[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C9S3PCMCKWXS4T2AMCK0C4":3,"skill-same:01M7C9S3PCMCKWXS4T2AMCK0C4":101},{"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},"01M7C9S3PCMCKWXS4T2AMCK0C4","loop-economist",null,"Reviews what an engineering loop's actual runs cost and produced — reading Claude Code session transcripts and git history rather than the loop's documents. Reports tokens per shipped change, cache and subagent spend, rework and thrash, how often a human had to intervene, and whether the right agent, model or persona did each kind of work; scores six dimensions (effectiveness, efficiency, budget, plan quality, agent fit, convergence) into a verdict, and prescribes what closes each finding — a verifier seam, routing the searching out, a bounded subloop or a gauntlet for items one run cannot converge on, or handing the queue to requeue when the input rather than the actor is the problem. Use when someone asks how efficient or effective their loop or agents are, what the loop is costing in tokens or budget, whether it is worth running, why it keeps redoing the same work, whether it is using the right model or agents, how to orchestrate multiple agents or personas across the work, or wants their multi-agent build loop analysed, scored or made cheaper.","Busy-Office\u002Fai-skills","Busy-Office","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F295484308?v=4","01M7C9Q87FQYDQSSBXBRNMQ1CW","01M7C9Q880V3XY9NB38T9ZCWFG",0,"MIT","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: loop-economist\ndescription: Reviews what an engineering loop's actual runs cost and produced — reading Claude Code session transcripts and git history rather than the loop's documents. Reports tokens per shipped change, cache and subagent spend, rework and thrash, how often a human had to intervene, and whether the right agent, model or persona did each kind of work; scores six dimensions (effectiveness, efficiency, budget, plan quality, agent fit, convergence) into a verdict, and prescribes what closes each finding — a verifier seam, routing the searching out, a bounded subloop or a gauntlet for items one run cannot converge on, or handing the queue to requeue when the input rather than the actor is the problem. Use when someone asks how efficient or effective their loop or agents are, what the loop is costing in tokens or budget, whether it is worth running, why it keeps redoing the same work, whether it is using the right model or agents, how to orchestrate multiple agents or personas across the work, or wants their multi-agent build loop analysed, scored or made cheaper.\n---\n\n# Loop Economist\n\n`loop-doctor` reads a loop's documents and asks whether the design is\nsound. Loop-economist reads what the loop **actually did** — every session\ntranscript and every commit in a window — and asks a different question:\n*what did a unit of shipped change cost, and was it produced by the right\ncrew?*\n\nA loop can score well on its documents and still spend forty thousand\ntokens a tick re-reading a codebase to change one line, or ship and then\nun-ship the same file four times. Only the runs show that.\n\n```mermaid\nflowchart LR\n  A[\"runs.mjs\u003Cbr\u002F>transcripts + git · read-only\"] --> B[\"the economics\u003Cbr\u002F>tokens per commit\"]\n  B --> C[\"findings\u003Cbr\u002F>Leak · Misroute · No-converge · Ill-formed\"]\n  C --> D[\"score\u003Cbr\u002F>6 × 0–5 → verdict\"]\n  D --> E[\"prescribe\u003Cbr\u002F>verifier · routing · subloop · gauntlet\"]\n  E --> F{\"binding constraint?\"}\n  F -- the actor --> G[\"review\u003Cbr\u002F>economics · score · findings · do-next\"]\n  F -- the input --> H[\"hand off to\u003Cbr\u002F>requeue\"]\n```\n\n## The guard-rail: no overhead on the project\n\n- **Write nothing into the target repo.** The review goes to the\n  scratchpad, or to a published page if that is how they read things.\n- **Run nothing of the project's.** No install, build or test run. Only\n  `git log` and the collector.\n- **Never quote a transcript.** Sessions contain the user's own words and\n  their code. Cite a session by id and a metric, never by content.