[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78D35J2PTB2EWVJD59ZVKBJ":3,"skill-same:01M78D35J2PTB2EWVJD59ZVKBJ":61},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"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":24,"commitSha":25,"skillDirectory":26,"forksCount":13,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":9,"updatedAt":29,"install":30,"resources":42},"01M78D35J2PTB2EWVJD59ZVKBJ","parley-delegate",null,"Delegate tasks to other agent CLIs (e.g codex, grok, opencode, pi, etc) with the parley CLI. Use when the user asks to delegate to other agents, or mentions parley.","femoral\u002Fparley","femoral","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F50255779?v=4","01M78D245DQ5V4YYJE7Y3XCBTV","01M78D2464YVC6B4Y2DXD26FTK",0,"MIT","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-delegate\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: parley-delegate\ndescription: Delegate tasks to other agent CLIs (e.g codex, grok, opencode, pi, etc) with the parley CLI. Use when the user asks to delegate to other agents, or mentions parley.\ndisable-model-invocation: true\n---\n\n# Delegating to parley\n\nParley runs a child agent CLI in an isolated git worktree and hands you back a schema-validated **report envelope**. You are the orchestrator: you write the brief, answer the child's questions, and review\u002Fmerge the branch. Parley never merges.\n\n**Orchestrate directly — don't wrap parley in subagents.** Unless the user explicitly asks for subagents, the session reading this skill is the orchestrator: it calls `delegate`, blocks on `watch`, answers questions, and reviews branches itself. Per-task babysitter subagents add a token layer, lose the question-answering context you already have, and tend to idle-stop waiting for notifications that never come.\n\n## Step 0 — load live config\n\nAt session start, before the first `delegate`:\n\n```\nparley info\n```\n\nTreat that output as **authoritative** for this project: vendors and profiles, valid `--type` ids, size\u002Fdifficulty classification, whether evaluation is on (and how to record evals), and fix\u002Fretry policy with error codes. Do not invent those values; re-run `parley info` if the project may have changed.\n\n**When evaluation is on**, install the harness plugin for your orchestrator (exports `PARLEY_SESSION_ID` \u002F `PARLEY_HARNESS` \u002F `PARLEY_MODEL` \u002F `PARLEY_EFFORT`), then register (or re-anchor after crash\u002Frestart):\n\n```\nparley session\n# re-anchor a known id (or rely on PARLEY_SESSION_ID from the plugin):\nparley session -s \u003Cid>\n```\n\nProvenance is env-only — models must not invent model\u002Feffort values. Missing env vars register as unknown. Prefer `PARLEY_SESSION_ID` from the plugin for later commands; `--session` is a fallback. See [sessions.md](sessions.md).\n\n**Setup problems** (missing vendors, unconfigured project, eval misconfig): stop and run `\u002Fparley-wizard` with the user. Do not invent config mid-orchestration.\n\n## The delegate loop\n\nThere is **one** flow for one task or many: `delegate` always returns immediately; `watch` is the only wait. The single-task case is the fan-out loop with n=1.\n\n1. **Write a self-contained brief.** The child sees only its worktree, your prompt, and `--context` files — none of your conversation. State the goal, constraints, and definition of done (including the project's real typecheck\u002Ftest commands) in the prompt; pass supporting files with `--context \u003Cfile>` (repeatable). Done when a stranger could execute the brief without asking you what it means.\n\n2. **Delegate** (returns immediately with pending-task JSON):\n\n   ```\n   parley delegate -v \u003Cvendor> -m \u003Cmodel> -n \u003Cshort-name> --session \u003Cid> \\\n     [--type \u003Cid>] [--size \u003Cid>] [--difficulty \u003Cid>] \"\u003Cbrief>\"\n   ```\n\n   (`-` as the prompt reads stdin — use a heredoc for long briefs.) A vendor (`-v`, or a `--profile` that names one) and a session id (`--session \u003Cid>`, or `PARLEY_SESSION_ID` in the environment) are both required — missing either is a usage error (exit 2). `delegate` and `answer` exit only `0` (accepted) or `2` (usage).\n\n   Choose vendor, profile, model, and classification from what `parley info` listed:\n\n   - **`--type \u003Cid>`** — work-domain type; valid ids come from info (omit ⇒ the automatic fallback listed there).\n   - **`--size \u003Cid>` \u002F `--difficulty \u003Cid>`** — optional classification for metrics; ids and guidance from info.