[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71RFN249M21NY2RFT4Q1P1G":3,"skill-same:01M71RFN249M21NY2RFT4Q1P1G":107},{"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":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":49,"commitSha":50,"skillDirectory":51,"forksCount":52,"repositoryHtmlUrl":53,"repositoryHomepageUrl":6,"ownerHtmlUrl":54,"ownerBlogUrl":6,"ownerDescription":55,"updatedAt":56,"install":57,"resources":70},"01M71RFN249M21NY2RFT4Q1P1G","autospec-loop",null,"Use when the user wants to run a single task repeatedly in a loop until a goal is reached — \"loop until the build passes\", \"keep going until done\", \"run X in a loop until Y\". Interactively refines the request into a frozen contract, then runs a goal-conditioned loop with conservative guardrails. Claims goal\u002Fcompletion loop phrasings; defers bare interval polling (\"loop every 5m\") to the native \u002Floop.","berlinguyinca\u002Fautospec","berlinguyinca","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F16364?v=4","01M71R7C0SWBG05FNHQE4XQ7Y3","Gert Wohlgemuth","01M71R7C1VY5K3Q715RNDDBVGQ",3,"MIT","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000037","律师与法务","PROFESSION","适用职业",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"autospec","loop","autospec loop","user","request","run","goal","build","native","claims","with","task","done","keep","when","use","contract","single","Guardrails","bare","2026-11-10 13:02:02",false,"---\nname: autospec-loop\ndescription: Use when the user wants to run a single task repeatedly in a loop until a goal is reached — \"loop until the build passes\", \"keep going until done\", \"run X in a loop until Y\". Interactively refines the request into a frozen contract, then runs a goal-conditioned loop with conservative guardrails. Claims goal\u002Fcompletion loop phrasings; defers bare interval polling (\"loop every 5m\") to the native \u002Floop.\n---\n\n# autospec-loop workflow (harness-neutral)\n\nTake an operator request to \"run something in a loop until a goal is reached\"\nand turn it into (1) an interactive **refine-until-\"go\"** gate that freezes a\ntestable contract, then (2) a **goal-conditioned loop** that re-dispatches the\nrefined prompt with carried state each iteration until a deterministic check —\nor, failing that, an independent verifier — confirms the goal is met, with\nconservative guardrails so it never spins unbounded.\n\nGoal: a harness-neutral, autospec-family way to \"keep doing X until Y\" that\nstops on the *goal* (not a clock), prefers a deterministic CHECK over LLM\nself-assessment, and is safe to run on small-context local models.\n\nManage your own context — never exceed 60%. Delegate per-iteration work to\nsubagents whenever your harness supports it (Codex runs iterations inline); do\nnot carry raw iteration transcripts in the orchestrator context.\n\n\u003C!-- autospec-block:startup-self-update SKILL_NAME=autospec-loop -->\n\n## Self-update mode\n\nIf the argument matches the regex `^\\s*update\\s*$` (case-insensitive,\nwhitespace-padded), this skill enters self-update mode and does not run the\nnormal pipeline. This section is pure prose: never interpolate or shell out the\noperator's argument text.\n\n1. **Detect harness** by checking which install path exists for this skill:\n   - Claude Code: `~\u002F.claude\u002Fskills\u002Fautospec-loop\u002FSKILL.md`\n   - OpenCode:    `~\u002F.config\u002Fopencode\u002Fagent\u002Fautospec-loop.md`\n   - Codex CLI:   `~\u002F.codex\u002Fprompts\u002Fautospec-loop.md`\n2. **Re-install the full autospec suite from `main`** by piping the canonical installer:\n   ```bash\n   curl -fsSL https:\u002F\u002Fraw.githubusercontent.com\u002Fberlinguyinca\u002Fautospec\u002Fmain\u002Fbootstrap.sh | bash -s -- --skill all --harness all --update\n   ```\n   Run this one-liner once; it refreshes all autospec skills across all harnesses.\n3. **Show the diff** between the prior installed file(s) and the freshly fetched copy.\n4. **Stop.** Do not enter the loop pipeline. Print the upgrade summary and return to the user.\n\nIf no install path is detected, print `Self-update: no installed copy of autospec-loop found; run install.sh first.` and exit.