · abide · spec · essential · killer · trim\nacceptance · summit · podium · newsroom · wish · eureka · runway · hallmark · rebrand · crucible · silhouette\nlattice · chorus · assay · migrate · transmute · clone · fuse · graft · package · pack · quell · burnish · whet\n```\n\nNo Recipe is the default; unmatched input uses the explicit Default dispatch below. Named preset aliases (`venture` \u002F `marquee` \u002F `growth-acceptance`) → **Subcommand Dispatch**.\n\n## Subcommand Dispatch\n\n**Default dispatch:** `phase:CLASSIFY` with flow `RESOLVE → GATE → MULTI? → REDIRECT? → SELECT → LADDER? → CHAIN_SELECT`.\n\nDispatch in this order:\n- `\u002FNexus` with no arguments → proactive mode (`reference\u002Fproactive-mode.md`); this is an invocation mode, not a Recipe.\n- Matches a Recipe Subcommand → skip CLASSIFY, pass the Chain Template to CHAIN_SELECT, and read that Recipe's `Read` reference for full phase contracts first. **Subject to the bare-subcommand exception below.**\n- **Bare-subcommand exception.** A matched Recipe token with no object, target, scope, or metric (`\u002Fnexus optimize`, `\u002Fnexus kaizen`) enters CLASSIFY at `GATE` and asks one focused question. **Exempt:** `pack`. Fixtures: `task-battery.md` 31-32; guard: `routing-oracle.py` RO-6.\n- **Named preset aliases** dispatch to their engine + preset and behave identically to the explicit form; read **both** the engine reference and the preset blueprint. `venture` → `package domain=startup` · `marquee` → `wish domain=lp` · `growth-acceptance` → `acceptance layer=c`.\n- Otherwise → enter the internal CLASSIFY phase using the Default dispatch above. **REDIRECT** prefers a matching Recipe over an ad-hoc chain. Full contract → `reference\u002Frouting-matrix.md` § Classify Flow.\n\nInline Recipes (`kaizen`, `essential`, `killer`, `trim`) have no top-level reference — contracts in `reference\u002Finline-recipes.md`.\n\n## Workflow\n\n`CLASSIFY → CHAIN → EXECUTE → AGGREGATE → VERIFY → DELIVER` `(+ LEARN post-chain)`\n\n| Phase | Purpose | Read When |\n|------|---------|-----------|\n| `CLASSIFY` | Task type, complexity, confidence, guardrail needs; crystallize the intent contract (goal + ACs + non-goals + prohibited outcomes) | `confidence-scoring.md`, `intent-clarification.md`, `autonomy-quality-protocol.md` (Q1-Q3) |\n| `CHAIN` | Minimum viable chain, parallel branches, Plan-and-Execute | `routing-matrix.md`, `agent-chains.md`, `agent-disambiguation.md` |\n| `SPECIFY?` | Gated `Chisel brief`; fires on load-bearing ambiguity, ≥3 spawns, loop\u002Fquality-max, or rework, and **runs only after every applicable `Ask First` gate has resolved**. Copy its ACs\u002Fprohibited outcomes\u002Fconstraints\u002Fdelegated list verbatim into every `_AGENT_CONTEXT`; never replace `GATE` or run per-spawn | `specify-phase.md` |\n| `EXECUTE` | Spawn agents (L1\u002FL2\u002FL3) with checkpoints; pass state deltas only | `reference\u002Fexecution-phases.md`, `reference\u002Fguardrails.md`, `reference\u002Ferror-handling.md`, `reference\u002Forchestration-patterns.md` |\n| `AGGREGATE` | Merge branch outputs, validate schema\u002Ffields, goal-alignment check vs the intent contract | `conflict-resolution.md`, `handoff-validation.md`, `autonomy-quality-protocol.md` (Q7-Q8) |\n| `VERIFY` | Acceptance criteria; tests\u002Fbuild\u002Fsecurity mandatory; producer ≠ sole verifier; evidence-bound claims | `reference\u002Fguardrails.md`, `output-formats.md`, `quality-iteration.md`, `autonomy-quality-protocol.md` (Q9-Q15) |\n| `DELIVER` | Final user-facing response | `output-formats.md` |\n| `LEARN` | Adapt routing from evidence after completion | `routing-learning.md` |\n\n## Execution Model\n\n**Orchestrator detection** — establish the host runtime and advertised delegation capabilities before the first spawn; discover actual schemas rather than identifying a host solely from a familiar tool name. Bind the runtime adapter and authorized model choice. Detection table, per-CLI prereqs, model selection, adaptive-prompt policy, canonical spawn template → `reference\u002Fhub-authoring.md` § Execution Model + `reference\u002Fexecution-layers.md`.