\n- **Read bounded.** `derived` and `warnings` fully; `sessions` and\n  `git.commits` by id only, when a finding needs that one entry's\n  evidence. Never read the full `runs.json` — it scales with the window,\n  `derived` doesn't.\n- **One context, one pass.** No subagents by default — reviewing a\n  fan-out problem by fanning out is its own joke.\n- **Leave a receipt** in the footer: transcripts read, window, commits,\n  repo untouched.\n\n## Workflow\n\n### 1. Collect\n\n```bash\nnode \u003Cskill-dir>\u002Fscripts\u002Fruns.mjs \u003Crepo-path> --since 14d > \u003Cscratchpad>\u002Fruns.json\njq '{derived, warnings, records, git: {commits: (.git.commits | length), rework: .git.rework}}' \u003Cscratchpad>\u002Fruns.json\n```\n\n`derived` already carries every rate this review reports — tokens per\ncommit, cache hit rate, subagent share, rework rate, the session-id lists\nfor thrash and zero-commit — and costs a few hundred tokens regardless of\nwindow. **Read that slice fully; do not read the full `runs.json` top to\nbottom.** The `sessions` array and `git.commits` are per-run raw detail\nthat scales with the window (every session's tool\u002Fsubagent breakdown,\nevery commit's numstat) and the review never quotes them — findings cite\na session id or a sha, not their contents. Treat the full file as an\nindex: pull one entry only when a finding needs to cite that specific\nsession or commit —\n\n```bash\njq --arg id \"\u003Csession-id>\" '.sessions[] | select(.id == $id)' \u003Cscratchpad>\u002Fruns.json\n```\n\n— never `Read` the whole `runs.json` into context. Sessions (tokens by\nkind, models, tools, subagents by type, human turns, interrupts, tool\nerrors, repeated identical calls, files edited), git (commits, churn,\nrework commits), and the loop's records (open\u002Fdone\u002Fvague items) are the\nraw material `derived` is computed from; the numbers are the evidence,\nnot the raw rows.\n\nPick the window from the loop's cadence: at least **20 ticks or 14 days**,\nwhichever is longer. Say the window in the first line of the review; every\nrate is meaningless without it.\n\n**Write the review from one collection, and quote that collection's numbers.**\nA rolling window moves: a review drafted from Monday's run and finished on\nTuesday will state a commit count that no longer exists, and every derived rate\ninherits it. Re-run the collector immediately before writing, and verify with\n`node evals\u002Fbar-check.mjs loop-economist \u003Cartifact> --collector \u003Crun.json>`,\nwhich compares the headline figures against the collection.\n\nNo transcripts (a loop that runs elsewhere, or a fresh machine) → the\nbudget half is `NOT MEASURED`. Score effectiveness, plan quality and\nconvergence from git and the records, and say plainly which four numbers\nyou could not see. Do not infer tokens from diff sizes.\n\n### 2. Read the economics\n\n`references\u002Fmetrics.md` — each metric with its band and the cause a\nreading out of band implies. Establish **tokens per commit** first; it is\nthe number the reader will remember. Then the three that explain it:\ncache hit rate, subagent share, tool calls per edit.\n\n### 3. Find the findings\n\nOne row per finding, classified:\n\n- **Leak** — spend that produced no kept change (a zero-commit session, a\n  fan-out whose output was re-derived, state re-read every tick).\n- **Misroute** — the wrong actor did the work: `agent-fit.md`.\n- **No-converge** — thrash, rework, a file the loop cannot leave alone.\n- **Ill-formed** — the item had no acceptance test; no actor could have\n  converged.\n\nEach row cites a session id, a file or a metric. \"Could be more efficient\"\nis not a row.\n\n### 4. Score\n\n`references\u002Fscorecard.md`. Six dimensions 0–5, one-line reason each citing\na number, mean → verdict (`compounding · productive · expensive ·\nspinning`), then the caps: nothing shipped caps at *spinning*; more than\nthree human turns a run caps at *productive*; rework over 0.4 caps at\n*expensive*.\n\n**Show the mean and name every cap, including the ones that do not bind.