\n   - **`--profile \u003Cname>`** — prefer a named profile when the user has them (tracked per task; makes `parley metrics` comparisons meaningful). Explicit flags win over profile defaults.\n   - **`--runner \u003Cname>`** — targets a configured remote runner; the task's commits come back as a pushed branch (no local worktree to review — fetch it).\n\n3. **Run the watch ack-loop** until exit 0 — even for a single task. This is a workflow you step through, not a script: each `watch` call returns one event, you do the real work it demands (answer, review, merge), then call `watch` again.\n\n   ```\n   run `parley watch --json`\n   repeat:\n     if exit 0            → all-done: every task terminal, every event acked. Stop.\n     otherwise            → one event arrived; note its `.seq` (the event id)\n       handle it per the table below (answer \u002F triage \u002F review-and-merge)\n       questions (3\u002F4)    → run `parley watch --json`            (answering acked it)\n       failed\u002Fcompleted (5\u002F6) → run `parley watch --json --ack \u003Cseq>`  (you ack it, now that it's handled)\n   ```\n\nWhat each exit code means and what you must do before acking:\n\n| `$?` | event | your move | how it gets acked |\n|---|---|---|---|\n| 3 | `task.question` | `parley answer \u003Ctask> \"\u003Canswer>\"` | **automatic** — answering moves the task out of `awaiting_answer`, which resolves the event. Do not pass `--ack` for it. |\n| 4 | `task.stalled` | `parley answer \u003Ctask> \"…\"` resumes it | **automatic** on resume, same as 3 |\n| 5 | `task.failed` | triage (see \"When a task fails\") | **explicit**: next `watch --ack \u003Cseq>` — only after you've triaged |\n| 6 | `task.completed` | review the branch, merge-or-reject, typecheck, `parley clean` (step 4) | **explicit**: next `watch --ack \u003Cseq>` — only after the review is done |\n| 0 | — (all-done) | stop looping; nothing is pending and nothing is running | — |\n| 2 | usage error | fix your invocation | — |\n\nRules that leave no room for interpretation:\n\n- **Ack means \"I handled this\", never \"I saw this\".** Ack a `completed` event only after its branch is reviewed and merged-or-rejected; ack a `failed` event only after triage. Acking early deletes your only reminder — the task drops out of the inbox and nothing will resurface it.\n- **Un-acked events redeliver.** If you crash or forget between delivery and ack, the next `watch` hands you the same event again. That is the safety net — lean on it; never ack defensively \"to clear the queue\".\n- **Exit 6 is not \"done\".** A completed task is *work for you* (review, merge, verify, clean). The loop is finished only at exit 0.\n- **Level-triggered, race-free.** An event already pending when `watch` starts returns immediately. There is no startup race and no sequence bookkeeping on your side; the only seq you ever touch is the one you pass back to `--ack`.\n- Positional task refs (`parley watch t1 t2`) narrow the inbox to those tasks; the default is every task in the session.\n\n4. **Review and integrate.** On exit 6 the envelope carries the worktree path, branch (`parley\u002F\u003Cid>-\u003Cname>`), and the report body. Review the diff on the branch, merge if it holds up, then `parley clean \u003Ctask>` (removes the worktree, keeps the branch). Ack only after that review. Done when the branch is merged-or-rejected and the worktree cleaned.\n\n   **A green report isn't proof correctness.** `outcome: success` only means the child's own verification passed. Verify yourself after every merge, not once at the end of a fan-out unless instructed. A later branch can reintroduce what an earlier one had cleared.\n\n   **Eval when expected.** When step 0 showed evaluation on, record an eval on every reviewed task:\n\n   ```\n   parley eval \u003Ctask> --answers '\u003Cjson>' --feedback \"\u003Cone line per criterion>\"\n   ```\n\n   Map every rubric criterion id (from `parley info`) to a boolean. The daemon computes score and baseline — do not pass a free-form score. Answer honestly, including failures. A later call overwrites the previous result for that task.\n\n5. **Fix loop.** When a completed task needs more work (review found gaps, or you have a concrete fix brief):\n\n   ```\n   parley fix \u003Ctask> \"\u003Cwhat to fix>\"\n   ```\n\n   That creates a linked reattempt (inherits profile\u002Fworkspace; may resume the parent vendor session — policy from `parley info`). Drive it with the same watch loop.