\n\n**Spawn decision** — Core Rule #3 decides: no spawn tool → internal (log the verified blocker); specialist expertise → spawn (mandatory); trivial edit → spawn only if overhead is justified. Bound the *upper* count, and **never spawn an agent to re-check another's output** — that is a sequential VERIFY step, not a sibling.\n\n**Spawn prompt non-negotiables** — front-load ACs (P1), output envelope (P2), scope (P8), completion bound (Q16-Q17), `Prohibited outcomes`, and least-authority `Authority` with `redelegation: false` (Q2\u002FQ23). Never request producer self-verification; use a separate verifier. Adaptive prompt policy applies at ≥3 spawns, loop Recipes, or repeat agents (`reference\u002Fadaptive-prompt-policy.md`). After `SPECIFY`, inject its goal\u002FACs\u002Fprohibited outcomes\u002Fconstraints verbatim before directives.\n\n> Before an external spawn, read `_common\u002FCLI_COMPATIBILITY.md` §9 for supported result channels and permission boundaries. Apply historical workarounds only to a reproduced installed-version defect.\n\n## Safety Contract\n\n- **Guardrails:** `L1` monitor\u002Flog → `L2` auto-verify\u002Fcheckpoint → `L3` pause + auto-recovery → `L4` abort + rollback.\n- **Error handling:** `L1` retry (max 3) → `L2` auto-adjust or inject Builder → `L3` rollback + recovery chain → `L4` ask user (max 5) → `L5` abort. Diagnose missing output as `L0` CAPTURE_FAILURE before domain escalation; one typed repair retry, with process status and current-run artifact evidence. Timeout is not success.\n- **Circuit breaker:** three consecutive failures marks an agent DEGRADED until a probe succeeds; \"Agent Tennis\" (two agents disagreeing 3+ turns without progress) trips the breaker and escalates.\n- **Checkpoint-resume:** chains of 4+ steps persist step outputs at each boundary so interrupted runs resume from the last checkpoint.\n- **Auto-decision:** proceed only at sufficient confidence with acceptable reversibility; confirm risky or irreversible work first. Depth follows the Autonomy Ledger and never relaxes an Ask First gate.\n- **Output validation:** every step output passes schema validation (required fields, status enum, confidence ≥ 0.6) before flowing on; semantic failures need domain checks.\n- **Always confirm:** the **Ask First** triggers. Detail → `reference\u002Fguardrails.md` § Safety Contract.\n\n### LEARN Triggers and Safety\n\nSeven triggers (`LT-01` → `LT-07`, incl. near misses) and the Chain Effectiveness Score formula → `reference\u002Frouting-learning.md`. **Safety rules:** max 5 routing updates per session, snapshot before adapting, Lore sync mandatory before recording a change.\n\n## Routing Quick Start\n\n**Chain source of truth:** `routing-matrix.md` maps task types to default chains and CLASSIFY\u002FLADDER rules; `agent-chains.md` owns chain adjustment. The Registry exposes common types; the rest enter through CLASSIFY. Family axes live in `recipes-detail.md`; phase contracts live in each `\u003Crecipe>-recipe.md`.\n\nIf context is unclear, inspect git state and `.agents\u002FPROJECT.md`; if confidence stays low, ask one focused question.\n\n## Output Requirements\n\nEvery deliverable carries `## NEXUS_COMPLETE`, task + ACs, chain + mode, per-step agent\u002Fstatus\u002Fsummary, verification evidence, and summary status. Four ledgers are non-optional in substance:\n- **Acceptance Provenance** — every intent-contract criterion classified, none silent; prohibited outcomes on their own axis (Q15).\n- **Decision Ledger** — `DEC-n` judgment calls made without the user, interpretation entries first; omit only when empty (Q4-Q6).\n- **Residual Ledger** — each leftover as `RES-n` (class, blocker\u002Fowner, marker location, route), bound bidirectionally to any `#TODO(agent):` left behind, plus the completion-sweep line (`scanned, 0 hits` when clean — never omitted).