** Add\nthe numbers up and check the division — a mean that disagrees with its own six\nscores is the easiest thing in the review to falsify. Then one line: which caps\nwere tested, which applied, and which did not. A verdict word alone does not\nshow the caps were considered.\n\nWrite the reason before the digit. A reason citing two out-of-band metrics\ncannot carry a 3.\n\n### 5. Prescribe\n\n`references\u002Fpatterns.md`, and only what closes a finding you wrote:\nverifier seam, route the reading out, cache discipline, per-tick budget,\npersona split, and for the items one pass cannot reach — a **bounded\nsubloop** (try · verify · adjust, N rounds, each round must change the\napproach) or a **gauntlet** (k independent attempts, a bar written first,\na judge that wrote none of them). Give every prescription its cost.\n\n### 6. Name the binding constraint\n\nOne sentence, and it decides the handoff:\n\n- **the actor** — wrong model, no verifier, unrouted searching → the\n  prescriptions above.\n- **the input** — thrash and rework concentrate on items with no\n  acceptance test → say *the input, not the actor, is the problem* and\n  hand off to **`requeue`**. A stronger model on an unstated target\n  converges on nothing, more expensively.\n- **the design** — the tick itself is unsound (two drivers, no stop, no\n  gate) → hand off to **`loop-doctor`**; that is its question, not this\n  one.\n\n### 6b. Compare against the last one, if there is one\n\nEvery number here is a point on a curve, and the curve is the argument. Before\nwriting, look for the previous review — a project keeping these in a notebook or\na `docs\u002Freviews\u002F` directory can be asked directly. Where one exists, lead with\nthe movement: *\"296k per shipping commit, up from 245k; rework 0.23 unchanged.\"*\nWhere none does, say so, and say the review is a baseline rather than a verdict —\na first measurement cannot tell anyone whether things are getting better.\n\n`docs\u002Fsecond-brain.md` in this repo is one optional way to keep them.\n\n### 7. Write the review\n\n`references\u002Freport-template.md`, exactly: ≤ 60 words, the economics table,\nthe crew diagram drawn from the runs (including the edges that should not\nexist), the score block, findings, non-converging runs, prescriptions,\ndo-next (≤ 5), footer receipt. Then a terminal summary of ≤ 8 lines.\n\n### 8. Apply — only when asked\n\nPrescriptions change how the loop behaves. Propose; apply on request, one\ncommit per prescription, diff shown first.\n\nFor the verifier seam specifically, applying means writing\n`references\u002Fpersonas\u002Fverifier.md` into the target repo's\n`.claude\u002Fagents\u002Fverifier.md` with its four slots filled from that project, and\nadding the one rule that calls it before the commit step. Fill the slots from\nthe project's own task file; **if the check command cannot be established, do\nnot install the agent** and say why — an unwired verifier returns green without\nrunning anything, and the loop will believe it.\n\n## Judgement calls\n\n**Tokens per commit is a ratio, and both halves lie alone.** Cheap runs\nthat ship nothing and expensive runs that ship a migration look the same\nin one number. Always report the pair.\n\n**Cache reads are not the bill.** Report them beside the billable total,\nnever inside it. Overstating spend to make a point is the same failure as\nunderstating it.\n\n**A retry that changes nothing is not persistence.** Three identical calls\nin one session is the evidence for a subloop; the subloop's whole content\nis the rule that each round must change the approach.\n\n**A context cannot verify itself.** If no separate actor ever checked the\nwork, agent fit is 0 — not 2 — however good the code turned out.\n\n**Fan-out is a cost, not a virtue.** Subagent share above half with flat\ncommits is a finding, and the prescription is fewer agents with stated\noutput shapes, not better prompts.\n\n**One window is a sample.** Quote the session count next to every rate,\nand prefer \"in this window\" to \"always\".