\n\n## Model tier\n\nTier A\u002FB model resolution and the harness-detection protocol are inherited\nverbatim from `AGENTS.md` (`## Subagent model selection (two-tier, cost-aware)`\nand `## Subagent vs inline decision matrix`). Refinement-gate reasoning runs at\nTier A (spec work); per-iteration loop work and the independent verifier run at\nTier B (implementation work), escalating to Tier A for high-stakes goals.\n\n> **Model tier:** Phase 1 refine-gate dispatches at Tier A (spec work); Phase 2\n> per-iteration worker and verifier dispatch at TIER_B (implementation work),\n> resolved at startup per the harness detection below.\n\n### Required capabilities & harness adapter\n\nThis workflow assumes the following capabilities. Map each to your harness's\nactual tool; if a capability is missing, use the listed fallback.\n\n| Capability                  | Claude Code                             | OpenCode                              | Codex CLI                                | Fallback if missing                              |\n|-----------------------------|-----------------------------------------|---------------------------------------|------------------------------------------|--------------------------------------------------|\n| Per-iteration worker        | `Agent` (subagent_type=general-purpose) | `task` agent, await output            | run the iteration inline (no subagents)  | Run the iteration in-thread (more context cost)  |\n| Independent verifier        | a **separate** `Agent` dispatch         | a **separate** `task` agent           | a separate inline judging pass           | Judge in-thread, but never reuse the worker pass |\n| Ask the user a question     | `AskUserQuestion`                       | inline prompt                         | inline prompt                            | Ask in the response and wait for the next turn   |\n| Subagent model tier         | Tier A: `opus` + `ultrathink`; Tier B: `sonnet` + medium thinking | Tier A: top `task` model + high reasoning; Tier B: smaller-tier `task` + medium reasoning | Tier A: top GPT + `reasoning_effort=high`; Tier B: current spark\u002Fcost-optimized Codex + `reasoning_effort=medium` | Honor the per-phase tier mapping in AGENTS.md; retry the same subagent UP on unavailability |\n| Subagent dispatch policy    | per AGENTS.md decision matrix           | per AGENTS.md decision matrix         | per AGENTS.md decision matrix            | inline with main-session token cost              |\n\n**Persistent project notes**: write durable preferences to **`AGENTS.md`** in\nthe repo root — recognized by Claude Code, OpenCode, and Codex.\n\n## Harness detection (run once at skill start, before the gate begins)\n\nDetect your harness by checking available tools before any dispatch:\n\n1. **Claude Code** — the `Agent` tool with a `subagent_type` parameter is available.\n   - `TIER_A` = `opus` + `ultrathink`  (the alias, which resolves to the current Claude Opus generation)\n   - `TIER_B` = `sonnet`               (the alias, which resolves to the current Claude Sonnet generation)\n2. **OpenCode** — a `task` tool with model\u002Ftier configuration is available (no `subagent_type`).\n   - `TIER_A` = top-tier task model + high reasoning\n   - `TIER_B` = smaller-tier task model + medium reasoning\n3. **Codex CLI** — neither `Agent` nor a configurable `task` tool is available; iterations run inline.\n   - `TIER_A` = current top GPT model + `reasoning_effort=high`\n   - `TIER_B` = the spark \u002F cost-optimized variant of the configured model when one exists, else the configured model + `reasoning_effort=medium`\n\n**Fallback rule:** If `TIER_B` is not available in your harness, silently retry\nthe same subagent dispatch with `TIER_A`. Preserve parent context on retry.\nNever ask the user.\n\nHold `TIER_A` and `TIER_B` for the entire skill run.\n\n## Trigger disambiguation\n\n`autospec-loop` and the native `\u002Floop` cover different intents; route by\ngoal-vs-interval:\n\n- **autospec-loop claims** goal\u002Fcompletion phrasings: *\"loop until …\"*, *\"in a\n  loop until …\"*, *\"keep going until done\"*, *\"do a loop\"*, *\"execute a loop\"*,\n  *\"run X in a loop until Y\"*, *\"keep doing X until Y\"*.\n- **Native `\u002Floop` keeps** bare interval phrasings: *\"loop every 5m\"*,\n  *\"\u002Floop 5m \u002Ffoo\"*, *\"poll every N minutes\"*, *\"keep running … on an interval\"*.