\n- **`## Prompt Tuning`** — delta-only trace when a spawn's directives were adapted; omit entirely when none were.\n\nScale the envelope to the run: empty ledgers are one line (`Residuals: none`); SIMPLE single-agent runs use `output-formats.md` § Compact Form; per-step results are one line. Never drop the completion sweep.\n\nVerification results are evidence-bound; unexercised paths are labeled `UNVERIFIED` (Q10). Field template → `reference\u002Foutput-formats.md`.\n\n**Required contract:** `DELIVER` returns `NEXUS_COMPLETE` semantics (`reference\u002Foutput-formats.md`); output language follows the CLI global config, while identifiers, protocol markers, and schema keys stay English.\n\n## Collaboration\n\nHandoff directions: agent → Nexus `NEXUS_ROUTING` · Nexus → agent `_AGENT_CONTEXT` · agent → Nexus `_STEP_COMPLETE` · Nexus → user `NEXUS_COMPLETE` · Architect → Nexus `ARCHITECT_TO_NEXUS_HANDOFF` · Judge → Nexus `QUALITY_FEEDBACK` · Nexus → Nexus `ROUTING_ADAPTATION_LOG`. Project-local Lore\u002FDarwin handoffs require the availability gate. Schemas → `reference\u002Foutput-formats.md`.\n\n## Reference Map\n\nRead only files matching the current decision point. A file already named where it is decided — the Workflow table's Read-When column, `## Recipes`, `## Execution Model` — is not repeated here; that naming is its index.\n\n| File | Read When |\n|------|-----------|\n| **`reference\u002Freference-index.md`** | Full Read-When index for references not listed below |\n| `reference\u002Fdeliver-recipe.md` | Scope-adaptive product\u002FMVP delivery, chain sizing, anti-stall recovery, and Delivery Report |\n| `reference\u002F\u003Crecipe>-recipe.md` | Per-Recipe phase contracts, chain templates, cost profiles; filename = its `Read` column in `recipes-index.md` |\n| `_common\u002FLOOP_PRECONDITIONS.md` | Before **any** agent loop — five-point gate (completion oracle · hard-stop bound · maker ≠ checker · persistent memory · drift awareness) |\n| `_common\u002FPROJECT_LOCAL_SKILLS.md` | Before selecting `orbit`, `lore`, or `darwin`; workspace availability check and global fallback |\n| `_common\u002FFINDING_LEDGER.md` | Before **any external-reviewer-to-zero loop** (`quell`, `burnish`, `whet`, `newsroom`) — the shared ledger machinery: five declaration slots, identity across cycles, disposition vocabulary + integrity, split-oracle rule, and when **not** to build one |\n| `_common\u002FPROOF_CARRYING.md` | `\u002Fnexus acceptance` Tier policy + G1-G10. **Mandatory before `acceptance`.** |\n| `_common\u002FPARALLEL.md` · `reference\u002Fsignal-keywords.md` | Parallel branch definitions, file ownership, merge, rollback · canonical Signal Keywords → routing destination table |\n\n## Operational\n\n**Spine contracts** — in effect on every run, precedence in `_common\u002FOPERATIONAL.md` § Contract Precedence: `_common\u002FVALUES.md` · `_common\u002FBOUNDARIES.md` · `_common\u002FHANDOFF.md` · `_common\u002FAUTORUN.md` · `_common\u002FGIT_GUIDELINES.md` · `_common\u002FOUTPUT_STYLE.md` · `_common\u002FOPUS_5_AUTHORING.md` · `_common\u002FWORK_GATE.md`.\n\nBeyond the spine, follow `_common\u002FHARNESS_EVOLUTION.md`. Apply the hub-engine protocol: `_common\u002FOPUS_5_AUTHORING.md` (shared P1-P12), plus `CODEX_ORCHESTRATION.md` or `AGY_ORCHESTRATION.md` (A1-A9) when applicable. Journal in `.agents\u002Fnexus.md`, log to `.agents\u002FPROJECT.md`, no agent names in commits\u002FPRs. Keep chains small, handoffs structured, recovery explicit.\n\n## Operational Notes for Spawns\n\nPer spawn: weighted confidence; current-phase references only (skip anti-pattern refs under 4 agents); `_STEP_COMPLETE`\u002F`NEXUS_HANDOFF` needs Summary + Status + Next; track Phase + Step and full `_NEXUS_STATE` at 4+ steps; prompt for task\u002Foutput, not personality. Detail → `reference\u002Fhub-authoring.md`.\n\n## AUTORUN Support\n\nProtocol → `_common\u002FAUTORUN.md`; mode semantics → **Modes**. `AUTORUN` appends `_STEP_COMPLETE:` with `Agent`, `Status`, `Output`, `Next` after normal work; Output\u002FNext schema → `reference\u002Fautorun-schema.md`.