\n\n**Never paste what a session said.** Ids and numbers only.\n\n## Files\n\n- `scripts\u002Fruns.mjs` — transcripts + git collector; `--self-test`.\n- `references\u002Fmetrics.md` — every metric, its band, what a bad reading\n  means.\n- `references\u002Fscorecard.md` — six dimensions, anchors, verdict and caps.\n- `references\u002Fagent-fit.md` — routing table, the four personas, fan-out\n  rules.\n- `references\u002Fpatterns.md` — subloop, gauntlet, verifier seam, routing,\n  cache, budget, escalate-to-input.\n- `references\u002Fpersonas\u002Fverifier.md` — the verifier as a template for the\n  target repo, with the refusal to install it unwired.\n- `references\u002Freport-template.md` — the review, exactly.\n- `evals\u002F` — the prompts, and `gauntlet\u002FBAR.md`, the bar this skill's\n  reviews are graded against by a blind critic.\n\n## Related\n\n`loop-doctor` reads the same loop's documents and judges the design;\n`loop-atlas` draws the flow and the crew as they are; `requeue` fixes the input\nwhen the items were never tasks; `sharpen-intent` fixes what the items are\nranked against. Step 6 names which of them the findings actually belong to —\nthat hand-off is part of the review, not an afterthought.\n\nWhen the binding constraint is *the actor* and the fix is the shape of the\nwork inside a run — a fan-out that should be a pipeline, a verifier that\nneeds its own node, a discovery loop that never converges — that is\n`graph-engineer`'s question: this skill measures the bill, that one redraws\nthe graph.\n","# Loop Economist\n\n`loop-doctor` reads a loop's documents and asks whether the design is\nsound. Loop-economist reads what the loop **actually did** — every session\ntranscript and every commit in a window — and asks a different question:\n*what did a unit of shipped change cost, and was it produced by the right\ncrew?*\n\nA loop can score well on its documents and still spend forty thousand\ntokens a tick re-reading a codebase to change one line, or ship and then\nun-ship the same file four times. Only the runs show that.\n\n```mermaid\nflowchart LR\n  A[\"runs.mjs\u003Cbr\u002F>transcripts + git · read-only\"] --> B[\"the economics\u003Cbr\u002F>tokens per commit\"]\n  B --> C[\"findings\u003Cbr\u002F>Leak · Misroute · No-converge · Ill-formed\"]\n  C --> D[\"score\u003Cbr\u002F>6 × 0–5 → verdict\"]\n  D --> E[\"prescribe\u003Cbr\u002F>verifier · routing · subloop · gauntlet\"]\n  E --> F{\"binding constraint?\"}\n  F -- the actor --> G[\"review\u003Cbr\u002F>economics · score · findings · do-next\"]\n  F -- the input --> H[\"hand off to\u003Cbr\u002F>requeue\"]\n```\n\n## The guard-rail: no overhead on the project\n\n- **Write nothing into the target repo.** The review goes to the\n  scratchpad, or to a published page if that is how they read things.\n- **Run nothing of the project's.** No install, build or test run. Only\n  `git log` and the collector.\n- **Never quote a transcript.** Sessions contain the user's own words and\n  their code. Cite a session by id and a metric, never by content.\n- **Read bounded.** `derived` and `warnings` fully; `sessions` and\n  `git.commits` by id only, when a finding needs that one entry's\n  evidence. Never read the full `runs.json` — it scales with the window,\n  `derived` doesn't.\n- **One context, one pass.** No subagents by default — reviewing a\n  fan-out problem by fanning out is its own joke.\n- **Leave a receipt** in the footer: transcripts read, window, commits,\n  repo untouched.\n\n## Workflow\n\n### 1. Collect\n\n```bash\nnode \u003Cskill-dir>\u002Fscripts\u002Fruns.mjs \u003Crepo-path> --since 14d > \u003Cscratchpad>\u002Fruns.json\njq '{derived, warnings, records, git: {commits: (.git.commits | length), rework: .git.rework}}' \u003Cscratchpad>\u002Fruns.json\n```\n\n`derived` already carries every rate this review reports — tokens per\ncommit, cache hit rate, subagent share, rework rate, the session-id lists\nfor thrash and zero-commit — and costs a few hundred tokens regardless of\nwindow. **Read that slice fully; do not read the full `runs.json` top to\nbottom.