[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75942DP7FZWXFNJ9ECAEWSD":3,"skill-same:01M75942DP7FZWXFNJ9ECAEWSD":43},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M75942DP7FZWXFNJ9ECAEWSD","freeforge-orchestration",null,"Primary multi-agent workflow for complex changes to this repository. Use when a task is large, unfamiliar, cross-cutting, or spans multiple files\u002Fmodules and independent exploration, security specialization, or fresh verification would materially reduce risk. Defines the default topology, delegation contract, and adjudication procedure for coordinating subagents without expanding their authority beyond AGENTS.md.","JackSmack1971\u002Ffreeforge-chat","JackSmack1971","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128181597?v=4","01M6KS1FH32N6WNWZ68JC9FFDJ","01M7593JRZ0MV9B0VVTRGWZ0WC",0,"MIT","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Ffreeforge-orchestration\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: freeforge-orchestration\ndescription: Primary multi-agent workflow for complex changes to this repository. Use when a task is large, unfamiliar, cross-cutting, or spans multiple files\u002Fmodules and independent exploration, security specialization, or fresh verification would materially reduce risk. Defines the default topology, delegation contract, and adjudication procedure for coordinating subagents without expanding their authority beyond AGENTS.md.\n---\n\n# Freeforge Orchestration\n\n## Purpose\n\nCoordinate subagents for complex changes using a fixed, low-fan-out topology:\nread-only explorers → synthesis → one implementation owner → fresh reviewer +\nverifier → primary adjudication. This skill governs *how* to delegate; it does\nnot grant any subagent authority beyond what `AGENTS.md` already permits for\nits role (Explorer, Implementer, Verifier, Security reviewer, Documentation\nmaintainer, Release operator).\n\n## When to delegate\n\nDelegation is justified only when at least one of these holds:\n\n- **Genuine isolation** — a subtask benefits from not sharing context\u002Fbias\n  with the primary thread (e.g. an adversarial verifier that should not see\n  the implementer's self-report).\n- **Independent exploration** — the codebase surface is unfamiliar or large\n  enough that parallel read-only reconnaissance is faster and safer than\n  serial exploration.\n- **Security specialization** — the change touches auth, secrets, crypto,\n  external input, deserialization, exec, network, migrations, concurrency,\n  CI\u002FCD, or dependencies (see `AGENTS.md` security constraints and the\n  `security-and-tools` skill).\n- **Fresh verification** — the implementer should not grade its own work;\n  a reviewer\u002Fverifier with no stake in the diff must check it.\n\nDo not delegate a task that is small, single-file, or already well\nunderstood — do it directly. Do not use majority vote across multiple\nimplementers, recursive subagent spawning, or more than one writer touching\noverlapping code at the same time.\n\n## Default topology\n\n```\nread-only explorers  →  synthesis  →  one implementation owner\n                                            │\n                                            ▼\n                          fresh reviewer  +  fresh verifier\n                                            │\n                                            ▼\n                              primary adjudication\n```\n\n- **Explorers** (0–4, read-only): independent reconnaissance of distinct\n  scopes. Never write, never overlap scope with another explorer.\n- **Synthesis**: the primary thread reconciles explorer findings against\n  source evidence (files, tests, history) — not against explorer claims\n  alone — and resolves conflicts before any code is written.\n- **Implementation owner**: exactly one subagent (or the primary thread\n  itself) owns any given piece of overlapping code. Never split one\n  cross-cutting change across multiple concurrent writers.\n- **Reviewer + verifier**: dispatched fresh (no prior exposure to the\n  implementer's reasoning), working from the actual diff and executed\n  evidence, not the implementer's narrative. Add a `security_reviewer` when\n  the change is security-sensitive per `AGENTS.md`.\n- **Primary adjudication**: the primary thread — not a subagent — makes the\n  final accept\u002Frevise\u002Fescalate call, using the diff and evidence, not\n  agreement counts.\n\n## Procedure\n\n1. **Inspect baseline\u002Fdirty state.