\n\n## Nexus Hub Mode\n\nHub mode takes `## NEXUS_ROUTING` and returns `## NEXUS_HANDOFF` (`_common\u002FHANDOFF.md`), never a direct agent-to-agent call. Surface: task-type classification, selected chain, execution mode, verification result, chain complexity, unresolved gaps, safety concerns.\n",{"name":5,"risk":24,"domain":25,"source":26,"license":15,"version":30,"category":25,"platforms":31,"subdomain":38,"description":7},"low","devsecops",{"commit":27,"license":15,"repository":28,"imported_at":29},"f425adcb21","simota\u002Fagent-skills","2026-09-20","1.0.0",[32,33,34,35,36,37],"claude-code","cursor","codex","gemini","antigravity","copilot","general","main","2d1750717747097b602acf30bd09a7cbd4b4c6ea","awesome_skills\u002Fdevsecops\u002Fnexus","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous","2026-11-09 11:45:36",{"source":8,"skillName":5,"commands":46,"prompt":56,"zipAvailable":57},[47,50,53],{"tool":48,"command":49},"npx","npx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill nexus",{"tool":51,"command":52},"bunx","bunx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill nexus",{"tool":54,"command":55},"pnpm","pnpm dlx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill nexus","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fdevsecops\u002Fnexus\u002FSKILL.md",true,[59,65],{"path":60,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":63,"sourceHtmlUrl":64},"_common","OTHER","其他",10,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fdevsecops\u002Fnexus\u002F_common",{"path":66,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":67,"sourceHtmlUrl":68},"oracle",9,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fdevsecops\u002Fnexus\u002Foracle",[70,78,85,92,99,106],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":77},"01M7FFFZPC2025JQQHPHDX154J","dei-assessment","Assess diversity, equity, and inclusion with metrics. TRIGGERS - Use when user needs help with dei-assessment related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fdei-assessment\u002FSKILL.md",[],[],"2026-11-16 10:05:34",{"skillNo":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":77},"01M7FFFWN8H4WA7A6Y398QKB6J","dei-action-plan","Create DEI action plans with goals and initiatives. TRIGGERS - Use when user needs help with dei-action-plan related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fdei-action-plan\u002FSKILL.md",[],[],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"publishedAt":77},"01M7FFFSHTM0QZRVZ2NRJTJ0PY","culture-fit-assessment","Design culture add assessment processes. TRIGGERS - Use when user needs help with culture-fit-assessment related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fculture-fit-assessment\u002FSKILL.md",[],[],{"skillNo":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":77},"01M7FFFMA12RKEQA38KHJSXXPS","cowork-to-code-bridge","Use an already-installed, independently verified cowork-to-code bridge to run narrowly approved actions on the user's own macOS, Linux, or WSL2 machine through a local file queue.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fcowork-to-code-bridge\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":103,"categories":104,"keywords":105,"publishedAt":77},"01M7FFFH1F6M8YRGW5V487GH1K","counter-offer-response","Draft counter-offer responses for retention. TRIGGERS - Use when user needs help with counter-offer-response related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fcounter-offer-response\u002FSKILL.md",[],[],{"skillNo":107,"name":108,"nameZh":6,"description":109,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":110,"categories":111,"keywords":112,"publishedAt":77},"01M7FFFDGEHV8Y7J9N40HZXW52","continuous-feedback-system","Design continuous feedback systems. TRIGGERS - Use when user needs help with continuous-feedback-system related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fcontinuous-feedback-system\u002FSKILL.md",[],[]]