** The `sessions` array and `git.commits` are per-run raw detail\nthat scales with the window (every session's tool\u002Fsubagent breakdown,\nevery commit's numstat) and the review never quotes them — findings cite\na session id or a sha, not their contents. Treat the full file as an\nindex: pull one entry only when a finding needs to cite that specific\nsession or commit —\n\n```bash\njq --arg id \"\u003Csession-id>\" '.sessions[] | select(.id == $id)' \u003Cscratchpad>\u002Fruns.json\n```\n\n— never `Read` the whole `runs.json` into context. Sessions (tokens by\nkind, models, tools, subagents by type, human turns, interrupts, tool\nerrors, repeated identical calls, files edited), git (commits, churn,\nrework commits), and the loop's records (open\u002Fdone\u002Fvague items) are the\nraw material `derived` is computed from; the numbers are the evidence,\nnot the raw rows.\n\nPick the window from the loop's cadence: at least **20 ticks or 14 days**,\nwhichever is longer. Say the window in the first line of the review; every\nrate is meaningless without it.\n\n**Write the review from one collection, and quote that collection's numbers.**\nA rolling window moves: a review drafted from Monday's run and finished on\nTuesday will state a commit count that no longer exists, and every derived rate\ninherits it. Re-run the collector immediately before writing, and verify with\n`node evals\u002Fbar-check.mjs loop-economist \u003Cartifact> --collector \u003Crun.json>`,\nwhich compares the headline figures against the collection.\n\nNo transcripts (a loop that runs elsewhere, or a fresh machine) → the\nbudget half is `NOT MEASURED`. Score effectiveness, plan quality and\nconvergence from git and the records, and say plainly which four numbers\nyou could not see. Do not infer tokens from diff sizes.\n\n### 2. Read the economics\n\n`references\u002Fmetrics.md` — each metric with its band and the cause a\nreading out of band implies. Establish **tokens per commit** first; it is\nthe number the reader will remember. Then the three that explain it:\ncache hit rate, subagent share, tool calls per edit.\n\n### 3. Find the findings\n\nOne row per finding, classified:\n\n- **Leak** — spend that produced no kept change (a zero-commit session, a\n  fan-out whose output was re-derived, state re-read every tick).\n- **Misroute** — the wrong actor did the work: `agent-fit.md`.\n- **No-converge** — thrash, rework, a file the loop cannot leave alone.\n- **Ill-formed** — the item had no acceptance test; no actor could have\n  converged.\n\nEach row cites a session id, a file or a metric. \"Could be more efficient\"\nis not a row.\n\n### 4. Score\n\n`references\u002Fscorecard.md`. Six dimensions 0–5, one-line reason each citing\na number, mean → verdict (`compounding · productive · expensive ·\nspinning`), then the caps: nothing shipped caps at *spinning*; more than\nthree human turns a run caps at *productive*; rework over 0.4 caps at\n*expensive*.\n\n**Show the mean and name every cap, including the ones that do not bind.** Add\nthe numbers up and check the division — a mean that disagrees with its own six\nscores is the easiest thing in the review to falsify. Then one line: which caps\nwere tested, which applied, and which did not. A verdict word alone does not\nshow the caps were considered.\n\nWrite the reason before the digit. A reason citing two out-of-band metrics\ncannot carry a 3.\n\n### 5. Prescribe\n\n`references\u002Fpatterns.md`, and only what closes a finding you wrote:\nverifier seam, route the reading out, cache discipline, per-tick budget,\npersona split, and for the items one pass cannot reach — a **bounded\nsubloop** (try · verify · adjust, N rounds, each round must change the\napproach) or a **gauntlet** (k independent attempts, a bar written first,\na judge that wrote none of them). Give every prescription its cost.\n\n### 6. Name the binding constraint\n\nOne sentence, and it decides the handoff:\n\n- **the actor** — wrong model, no verifier, unrouted searching → the\n  prescriptions above.