\n- **Ambiguous** bare *\"do a loop\"* \u002F *\"execute a loop\"* with no interval and no\n  explicit goal → autospec-loop claims it, and the refine gate's **first\n  question** is goal-or-interval. If the operator wants interval polling, the\n  skill **redirects to native `\u002Floop`** and exits without starting its own loop.\n\nDecision rule, applied to the raw request before anything else runs:\n\n1. **Explicit interval, no goal** — the request names a cadence (*\"every 5m\"*,\n   *\"every N minutes\"*, *\"on an interval\"*, or the literal `\u002Floop \u003Cinterval>`\n   form) and no completion criterion → **defer to native `\u002Floop`**; do not\n   enter this skill's gate.\n2. **Explicit goal \u002F completion** — the request names a stopping condition\n   (*\"until …\"*, *\"until done\"*, *\"until it passes\"*, *\"keep doing X until Y\"*)\n   → **autospec-loop claims it**; enter the Phase 1 gate.\n3. **Ambiguous** — a bare *\"do a loop\"* \u002F *\"execute a loop\"* with neither an\n   interval nor a goal → autospec-loop claims it, and the gate's **first\n   clarifying question is goal-or-interval**: on a goal answer, continue the\n   gate; on an interval answer, **redirect to native `\u002Floop`** and exit without\n   starting a loop here.\n\nThe frontmatter `description` encodes this same split so auto-invocation routes\ngoal-intent phrasings here and leaves bare interval phrasings to native `\u002Floop`.\n\n## Phase 1 — refine-until-go gate\n\nThe interactive gate that turns a raw operator request into a frozen,\noperator-approved contract before any loop iteration runs. It reuses\n`autospec-refine --interactive` for the refinement rounds (do **not** fork the\nlens logic — per the ROI-check rule, invoke upstream), extracts the\nGOAL \u002F CHECK \u002F MODE contract, shows-and-explains what was understood, and keeps\niterating with the operator until they type the literal unlock token `go`.\n\nThe gate is the core requested behaviour, so it is **on by default**; only\n`--yes` skips it (autonomous mode).\n\n### 1. Trigger disambiguation handoff\n\nApply the **Trigger disambiguation** section above. If the request is bare\ninterval polling intent (*\"loop every 5m\"*, *\"poll every N minutes\"*), redirect\nto the native `\u002Floop` and **stop** — do not enter this gate. If the request is\nambiguous (*\"do a loop\"* with no interval and no goal), the gate's **first\nclarifying question** is goal-or-interval; on an interval answer, redirect to\nnative `\u002Floop` and stop. Otherwise (a goal\u002Fcompletion intent) continue.\n\n### 2. Refinement rounds (reuse autospec-refine)\n\nRun the repo-grounded prompt-improvement rounds by invoking\n**`autospec-refine --interactive`** on the raw request. This reuses the upstream\nlens pipeline (repo-grounding, clarity-ac, sizing, adversarial); do not\nreimplement or fork it. Take the refined prompt that `autospec-refine` produces\nas the input to contract extraction below.\n\nIf the operator passed `--goal '\u003Ccriterion>'`, treat that as the authoritative\nGOAL and skip GOAL inference; still run the refinement rounds to sharpen the\nper-iteration prompt unless `--yes` is also set.\n\n### 3. Contract extraction (GOAL \u002F CHECK \u002F MODE)\n\nFrom the refined prompt, extract three contract artifacts:\n\n- **GOAL** — an explicit, **testable** success criterion (one or more\n  sentences). It must be phrased so that \"done\" is unambiguous. When `--goal`\n  was supplied, use it verbatim.\n- **CHECK** — a **deterministic verifier**: a shell command plus its expected\n  exit status (e.g. a test command, a `grep`, or a `gh` query that exits `0`\n  exactly when the goal holds), **when one is derivable from the refined\n  prompt**. If no deterministic check can be derived, set CHECK to **`null`** —\n  the loop will then rely on a **separate** verifier subagent (never the worker)\n  to judge each iteration. If the operator passed `--check '\u003Ccmd>'`, use that\n  command verbatim and skip CHECK inference; an explicit `--check ''` (empty) or\n  the operator declining a check forces CHECK = `null`.