** Run read-only checks (`git status`,\n   `git diff`, relevant file reads) before any delegation or mutation.\n   Treat prior plans, summaries, or tickets as hypotheses until confirmed\n   against current repository state.\n2. **Decide whether delegation is justified** using the criteria above. If\n   not justified, do the work directly and skip the rest of this procedure.\n3. **Dispatch at most 2–4 bounded workers.** Fewer, tightly scoped workers\n   beat many broad ones.\n4. **Give each worker an explicit assignment** containing:\n   - `baseline` — the commit\u002Fstate it is starting from;\n   - `scope` — exactly which files\u002Fareas are in bounds;\n   - `objective` — what it must produce;\n   - `non_goals` — what it must not do or touch;\n   - `applicable_instructions` — the relevant `AGENTS.md` sections, nested\n     `AGENTS.md` files, and skills (e.g. `security-and-tools`,\n     `surgical-change-control`) it must follow;\n   - `evidence_required` — what proof of work it must return (file paths,\n     command output, diff hunks — not prose claims);\n   - `return_schema` — the schema in \"Delegation result schema\" below.\n5. **Wait for independent read-only findings** from all explorers before\n   synthesizing. Do not let explorers see each other's in-progress output.\n6. **Synthesize conflicts against source evidence.** Where explorers\n   disagree, resolve by re-checking the repository\u002Ftests directly, not by\n   picking a majority.\n7. **Assign exactly one implementation owner** for any overlapping code\n   surface identified during synthesis.\n8. **Obtain fresh reviewer and verifier passes** after implementation,\n   working from the Git diff and independently executed checks. Add\n   `security_reviewer` for security-sensitive changes.\n9. **Adjudicate** from the Git diff plus executed evidence — never from\n   subagent self-reports alone. Confirm the diff matches the stated scope\n   and that no unrelated files changed.\n10. **Stop** when acceptance criteria are met, or report the specific\n    blocker if one remains (per `AGENTS.md` \"Mandatory escalation\").\n\n## Delegation result schema\n\nEvery dispatched worker returns:\n\n```json\n{\n  \"status\": \"ok | blocked | partial\",\n  \"baseline\": \"commit\u002Fstate the worker started from\",\n  \"scope\": \"files\u002Fareas actually touched or examined\",\n  \"observations\": \"findings relevant to the objective, with evidence pointers\",\n  \"changes\": \"diff summary — only present if the worker is a writer\",\n  \"verification\": \"commands\u002Fchecks run and their results\",\n  \"risks\": \"known gaps, assumptions, or residual risk\",\n  \"recommended_next_action\": \"what the primary thread should do next\"\n}\n```\n\nWorkers that are not writers omit `changes`. A worker reporting `status:\nok` with no `verification` evidence has not actually verified anything —\ntreat that as `partial`.\n\n## Non-goals\n\n- This skill does not grant subagents any tool access, filesystem scope, or\n  network\u002Fsecret permission beyond what `AGENTS.md` and the runtime already\n  allow for their role. It is a coordination contract, not a permission\n  grant.\n- Do not port the legacy 7-axes workflow (previously under `.claude\u002F` \u002F\n  `.codex\u002F`) into this skill. Only add topology elements here when a\n  concrete repository need demonstrates they are missing — this skill stays\n  minimal by default.\n- No majority vote, no recursive subagent spawning, no overlapping writers.\n\n## Escalation\n\nIf applicable instructions materially conflict, a worker requests authority\noutside its role, or scope grows substantially beyond the original task,\nstop and escalate per `AGENTS.md` \"Mandatory escalation\" instead of\nresolving it silently inside the orchestration loop.\n","# Freeforge Orchestration\n\n## Purpose\n\nCoordinate subagents for complex changes using a fixed, low-fan-out topology:\nread-only explorers → synthesis → one implementation owner → fresh reviewer +\nverifier → primary adjudication. This skill governs *how* to delegate; it does\nnot grant any subagent authority beyond what `AGENTS.md` already permits for\nits role (Explorer, Implementer, Verifier, Security reviewer, Documentation\nmaintainer, Release operator).\n\n## When to delegate\n\nDelegation is justified only when at least one of these holds:\n\n- **Genuine isolation** — a subtask benefits from not sharing context\u002Fbias\n  with the primary thread (e.g. an adversarial verifier that should not see\n  the implementer's self-report).