\n- **the input** — thrash and rework concentrate on items with no\n  acceptance test → say *the input, not the actor, is the problem* and\n  hand off to **`requeue`**. A stronger model on an unstated target\n  converges on nothing, more expensively.\n- **the design** — the tick itself is unsound (two drivers, no stop, no\n  gate) → hand off to **`loop-doctor`**; that is its question, not this\n  one.\n\n### 6b. Compare against the last one, if there is one\n\nEvery number here is a point on a curve, and the curve is the argument. Before\nwriting, look for the previous review — a project keeping these in a notebook or\na `docs\u002Freviews\u002F` directory can be asked directly. Where one exists, lead with\nthe movement: *\"296k per shipping commit, up from 245k; rework 0.23 unchanged.\"*\nWhere none does, say so, and say the review is a baseline rather than a verdict —\na first measurement cannot tell anyone whether things are getting better.\n\n`docs\u002Fsecond-brain.md` in this repo is one optional way to keep them.\n\n### 7. Write the review\n\n`references\u002Freport-template.md`, exactly: ≤ 60 words, the economics table,\nthe crew diagram drawn from the runs (including the edges that should not\nexist), the score block, findings, non-converging runs, prescriptions,\ndo-next (≤ 5), footer receipt. Then a terminal summary of ≤ 8 lines.\n\n### 8. Apply — only when asked\n\nPrescriptions change how the loop behaves. Propose; apply on request, one\ncommit per prescription, diff shown first.\n\nFor the verifier seam specifically, applying means writing\n`references\u002Fpersonas\u002Fverifier.md` into the target repo's\n`.claude\u002Fagents\u002Fverifier.md` with its four slots filled from that project, and\nadding the one rule that calls it before the commit step. Fill the slots from\nthe project's own task file; **if the check command cannot be established, do\nnot install the agent** and say why — an unwired verifier returns green without\nrunning anything, and the loop will believe it.\n\n## Judgement calls\n\n**Tokens per commit is a ratio, and both halves lie alone.** Cheap runs\nthat ship nothing and expensive runs that ship a migration look the same\nin one number. Always report the pair.\n\n**Cache reads are not the bill.** Report them beside the billable total,\nnever inside it. Overstating spend to make a point is the same failure as\nunderstating it.\n\n**A retry that changes nothing is not persistence.** Three identical calls\nin one session is the evidence for a subloop; the subloop's whole content\nis the rule that each round must change the approach.\n\n**A context cannot verify itself.** If no separate actor ever checked the\nwork, agent fit is 0 — not 2 — however good the code turned out.\n\n**Fan-out is a cost, not a virtue.** Subagent share above half with flat\ncommits is a finding, and the prescription is fewer agents with stated\noutput shapes, not better prompts.\n\n**One window is a sample.** Quote the session count next to every rate,\nand prefer \"in this window\" to \"always\".\n\n**Never paste what a session said.** Ids and numbers only.\n\n## Files\n\n- `scripts\u002Fruns.mjs` — transcripts + git collector; `--self-test`.\n- `references\u002Fmetrics.md` — every metric, its band, what a bad reading\n  means.\n- `references\u002Fscorecard.md` — six dimensions, anchors, verdict and caps.\n- `references\u002Fagent-fit.md` — routing table, the four personas, fan-out\n  rules.\n- `references\u002Fpatterns.md` — subloop, gauntlet, verifier seam, routing,\n  cache, budget, escalate-to-input.\n- `references\u002Fpersonas\u002Fverifier.md` — the verifier as a template for the\n  target repo, with the refusal to install it unwired.\n- `references\u002Freport-template.md` — the review, exactly.\n- `evals\u002F` — the prompts, and `gauntlet\u002FBAR.md`, the bar this skill's\n  reviews are graded against by a blind critic.\n\n## Related\n\n`loop-doctor` reads the same loop's documents and judges the design;\n`loop-atlas` draws the flow and the crew as they are; `requeue` fixes the input\nwhen the items were never tasks; `sharpen-intent` fixes what the items are\nranked against. Step 6 names which of them the findings actually belong to —\nthat hand-off is part of the review, not an afterthought.