\n\n   If `parley fix` exits with `retry_limit_exceeded` (exit 7) or `reattempt_window_expired` (exit 8):\n\n   ```\n   parley fix --fresh \u003Ctask> \"\u003Cwhat to fix>\"\n   ```\n\n   `--fresh` starts a blank session, is uncapped by resume retry limits, and stays in the attempt chain. Exact limits and window live in `parley info` — re-read them when unsure.\n\n## Fan-out: several tasks in parallel\n\nEach task gets its own worktree, so parallel tasks never collide. Batch them however it makes sense, then drive the **entire** set with the same watch loop above:\n\n```\nparley delegate -v \u003Cvendor> -n task-a --session \u003Cid> --type \u003Cid> \"\u003Cbrief A>\"\nparley delegate -v \u003Cvendor> -n task-b --session \u003Cid> --type \u003Cid> \"\u003Cbrief B>\"\n```\n\nDo not poll `status` on an interval and do not sleep-and-check. One mechanism for n=1 and n=N.\n\n## Session ID\n\nThe session ID identifies the current orchestration session. Resolution is env-first: `PARLEY_SESSION_ID` > `--session \u003Cid>` > session-state file > ancestry binding to a registered session. Install the harness plugin so provenance (session id + harness\u002Fmodel\u002Feffort) is set for you — via env vars or the INTERIM state-file channel; see [sessions.md](sessions.md).\n\n## Context files\n\n`--context \u003Cfile>` is repeatable; each file lands in the worktree under `.parley\u002Fcontext\u002F`, materialized by **basename**.\n\n```\nparley delegate -v \u003Cvendor> -n task-a --session \u003Cid> \\\n  --context \u002Fpath\u002Fto\u002Fconfig.json \\\n  --context \u002Fpath\u002Fto\u002Fother.json \\\n  \"\u003Cbrief A>\"\n```\n\n### Integrating fan-out branches\n\nWhen several branches share a fork point, review each branch on its own and resolve conflicts at merge time. When tasks share files or one builds on another, prefer **dependency waves** over blind parallelism: merge task A first, then delegate task B with `--base-ref` on the freshly-merged target so it forks from its actual prerequisite.\n\n## Beyond the golden path\n\nOne-liner pointers — read the linked file only when its condition fires:\n\n- **Non-default task shapes** — structured `--report-schema` results, no git worktree `--cwd`, sandbox postures: read [task-shaping.md](task-shaping.md).\n- **Harness plugins and session provenance** (`PARLEY_SESSION_ID` \u002F `HARNESS` \u002F `MODEL` \u002F `EFFORT`): read [sessions.md](sessions.md).\n\n## When a task fails\n\nCheck the task's `error` field first (`parley status \u003Ctask> --json | jq '.error'`), then `diag.log` in its `logs_dir`, before touching the raw vendor stream — full order and what each layer means: [docs\u002Fagents\u002Ftroubleshooting.md](..\u002F..\u002Fdocs\u002Fagents\u002Ftroubleshooting.md). `parley logs \u003Ctask>` is the last resort; it burns a lot of context on long tasks.\n\nAfter triage, if the failure is fixable with a clearer brief, use the fix loop above (`parley fix` \u002F `parley fix --fresh`) rather than starting an unrelated new delegate when you want a linked attempt chain.\n\n## Reporting parley bugs\n\nParley is in early testing — when parley itself misbehaves (wrong state transition, lost report, worktree damage, a `PARLEY-DIAG` you can't act on), file it upstream after the triage above, whether or not you found a workaround: open an issue on [github.com\u002Ffemoral\u002Fparley](https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley) labelled `needs-triage`, body per [bug-report.md](bug-report.md). Done when the issue URL exists and you've told the user.\n","# Delegating to parley\n\nParley runs a child agent CLI in an isolated git worktree and hands you back a schema-validated **report envelope**. You are the orchestrator: you write the brief, answer the child's questions, and review\u002Fmerge the branch. Parley never merges.\n\n**Orchestrate directly — don't wrap parley in subagents.** Unless the user explicitly asks for subagents, the session reading this skill is the orchestrator: it calls `delegate`, blocks on `watch`, answers questions, and reviews branches itself. Per-task babysitter subagents add a token layer, lose the question-answering context you already have, and tend to idle-stop waiting for notifications that never come.\n\n## Step 0 — load live config\n\nAt session start, before the first `delegate`:\n\n```\nparley info\n```\n\nTreat that output as **authoritative** for this project: vendors and profiles, valid `--type` ids, size\u002Fdifficulty classification, whether evaluation is on (and how to record evals), and fix\u002Fretry policy with error codes. Do not invent those values; re-run `parley info` if the project may have changed.