\n- **Independent exploration** — the codebase surface is unfamiliar or large\n  enough that parallel read-only reconnaissance is faster and safer than\n  serial exploration.\n- **Security specialization** — the change touches auth, secrets, crypto,\n  external input, deserialization, exec, network, migrations, concurrency,\n  CI\u002FCD, or dependencies (see `AGENTS.md` security constraints and the\n  `security-and-tools` skill).\n- **Fresh verification** — the implementer should not grade its own work;\n  a reviewer\u002Fverifier with no stake in the diff must check it.\n\nDo not delegate a task that is small, single-file, or already well\nunderstood — do it directly. Do not use majority vote across multiple\nimplementers, recursive subagent spawning, or more than one writer touching\noverlapping code at the same time.\n\n## Default topology\n\n```\nread-only explorers  →  synthesis  →  one implementation owner\n                                            │\n                                            ▼\n                          fresh reviewer  +  fresh verifier\n                                            │\n                                            ▼\n                              primary adjudication\n```\n\n- **Explorers** (0–4, read-only): independent reconnaissance of distinct\n  scopes. Never write, never overlap scope with another explorer.\n- **Synthesis**: the primary thread reconciles explorer findings against\n  source evidence (files, tests, history) — not against explorer claims\n  alone — and resolves conflicts before any code is written.\n- **Implementation owner**: exactly one subagent (or the primary thread\n  itself) owns any given piece of overlapping code. Never split one\n  cross-cutting change across multiple concurrent writers.\n- **Reviewer + verifier**: dispatched fresh (no prior exposure to the\n  implementer's reasoning), working from the actual diff and executed\n  evidence, not the implementer's narrative. Add a `security_reviewer` when\n  the change is security-sensitive per `AGENTS.md`.\n- **Primary adjudication**: the primary thread — not a subagent — makes the\n  final accept\u002Frevise\u002Fescalate call, using the diff and evidence, not\n  agreement counts.\n\n## Procedure\n\n1. **Inspect baseline\u002Fdirty state.** Run read-only checks (`git status`,\n   `git diff`, relevant file reads) before any delegation or mutation.\n   Treat prior plans, summaries, or tickets as hypotheses until confirmed\n   against current repository state.\n2. **Decide whether delegation is justified** using the criteria above. If\n   not justified, do the work directly and skip the rest of this procedure.\n3. **Dispatch at most 2–4 bounded workers.** Fewer, tightly scoped workers\n   beat many broad ones.\n4. **Give each worker an explicit assignment** containing:\n   - `baseline` — the commit\u002Fstate it is starting from;\n   - `scope` — exactly which files\u002Fareas are in bounds;\n   - `objective` — what it must produce;\n   - `non_goals` — what it must not do or touch;\n   - `applicable_instructions` — the relevant `AGENTS.md` sections, nested\n     `AGENTS.md` files, and skills (e.g. `security-and-tools`,\n     `surgical-change-control`) it must follow;\n   - `evidence_required` — what proof of work it must return (file paths,\n     command output, diff hunks — not prose claims);\n   - `return_schema` — the schema in \"Delegation result schema\" below.\n5. **Wait for independent read-only findings** from all explorers before\n   synthesizing. Do not let explorers see each other's in-progress output.\n6. **Synthesize conflicts against source evidence.** Where explorers\n   disagree, resolve by re-checking the repository\u002Ftests directly, not by\n   picking a majority.\n7. **Assign exactly one implementation owner** for any overlapping code\n   surface identified during synthesis.\n8. **Obtain fresh reviewer and verifier passes** after implementation,\n   working from the Git diff and independently executed checks. Add\n   `security_reviewer` for security-sensitive changes.\n9. **Adjudicate** from the Git diff plus executed evidence — never from\n   subagent self-reports alone. Confirm the diff matches the stated scope\n   and that no unrelated files changed.\n10. **Stop** when acceptance criteria are met, or report the specific\n    blocker if one remains (per `AGENTS.md` \"Mandatory escalation\").