\n\nWhen the binding constraint is *the actor* and the fix is the shape of the\nwork inside a run — a fan-out that should be a pipeline, a verifier that\nneeds its own node, a discovery loop that never converges — that is\n`graph-engineer`'s question: this skill measures the bill, that one redraws\nthe graph.\n",{"name":5,"description":7},"main","ee822ad7e3f1e7dd3305c8e03565dc584906fb76","skills\u002Floop-economist","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills","https:\u002F\u002Fgithub.com\u002FBusy-Office","2026-09-27 17:58:58",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add Busy-Office\u002Fai-skills --skill loop-economist",{"tool":35,"command":36},"bunx","bunx skills add Busy-Office\u002Fai-skills --skill loop-economist",{"tool":38,"command":39},"pnpm","pnpm dlx skills add Busy-Office\u002Fai-skills --skill loop-economist","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002FSKILL.md",true,[43,49,53,57,61,65,71,75,79,83,87,91,95],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"evals\u002Fevals.json","REFERENCE","参考文档",2964,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Fevals\u002Fevals.json",{"path":50,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":51,"sourceHtmlUrl":52},"evals\u002Fgauntlet\u002FBAR.md",5886,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Fevals\u002Fgauntlet\u002FBAR.md",{"path":54,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":55,"sourceHtmlUrl":56},"evals\u002Fgauntlet\u002FROUNDS.md",2023,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Fevals\u002Fgauntlet\u002FROUNDS.md",{"path":58,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":59,"sourceHtmlUrl":60},"fixtures\u002Fthrash-run\u002Fdocs\u002FBACKLOG.md",119,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Ffixtures\u002Fthrash-run\u002Fdocs\u002FBACKLOG.md",{"path":62,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":63,"sourceHtmlUrl":64},"fixtures\u002Fthrash-run\u002Fexpect.json",429,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Ffixtures\u002Fthrash-run\u002Fexpect.json",{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"fixtures\u002Fthrash-run\u002Ftranscripts\u002Fs1.jsonl","OTHER","其他",3090,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Ffixtures\u002Fthrash-run\u002Ftranscripts\u002Fs1.jsonl",{"path":72,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":73,"sourceHtmlUrl":74},"references\u002Fagent-fit.md",6236,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Freferences\u002Fagent-fit.md",{"path":76,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":77,"sourceHtmlUrl":78},"references\u002Fmetrics.md",7196,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Freferences\u002Fmetrics.md",{"path":80,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":81,"sourceHtmlUrl":82},"references\u002Fpatterns.md",9012,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Freferences\u002Fpatterns.md",{"path":84,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":85,"sourceHtmlUrl":86},"references\u002Fpersonas\u002Fverifier.md",4576,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Freferences\u002Fpersonas\u002Fverifier.md",{"path":88,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":89,"sourceHtmlUrl":90},"references\u002Freport-template.md",2912,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Freferences\u002Freport-template.md",{"path":92,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":93,"sourceHtmlUrl":94},"references\u002Fscorecard.md",2980,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Freferences\u002Fscorecard.md",{"path":96,"resourceType":97,"resourceTypeDesc":98,"sizeBytes":99,"sourceHtmlUrl":100},"scripts\u002Fruns.mjs","SCRIPT","脚本",21152,"https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Floop-economist\u002Fscripts\u002Fruns.mjs",[102,109,115,122,128,135],{"skillNo":103,"name":104,"nameZh":6,"description":105,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":106,"categories":107,"keywords":108,"publishedAt":18},"01M7C9TK43J3W13Q83AHM3ZCWV","loop","one