\n- **MODE** — `cumulative` (carry `progress[]` state between iterations;\n  **default**) vs `polling` (stateless re-run of the same refined prompt each\n  iteration, judged on fresh repo state). Default to `cumulative` unless the\n  refined prompt clearly describes stateless polling.\n\n### 4. Show & explain\n\nPresent to the operator, in plain language:\n\n- the **refined prompt verbatim** (exactly what each iteration's worker\n  receives);\n- **what was understood** — the restated GOAL; the CHECK that will gate\n  termination, shown as the literal command + expected exit (or, when CHECK is\n  `null`, the sentence *\"no deterministic check — an independent verifier\n  subagent will judge each iteration against the GOAL\"*); the MODE\n  (`cumulative` \u002F `polling`) and why; and the active guardrails (`max_iters`,\n  `no_progress_K`, `stop.flag`, pre-loop short-circuit);\n- the **per-iteration plan** — what one iteration will do (dispatch worker →\n  persist outcome → evaluate goal → no-progress check → converge-or-continue).\n\n### 5. Clarify loop (unlock on `go`)\n\nAsk the operator to confirm or correct. If they correct or clarify anything,\n**re-run steps 2–4** with the new input and show the updated contract. Repeat\nuntil the operator types the literal unlock token **`go`** (case-insensitive;\nalso accept `\"go ahead\"` and `\"start\"`). Any other response is treated as\nfurther clarification and re-enters the loop — never start the loop on an\nambiguous or partial confirmation.\n\n`--yes` skips this entire gate (steps 4–5): the extracted contract is frozen\nimmediately and the loop starts without operator confirmation. It is **off by\ndefault** because the gate is the core requested behaviour.\n\n### 6. Freeze the contract\n\nInitialize the loop-state file at\n`~\u002F.autospec\u002Floop\u002F\u003Crepo-slug>\u002Floop-state.json` via `loop-state.sh init`, writing\nthe frozen `goal`, `check` (the command string or `null`), and `mode`, plus the\nguardrail defaults (`max_iters`, `no_progress_K` — overridable by `--max-iters`\n\u002F `--no-progress`) and the zeroed counters (`iter`, `no_progress_count`,\n`progress`, `last_result`, `done`, `exit_reason`). Once frozen, the contract is\nauthoritative for the run; proceed to Phase 2.\n\n## Phase 2 — goal-conditioned loop\n\nState lives at `~\u002F.autospec\u002Floop\u002F\u003Crepo-slug>\u002Floop-state.json` (path-scoped slug\nto avoid cross-repo collision; atomic `tmp`+`mv` writes). Drive it **only**\nthrough the scripts that ship with this skill — never hand-edit the JSON:\n\n- `loop-state.sh init --repo \u003Cowner\u002Fname> --goal \u003CGOAL> --check \u003CCMD|null>\n  [--mode cumulative|polling] [--max-iters N] [--no-progress-k N]` — already run\n  by Phase 1 step 6 to freeze the contract.\n- `loop-state.sh read --repo \u003Cowner\u002Fname>` — print the current JSON; parse the\n  fields you need with `jq`.\n- `loop-state.sh update --repo \u003Cowner\u002Fname> --key \u003Cfield> --value \u003Cval>` —\n  atomic single-field write (the script picks the JSON type: `true`\u002F`false` →\n  boolean, all-digits → number, else string).\n- `loop-check.sh '\u003CCMD>'` — run the deterministic CHECK; prints `met` and exits\n  `0` when satisfied, `not-met` and exits `1` otherwise (a missing, empty, or\n  literal `null` command is always `not-met`).\n\n`progress[]` is an array. `loop-state.sh update --value` infers the JSON type\nfrom the value (`true`\u002F`false` → boolean, all-digits → number, **everything else\n→ string**), so it cannot write a real array — passing a serialized array string\nwould store `progress` as a quoted string and corrupt the schema. To append an\niteration summary, read the state, append with `jq`, and write the **whole**\nstate file atomically yourself (`jq '.progress += [\"\u003Csummary>\"]' \"$(loop-state.sh\npath --repo \u003Cowner\u002Fname>)\"` into a `mktemp` then `mv` over the path) — or, the\nsimpler default, keep a human-readable rolling log in `last_result` (a plain\nstring `update` handles correctly) and let Phase 1's `init` seed `progress` as\nthe empty array, treating it as the exit-report accumulator you only append to\nvia the `jq`+`mv` path. Resolve `\u003Cowner\u002Fname>` once at loop start (e.g.\n`gh repo view --json nameWithOwner --jq .nameWithOwner`) and reuse it for every\nscript call so all reads and writes target the same state file.