\n\n## Delegation result schema\n\nEvery dispatched worker returns:\n\n```json\n{\n  \"status\": \"ok | blocked | partial\",\n  \"baseline\": \"commit\u002Fstate the worker started from\",\n  \"scope\": \"files\u002Fareas actually touched or examined\",\n  \"observations\": \"findings relevant to the objective, with evidence pointers\",\n  \"changes\": \"diff summary — only present if the worker is a writer\",\n  \"verification\": \"commands\u002Fchecks run and their results\",\n  \"risks\": \"known gaps, assumptions, or residual risk\",\n  \"recommended_next_action\": \"what the primary thread should do next\"\n}\n```\n\nWorkers that are not writers omit `changes`. A worker reporting `status:\nok` with no `verification` evidence has not actually verified anything —\ntreat that as `partial`.\n\n## Non-goals\n\n- This skill does not grant subagents any tool access, filesystem scope, or\n  network\u002Fsecret permission beyond what `AGENTS.md` and the runtime already\n  allow for their role. It is a coordination contract, not a permission\n  grant.\n- Do not port the legacy 7-axes workflow (previously under `.claude\u002F` \u002F\n  `.codex\u002F`) into this skill. Only add topology elements here when a\n  concrete repository need demonstrates they are missing — this skill stays\n  minimal by default.\n- No majority vote, no recursive subagent spawning, no overlapping writers.\n\n## Escalation\n\nIf applicable instructions materially conflict, a worker requests authority\noutside its role, or scope grows substantially beyond the original task,\nstop and escalate per `AGENTS.md` \"Mandatory escalation\" instead of\nresolving it silently inside the orchestration loop.\n",{"name":5,"description":7},"main","a5bed13dcca926e006d2681d548e722fffa499ab",".agents\u002Fskills\u002Ffreeforge-orchestration","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat","https:\u002F\u002Fgithub.com\u002FJackSmack1971","2026-10-02 05:58:57",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add JackSmack1971\u002Ffreeforge-chat --skill freeforge-orchestration",{"tool":35,"command":36},"bunx","bunx skills add JackSmack1971\u002Ffreeforge-chat --skill freeforge-orchestration",{"tool":38,"command":39},"pnpm","pnpm dlx skills add JackSmack1971\u002Ffreeforge-chat --skill freeforge-orchestration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Ffreeforge-orchestration\u002FSKILL.md",true,[],[44,51,57,62,67,72],{"skillNo":45,"name":46,"nameZh":6,"description":47,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":48,"categories":49,"keywords":50,"publishedAt":18},"01M7596K2F59EBMTV0GEA7G1DF","sample","Fixture skill.","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002Ftests\u002Fcontrol-plane\u002Ffixtures\u002Fsnapshot-good\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":52,"name":46,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":54,"categories":55,"keywords":56,"publishedAt":18},"01M7596HMJGJCK6C1BBBP6YFWM","A minimal fixture skill used only to exercise the verifier.","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002Ftests\u002Fcontrol-plane\u002Ffixtures\u002Fmalformed\u002Fwriter-configured-read-only\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":58,"name":46,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":18},"01M7596FVEYPJ99D4XJZZ07HB1","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002Ftests\u002Fcontrol-plane\u002Ffixtures\u002Fmalformed\u002Fstale-identifier-under-codex\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":63,"name":46,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":18},"01M7596ECKJNJE0G09SFTENYTW","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002Ftests\u002Fcontrol-plane\u002Ffixtures\u002Fmalformed\u002Freviewer-configured-write-enabled\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":68,"name":46,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":18},"01M7596BMRAMMD4P3MW4N25ZAV","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002Ftests\u002Fcontrol-plane\u002Ffixtures\u002Fmalformed\u002Fno-rules-dir\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[],{"skillNo":73,"name":46,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":18},"01M759695BRBKRSK4KDY4VCFR4","https:\u002F\u002Fgithub.com\u002FJackSmack1971\u002Ffreeforge-chat\u002Fblob\u002Fmain\u002Ftests\u002Fcontrol-plane\u002Ffixtures\u002Fmalformed\u002Fmissing-skill-md\u002F.agents\u002Fskills\u002Fsample\u002FSKILL.md",[],[]]