tick","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwake-weight\u002Ffixtures\u002Fheavy-wake\u002F.claude\u002Fskills\u002Floop\u002FSKILL.md",[],[],{"skillNo":110,"name":111,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":112,"categories":113,"keywords":114,"publishedAt":18},"01M7C9TG9ND44SG4X5735RAKVE","wake-weight","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwake-weight\u002FSKILL.md",[],[],{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":18},"01M7C9T6K5HNKMW49F9XXAPZVH","solo-flow","Prescribes and audits a trunk-based git workflow for a repository with exactly one writer — a single autonomous agent or loop, with no other humans committing. One branch (main), local-only short-lived work branches, CI as the sole independent check (no PRs, nothing to review socially), and releases as tags cut directly off main rather than a develop\u002Frelease branch dance. Use whenever someone is setting up or reviewing git\u002Fbranching\u002Frelease conventions for an agent-driven or loop-driven repo, asks whether to use GitFlow\u002FGitHub Flow\u002FGitLab Flow\u002Ftrunk-based development, wants AGENTS.md or CLAUDE.md git rules written, is deciding whether a develop branch or long-lived feature branches earn their keep, or reports branch drift, a forgotten back-merge, or an autonomous loop that got confused about which branch it was on. Also use to sanity-check an existing branching setup that looks like GitFlow but has no other human collaborators to justify it, and when deciding how the agent should isolate each unit of work (git worktrees, a primary checkout that never leaves main, running items in parallel).","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsolo-flow\u002FSKILL.md",[],[],{"skillNo":123,"name":124,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":125,"categories":126,"keywords":127,"publishedAt":18},"01M7C9T3HA83YMNZNR8VSM5694","sharpen-intent","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsharpen-intent\u002FSKILL.md",[],[],{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":18},"01M7C9SW8YC2TWRYH6BN7NRDH9","requeue","Reviews and reorders the queue an engineering loop or team works from — backlog, roadmap, todo, milestones — so the next work picked up is the right work. Pulls every open item with the attributes that decide order (acceptance stated or not, blocked, human-only, duplicated, contradicted between files, stale), ranks them against the project's intent by unblocking power, purpose, readiness, decay and cost, splits them into lanes (the loop can finish it · needs a bounded subloop · needs a gauntlet of approaches · needs a human), rewrites the top vague items into tasks with an object, a test and a bound, and proposes the new order without touching the repo. Use when someone asks what to work on next, wants the backlog or roadmap reprioritised, re-ranked, triaged or cleaned up, says the priorities are wrong or the roadmap is stale, asks which items the autonomous loop should be fed, or arrives from loop-economist because the loop's problem is its input rather than its actor.","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Frequeue\u002FSKILL.md",[],[],{"skillNo":136,"name":137,"nameZh":6,"description":138,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":139,"categories":140,"keywords":141,"publishedAt":18},"01M7C9SQXNBEVGRFZR40W24AMW","progress-dashboard","Publishes a one-page stakeholder progress dashboard for any git project — where it stands, what's waiting on people, whether it's stalled — from whatever records the project keeps (ADRs, decision notes, design graphs, roadmaps, gates, backlogs, changelogs, PRs). Use whenever someone asks how a project is going, wants a status update or progress report, asks \"where are we\", \"what's blocked\", \"what's waiting on me\u002Fus\", \"what's left\", or wants an existing dashboard refreshed.","https:\u002F\u002Fgithub.com\u002FBusy-Office\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fprogress-dashboard\u002FSKILL.md",[],[]]