\n\n**Pre-loop short-circuit:** read `check` from state; if it is non-null, run\n`loop-check.sh \"$check\"` once **before** the first iteration. If it prints `met`\n(exit 0), the goal already holds: `loop-state.sh update --key done --value true`,\n`--key exit_reason --value already-satisfied`, emit the exit report, and stop —\nno worker is ever dispatched. If `check` is `null`, skip the short-circuit and\nenter the loop (the in-loop verifier judges each iteration).\n\n**Each iteration** (stop-gates evaluated *before* doing work):\n\n1. **Halt gate** — before dispatching any worker, evaluate the guardrails (see\n   **Guardrails**) against current state. If `~\u002F.autospec\u002Fstop.flag` exists, or\n   `iter >= max_iters`, or `no_progress_count >= no_progress_K`, write the\n   matching `exit_reason` (`stopped-by-flag` \u002F `max-iters` \u002F `no-progress`) and\n   `done=true` via `loop-state.sh update`, emit the exit report, and stop\n   **without** dispatching another worker.\n2. **Work** — dispatch a **worker** subagent (Tier B; **inline on Codex**, which\n   has no subagents) with input = the refined prompt + GOAL + the carried\n   `progress[]` (cumulative mode) or the refined prompt + `last_result` only\n   (polling mode). Instruct it to make forward progress toward GOAL and report\n   concretely what changed (files touched, commands run, observable deltas).\n3. **Persist** — capture the worker's outcome. Set `last_result` to its summary\n   via `loop-state.sh update --key last_result --value \"\u003Csummary>\"`, and append a\n   one-line iteration summary to `progress[]` via the `jq`+`mv` whole-file path\n   above (never via `update`, which would stringify the array). Never carry the\n   raw worker transcript in your own context.\n4. **Goal eval** — if `check` is non-null, run `loop-check.sh \"$check\"`: `met`\n   (exit 0) means the goal holds, `not-met` (exit 1) means keep going. If `check`\n   is `null`, dispatch a **separate verifier** subagent (Tier B; Tier A for\n   high-stakes goals) — **never the worker subagent that produced this result**,\n   to honour the no-self-approval rule — to judge `last_result` \u002F current repo\n   state against GOAL and return met \u002F not-met plus a one-line reason.\n5. **No-progress detection** — did this iteration produce a measurable change\n   toward the goal (CHECK output flipped or moved, files changed, a non-empty\n   `progress[]` delta)? **No** → `no_progress_count++` (read, increment, write).\n   **Yes** → reset `no_progress_count` to `0`.\n6. **Converge** — goal met (`loop-check.sh` printed `met`, or the verifier\n   returned met) → `loop-state.sh update --key done --value true`,\n   `--key exit_reason --value goal-met`, break. Otherwise increment `iter`\n   (`loop-state.sh update --key iter --value \u003Citer+1>`) and continue to the next\n   iteration's halt gate.\n\n**Exit report.** On any loop exit, read the final state once\n(`loop-state.sh read`) and report to the operator, in plain language:\n\n- the **final status** drawn from `exit_reason` — one of `goal-met`,\n  `max-iters`, `no-progress`, `stopped-by-flag`, or `already-satisfied`;\n- the **iteration count** (`iter`) and, when relevant, the `no_progress_count`;\n- the **progress log** — the `progress[]` entries in order, so the operator can\n  see what each iteration accomplished and why the loop stopped.\n\nMake the status unambiguous: distinguish a true `goal-met` convergence from a\nguardrail halt (`max-iters` \u002F `no-progress` \u002F `stopped-by-flag`) so the operator\nknows whether to re-run, raise `--max-iters`, or intervene.\n\n## Guardrails\n\nConservative caps so the loop never spins unbounded (small-model target):\n\n- **stop.flag** — `~\u002F.autospec\u002Fstop.flag` exists (shared with `\u002Fautospec-stop`)\n  → halt with `exit_reason=\"stopped-by-flag\"`.\n- **max-iters** — `iter >= max_iters` (default 25, `--max-iters N`) →\n  `exit_reason=\"max-iters\"`.\n- **no-progress** — `no_progress_count >= no_progress_K` (default 3,\n  `--no-progress K`) → `exit_reason=\"no-progress\"`.\n- **pre-loop short-circuit** — CHECK already satisfied before iterating →\n  `exit_reason=\"already-satisfied\"`, no wasted iteration.