\n\n**When evaluation is on**, install the harness plugin for your orchestrator (exports `PARLEY_SESSION_ID` \u002F `PARLEY_HARNESS` \u002F `PARLEY_MODEL` \u002F `PARLEY_EFFORT`), then register (or re-anchor after crash\u002Frestart):\n\n```\nparley session\n# re-anchor a known id (or rely on PARLEY_SESSION_ID from the plugin):\nparley session -s \u003Cid>\n```\n\nProvenance is env-only — models must not invent model\u002Feffort values. Missing env vars register as unknown. Prefer `PARLEY_SESSION_ID` from the plugin for later commands; `--session` is a fallback. See [sessions.md](sessions.md).\n\n**Setup problems** (missing vendors, unconfigured project, eval misconfig): stop and run `\u002Fparley-wizard` with the user. Do not invent config mid-orchestration.\n\n## The delegate loop\n\nThere is **one** flow for one task or many: `delegate` always returns immediately; `watch` is the only wait. The single-task case is the fan-out loop with n=1.\n\n1. **Write a self-contained brief.** The child sees only its worktree, your prompt, and `--context` files — none of your conversation. State the goal, constraints, and definition of done (including the project's real typecheck\u002Ftest commands) in the prompt; pass supporting files with `--context \u003Cfile>` (repeatable). Done when a stranger could execute the brief without asking you what it means.\n\n2. **Delegate** (returns immediately with pending-task JSON):\n\n   ```\n   parley delegate -v \u003Cvendor> -m \u003Cmodel> -n \u003Cshort-name> --session \u003Cid> \\\n     [--type \u003Cid>] [--size \u003Cid>] [--difficulty \u003Cid>] \"\u003Cbrief>\"\n   ```\n\n   (`-` as the prompt reads stdin — use a heredoc for long briefs.) A vendor (`-v`, or a `--profile` that names one) and a session id (`--session \u003Cid>`, or `PARLEY_SESSION_ID` in the environment) are both required — missing either is a usage error (exit 2). `delegate` and `answer` exit only `0` (accepted) or `2` (usage).\n\n   Choose vendor, profile, model, and classification from what `parley info` listed:\n\n   - **`--type \u003Cid>`** — work-domain type; valid ids come from info (omit ⇒ the automatic fallback listed there).\n   - **`--size \u003Cid>` \u002F `--difficulty \u003Cid>`** — optional classification for metrics; ids and guidance from info.\n   - **`--profile \u003Cname>`** — prefer a named profile when the user has them (tracked per task; makes `parley metrics` comparisons meaningful). Explicit flags win over profile defaults.\n   - **`--runner \u003Cname>`** — targets a configured remote runner; the task's commits come back as a pushed branch (no local worktree to review — fetch it).\n\n3. **Run the watch ack-loop** until exit 0 — even for a single task. This is a workflow you step through, not a script: each `watch` call returns one event, you do the real work it demands (answer, review, merge), then call `watch` again.\n\n   ```\n   run `parley watch --json`\n   repeat:\n     if exit 0            → all-done: every task terminal, every event acked. Stop.\n     otherwise            → one event arrived; note its `.seq` (the event id)\n       handle it per the table below (answer \u002F triage \u002F review-and-merge)\n       questions (3\u002F4)    → run `parley watch --json`            (answering acked it)\n       failed\u002Fcompleted (5\u002F6) → run `parley watch --json --ack \u003Cseq>`  (you ack it, now that it's handled)\n   ```\n\nWhat each exit code means and what you must do before acking:\n\n| `$?` | event | your move | how it gets acked |\n|---|---|---|---|\n| 3 | `task.question` | `parley answer \u003Ctask> \"\u003Canswer>\"` | **automatic** — answering moves the task out of `awaiting_answer`, which resolves the event. Do not pass `--ack` for it. |\n| 4 | `task.stalled` | `parley answer \u003Ctask> \"…\"` resumes it | **automatic** on resume, same as 3 |\n| 5 | `task.failed` | triage (see \"When a task fails\") | **explicit**: next `watch --ack \u003Cseq>` — only after you've triaged |\n| 6 | `task.completed` | review the branch, merge-or-reject, typecheck, `parley clean` (step 4) | **explicit**: next `watch --ack \u003Cseq>` — only after the review is done |\n| 0 | — (all-done) | stop looping; nothing is pending and nothing is running | — |\n| 2 | usage error | fix your invocation | — |\n\nRules that leave no room for interpretation:\n\n- **Ack means \"I handled this\", never \"I saw this\".** Ack a `completed` event only after its branch is reviewed and merged-or-rejected; ack a `failed` event only after triage. Acking early deletes your only reminder — the task drops out of the inbox and nothing will resurface it.\n- **Un-acked events redeliver.** If you crash or forget between delivery and ack, the next `watch` hands you the same event again. That is the safety net — lean on it; never ack defensively \"to clear the queue\".\n- **Exit 6 is not \"done\".** A completed task is *work for you* (review, merge, verify, clean). The loop is finished only at exit 0.\n- **Level-triggered, race-free.** An event already pending when `watch` starts returns immediately. There is no startup race and no sequence bookkeeping on your side; the only seq you ever touch is the one you pass back to `--ack`.\n- Positional task refs (`parley watch t1 t2`) narrow the inbox to those tasks; the default is every task in the session.\n\n4. **Review and integrate.** On exit 6 the envelope carries the worktree path, branch (`parley\u002F\u003Cid>-\u003Cname>`), and the report body. Review the diff on the branch, merge if it holds up, then `parley clean \u003Ctask>` (removes the worktree, keeps the branch). Ack only after that review. Done when the branch is merged-or-rejected and the worktree cleaned.\n\n   **A green report isn't proof correctness.** `outcome: success` only means the child's own verification passed. Verify yourself after every merge, not once at the end of a fan-out unless instructed. A later branch can reintroduce what an earlier one had cleared.\n\n   **Eval when expected.** When step 0 showed evaluation on, record an eval on every reviewed task:\n\n   ```\n   parley eval \u003Ctask> --answers '\u003Cjson>' --feedback \"\u003Cone line per criterion>\"\n   ```\n\n   Map every rubric criterion id (from `parley info`) to a boolean. The daemon computes score and baseline — do not pass a free-form score. Answer honestly, including failures. A later call overwrites the previous result for that task.\n\n5. **Fix loop.** When a completed task needs more work (review found gaps, or you have a concrete fix brief):\n\n   ```\n   parley fix \u003Ctask> \"\u003Cwhat to fix>\"\n   ```\n\n   That creates a linked reattempt (inherits profile\u002Fworkspace; may resume the parent vendor session — policy from `parley info`). Drive it with the same watch loop.\n\n   If `parley fix` exits with `retry_limit_exceeded` (exit 7) or `reattempt_window_expired` (exit 8):\n\n   ```\n   parley fix --fresh \u003Ctask> \"\u003Cwhat to fix>\"\n   ```\n\n   `--fresh` starts a blank session, is uncapped by resume retry limits, and stays in the attempt chain. Exact limits and window live in `parley info` — re-read them when unsure.\n\n## Fan-out: several tasks in parallel\n\nEach task gets its own worktree, so parallel tasks never collide. Batch them however it makes sense, then drive the **entire** set with the same watch loop above:\n\n```\nparley delegate -v \u003Cvendor> -n task-a --session \u003Cid> --type \u003Cid> \"\u003Cbrief A>\"\nparley delegate -v \u003Cvendor> -n task-b --session \u003Cid> --type \u003Cid> \"\u003Cbrief B>\"\n```\n\nDo not poll `status` on an interval and do not sleep-and-check. One mechanism for n=1 and n=N.\n\n## Session ID\n\nThe session ID identifies the current orchestration session. Resolution is env-first: `PARLEY_SESSION_ID` > `--session \u003Cid>` > session-state file > ancestry binding to a registered session. Install the harness plugin so provenance (session id + harness\u002Fmodel\u002Feffort) is set for you — via env vars or the INTERIM state-file channel; see [sessions.md](sessions.md).\n\n## Context files\n\n`--context \u003Cfile>` is repeatable; each file lands in the worktree under `.parley\u002Fcontext\u002F`, materialized by **basename**.\n\n```\nparley delegate -v \u003Cvendor> -n task-a --session \u003Cid> \\\n  --context \u002Fpath\u002Fto\u002Fconfig.json \\\n  --context \u002Fpath\u002Fto\u002Fother.json \\\n  \"\u003Cbrief A>\"\n```\n\n### Integrating fan-out branches\n\nWhen several branches share a fork point, review each branch on its own and resolve conflicts at merge time. When tasks share files or one builds on another, prefer **dependency waves** over blind parallelism: merge task A first, then delegate task B with `--base-ref` on the freshly-merged target so it forks from its actual prerequisite.\n\n## Beyond the golden path\n\nOne-liner pointers — read the linked file only when its condition fires:\n\n- **Non-default task shapes** — structured `--report-schema` results, no git worktree `--cwd`, sandbox postures: read [task-shaping.md](task-shaping.md).\n- **Harness plugins and session provenance** (`PARLEY_SESSION_ID` \u002F `HARNESS` \u002F `MODEL` \u002F `EFFORT`): read [sessions.md](sessions.md).