\n- **no self-approval** — when CHECK is `null`, goal judgement is always done by\n  a **separate** verifier subagent, never the worker that produced the result.\n\nThe halt gate is evaluated *before* each iteration's work, so a set flag or an\nexhausted cap stops the loop without dispatching another worker.\n","# autospec-loop workflow (harness-neutral)\n\nTake an operator request to \"run something in a loop until a goal is reached\"\nand turn it into (1) an interactive **refine-until-\"go\"** gate that freezes a\ntestable contract, then (2) a **goal-conditioned loop** that re-dispatches the\nrefined prompt with carried state each iteration until a deterministic check —\nor, failing that, an independent verifier — confirms the goal is met, with\nconservative guardrails so it never spins unbounded.\n\nGoal: a harness-neutral, autospec-family way to \"keep doing X until Y\" that\nstops on the *goal* (not a clock), prefers a deterministic CHECK over LLM\nself-assessment, and is safe to run on small-context local models.\n\nManage your own context — never exceed 60%. Delegate per-iteration work to\nsubagents whenever your harness supports it (Codex runs iterations inline); do\nnot carry raw iteration transcripts in the orchestrator context.\n\n\u003C!-- autospec-block:startup-self-update SKILL_NAME=autospec-loop -->\n\n## Self-update mode\n\nIf the argument matches the regex `^\\s*update\\s*$` (case-insensitive,\nwhitespace-padded), this skill enters self-update mode and does not run the\nnormal pipeline. This section is pure prose: never interpolate or shell out the\noperator's argument text.\n\n1. **Detect harness** by checking which install path exists for this skill:\n   - Claude Code: `~\u002F.claude\u002Fskills\u002Fautospec-loop\u002FSKILL.md`\n   - OpenCode:    `~\u002F.config\u002Fopencode\u002Fagent\u002Fautospec-loop.md`\n   - Codex CLI:   `~\u002F.codex\u002Fprompts\u002Fautospec-loop.md`\n2. **Re-install the full autospec suite from `main`** by piping the canonical installer:\n   ```bash\n   curl -fsSL https:\u002F\u002Fraw.githubusercontent.com\u002Fberlinguyinca\u002Fautospec\u002Fmain\u002Fbootstrap.sh | bash -s -- --skill all --harness all --update\n   ```\n   Run this one-liner once; it refreshes all autospec skills across all harnesses.\n3. **Show the diff** between the prior installed file(s) and the freshly fetched copy.\n4. **Stop.** Do not enter the loop pipeline. Print the upgrade summary and return to the user.\n\nIf no install path is detected, print `Self-update: no installed copy of autospec-loop found; run install.sh first.` and exit.\n\n## Model tier\n\nTier A\u002FB model resolution and the harness-detection protocol are inherited\nverbatim from `AGENTS.md` (`## Subagent model selection (two-tier, cost-aware)`\nand `## Subagent vs inline decision matrix`). Refinement-gate reasoning runs at\nTier A (spec work); per-iteration loop work and the independent verifier run at\nTier B (implementation work), escalating to Tier A for high-stakes goals.\n\n> **Model tier:** Phase 1 refine-gate dispatches at Tier A (spec work); Phase 2\n> per-iteration worker and verifier dispatch at TIER_B (implementation work),\n> resolved at startup per the harness detection below.\n\n### Required capabilities & harness adapter\n\nThis workflow assumes the following capabilities. Map each to your harness's\nactual tool; if a capability is missing, use the listed fallback.\n\n| Capability                  | Claude Code                             | OpenCode                              | Codex CLI                                | Fallback if missing                              |\n|-----------------------------|-----------------------------------------|---------------------------------------|------------------------------------------|--------------------------------------------------|\n| Per-iteration worker        | `Agent` (subagent_type=general-purpose) | `task` agent, await output            | run the iteration inline (no subagents)  | Run the iteration in-thread (more context cost)  |\n| Independent verifier        | a **separate** `Agent` dispatch         | a **separate** `task` agent           | a separate inline judging pass           | Judge in-thread, but never reuse the worker pass |\n| Ask the user a question     | `AskUserQuestion`                       | inline prompt                         | inline prompt                            | Ask in the response and wait for the next turn   |\n| Subagent model tier         | Tier A: `opus` + `ultrathink`; Tier B: `sonnet` + medium thinking | Tier A: top `task` model + high reasoning; Tier B: smaller-tier `task` + medium reasoning | Tier A: top GPT + `reasoning_effort=high`; Tier B: current spark\u002Fcost-optimized Codex + `reasoning_effort=medium` | Honor the per-phase tier mapping in AGENTS.md; retry the same subagent UP on unavailability |\n| Subagent dispatch policy    | per AGENTS.md decision matrix           | per AGENTS.md decision matrix         | per AGENTS.md decision matrix            | inline with main-session token cost              |\n\n**Persistent project notes**: write durable preferences to **`AGENTS.md`** in\nthe repo root — recognized by Claude Code, OpenCode, and Codex.\n\n## Harness detection (run once at skill start, before the gate begins)\n\nDetect your harness by checking available tools before any dispatch:\n\n1. **Claude Code** — the `Agent` tool with a `subagent_type` parameter is available.\n   - `TIER_A` = `opus` + `ultrathink`  (the alias, which resolves to the current Claude Opus generation)\n   - `TIER_B` = `sonnet`               (the alias, which resolves to the current Claude Sonnet generation)\n2. **OpenCode** — a `task` tool with model\u002Ftier configuration is available (no `subagent_type`).\n   - `TIER_A` = top-tier task model + high reasoning\n   - `TIER_B` = smaller-tier task model + medium reasoning\n3. **Codex CLI** — neither `Agent` nor a configurable `task` tool is available; iterations run inline.\n   - `TIER_A` = current top GPT model + `reasoning_effort=high`\n   - `TIER_B` = the spark \u002F cost-optimized variant of the configured model when one exists, else the configured model + `reasoning_effort=medium`\n\n**Fallback rule:** If `TIER_B` is not available in your harness, silently retry\nthe same subagent dispatch with `TIER_A`. Preserve parent context on retry.\nNever ask the user.\n\nHold `TIER_A` and `TIER_B` for the entire skill run.\n\n## Trigger disambiguation\n\n`autospec-loop` and the native `\u002Floop` cover different intents; route by\ngoal-vs-interval:\n\n- **autospec-loop claims** goal\u002Fcompletion phrasings: *\"loop until …\"*, *\"in a\n  loop until …\"*, *\"keep going until done\"*, *\"do a loop\"*, *\"execute a loop\"*,\n  *\"run X in a loop until Y\"*, *\"keep doing X until Y\"*.\n- **Native `\u002Floop` keeps** bare interval phrasings: *\"loop every 5m\"*,\n  *\"\u002Floop 5m \u002Ffoo\"*, *\"poll every N minutes\"*, *\"keep running … on an interval\"*.\n- **Ambiguous** bare *\"do a loop\"* \u002F *\"execute a loop\"* with no interval and no\n  explicit goal → autospec-loop claims it, and the refine gate's **first\n  question** is goal-or-interval. If the operator wants interval polling, the\n  skill **redirects to native `\u002Floop`** and exits without starting its own loop.\n\nDecision rule, applied to the raw request before anything else runs:\n\n1. **Explicit interval, no goal** — the request names a cadence (*\"every 5m\"*,\n   *\"every N minutes\"*, *\"on an interval\"*, or the literal `\u002Floop \u003Cinterval>`\n   form) and no completion criterion → **defer to native `\u002Floop`**; do not\n   enter this skill's gate.\n2. **Explicit goal \u002F completion** — the request names a stopping condition\n   (*\"until …\"*, *\"until done\"*, *\"until it passes\"*, *\"keep doing X until Y\"*)\n   → **autospec-loop claims it**; enter the Phase 1 gate.\n3. **Ambiguous** — a bare *\"do a loop\"* \u002F *\"execute a loop\"* with neither an\n   interval nor a goal → autospec-loop claims it, and the gate's **first\n   clarifying question is goal-or-interval**: on a goal answer, continue the\n   gate; on an interval answer, **redirect to native `\u002Floop`** and exit without\n   starting a loop here.\n\nThe frontmatter `description` encodes this same split so auto-invocation routes\ngoal-intent phrasings here and leaves bare interval phrasings to native `\u002Floop`.