\n\n## When a task fails\n\nCheck the task's `error` field first (`parley status \u003Ctask> --json | jq '.error'`), then `diag.log` in its `logs_dir`, before touching the raw vendor stream — full order and what each layer means: [docs\u002Fagents\u002Ftroubleshooting.md](..\u002F..\u002Fdocs\u002Fagents\u002Ftroubleshooting.md). `parley logs \u003Ctask>` is the last resort; it burns a lot of context on long tasks.\n\nAfter triage, if the failure is fixable with a clearer brief, use the fix loop above (`parley fix` \u002F `parley fix --fresh`) rather than starting an unrelated new delegate when you want a linked attempt chain.\n\n## Reporting parley bugs\n\nParley is in early testing — when parley itself misbehaves (wrong state transition, lost report, worktree damage, a `PARLEY-DIAG` you can't act on), file it upstream after the triage above, whether or not you found a workaround: open an issue on [github.com\u002Ffemoral\u002Fparley](https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley) labelled `needs-triage`, body per [bug-report.md](bug-report.md). Done when the issue URL exists and you've told the user.\n",{"name":5,"description":7,"disable-model-invocation":23},true,"main","61d8c2ca816e611a5421fb7249215dada95d6727","packages\u002Fcli\u002Fskills\u002Fparley-delegate","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley","https:\u002F\u002Fgithub.com\u002Ffemoral","2026-10-15 08:25:59",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":23},[32,35,38],{"tool":33,"command":34},"npx","npx skills add femoral\u002Fparley --skill parley-delegate",{"tool":36,"command":37},"bunx","bunx skills add femoral\u002Fparley --skill parley-delegate",{"tool":39,"command":40},"pnpm","pnpm dlx skills add femoral\u002Fparley --skill parley-delegate","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-delegate\u002FSKILL.md",[43,49,53,57],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"bug-report.md","REFERENCE","参考文档",977,"https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-delegate\u002Fbug-report.md",{"path":50,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":51,"sourceHtmlUrl":52},"sessions.md",4146,"https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-delegate\u002Fsessions.md",{"path":54,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":55,"sourceHtmlUrl":56},"task-shaping.md",1610,"https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-delegate\u002Ftask-shaping.md",{"path":58,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":59,"sourceHtmlUrl":60},"troubleshooting.md",1682,"https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-delegate\u002Ftroubleshooting.md",[62,69,76,83,90,97],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":18},"01M78D3GNHM4K92ZMJ2AK5CN20","fixture-beta","Tiny fixture skill beta for multi-skill install tests.","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Ftests\u002Ffixtures\u002Fskills-bundle\u002Ffixture-beta\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":18},"01M78D3FCPMA1FYRDPB3WZSG33","fixture-alpha","Tiny fixture skill alpha for multi-skill install tests.","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Ftests\u002Ffixtures\u002Fskills-bundle\u002Ffixture-alpha\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":18},"01M78D38XE11MWJTSQJFRM1ZC0","parley-wizard","Configure a parley project by interview — full init (eval, types, rubrics, classification, retention, resume\u002Fretries, remote daemon, vendors) or targeted reconfig. Use when the user wants to set up parley, run \u002Fparley-wizard, reconfigure .parley, turn eval on\u002Foff, edit task types or rubrics, or fix project config.","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002Fpackages\u002Fcli\u002Fskills\u002Fparley-wizard\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":18},"01M78D33Q44YWB06FGZ9VEGGKP","writing-great-skills","Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fwriting-great-skills\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":18},"01M78D2ZG0FVDFDYQA726PKJSP","wayfinder","Plan a huge chunk of work — more than one agent session can hold — as a shared map of investigation tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fwayfinder\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":18},"01M78D2X3405WZAT45XTT4G9KV","triage","Move issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.","https:\u002F\u002Fgithub.com\u002Ffemoral\u002Fparley\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Ftriage\u002FSKILL.md",[],[]]