\n\n## Phase 1 — refine-until-go gate\n\nThe interactive gate that turns a raw operator request into a frozen,\noperator-approved contract before any loop iteration runs. It reuses\n`autospec-refine --interactive` for the refinement rounds (do **not** fork the\nlens logic — per the ROI-check rule, invoke upstream), extracts the\nGOAL \u002F CHECK \u002F MODE contract, shows-and-explains what was understood, and keeps\niterating with the operator until they type the literal unlock token `go`.\n\nThe gate is the core requested behaviour, so it is **on by default**; only\n`--yes` skips it (autonomous mode).\n\n### 1. Trigger disambiguation handoff\n\nApply the **Trigger disambiguation** section above. If the request is bare\ninterval polling intent (*\"loop every 5m\"*, *\"poll every N minutes\"*), redirect\nto the native `\u002Floop` and **stop** — do not enter this gate. If the request is\nambiguous (*\"do a loop\"* with no interval and no goal), the gate's **first\nclarifying question** is goal-or-interval; on an interval answer, redirect to\nnative `\u002Floop` and stop. Otherwise (a goal\u002Fcompletion intent) continue.\n\n### 2. Refinement rounds (reuse autospec-refine)\n\nRun the repo-grounded prompt-improvement rounds by invoking\n**`autospec-refine --interactive`** on the raw request. This reuses the upstream\nlens pipeline (repo-grounding, clarity-ac, sizing, adversarial); do not\nreimplement or fork it. Take the refined prompt that `autospec-refine` produces\nas the input to contract extraction below.\n\nIf the operator passed `--goal '\u003Ccriterion>'`, treat that as the authoritative\nGOAL and skip GOAL inference; still run the refinement rounds to sharpen the\nper-iteration prompt unless `--yes` is also set.\n\n### 3. Contract extraction (GOAL \u002F CHECK \u002F MODE)\n\nFrom the refined prompt, extract three contract artifacts:\n\n- **GOAL** — an explicit, **testable** success criterion (one or more\n  sentences). It must be phrased so that \"done\" is unambiguous. When `--goal`\n  was supplied, use it verbatim.\n- **CHECK** — a **deterministic verifier**: a shell command plus its expected\n  exit status (e.g. a test command, a `grep`, or a `gh` query that exits `0`\n  exactly when the goal holds), **when one is derivable from the refined\n  prompt**. If no deterministic check can be derived, set CHECK to **`null`** —\n  the loop will then rely on a **separate** verifier subagent (never the worker)\n  to judge each iteration. If the operator passed `--check '\u003Ccmd>'`, use that\n  command verbatim and skip CHECK inference; an explicit `--check ''` (empty) or\n  the operator declining a check forces CHECK = `null`.\n- **MODE** — `cumulative` (carry `progress[]` state between iterations;\n  **default**) vs `polling` (stateless re-run of the same refined prompt each\n  iteration, judged on fresh repo state). Default to `cumulative` unless the\n  refined prompt clearly describes stateless polling.\n\n### 4. Show & explain\n\nPresent to the operator, in plain language:\n\n- the **refined prompt verbatim** (exactly what each iteration's worker\n  receives);\n- **what was understood** — the restated GOAL; the CHECK that will gate\n  termination, shown as the literal command + expected exit (or, when CHECK is\n  `null`, the sentence *\"no deterministic check — an independent verifier\n  subagent will judge each iteration against the GOAL\"*); the MODE\n  (`cumulative` \u002F `polling`) and why; and the active guardrails (`max_iters`,\n  `no_progress_K`, `stop.flag`, pre-loop short-circuit);\n- the **per-iteration plan** — what one iteration will do (dispatch worker →\n  persist outcome → evaluate goal → no-progress check → converge-or-continue).\n\n### 5. Clarify loop (unlock on `go`)\n\nAsk the operator to confirm or correct. If they correct or clarify anything,\n**re-run steps 2–4** with the new input and show the updated contract. Repeat\nuntil the operator types the literal unlock token **`go`** (case-insensitive;\nalso accept `\"go ahead\"` and `\"start\"`). Any other response is treated as\nfurther clarification and re-enters the loop — never start the loop on an\nambiguous or partial confirmation.\n\n`--yes` skips this entire gate (steps 4–5): the extracted contract is frozen\nimmediately and the loop starts without operator confirmation. It is **off by\ndefault** because the gate is the core requested behaviour.\n\n### 6. Freeze the contract\n\nInitialize the loop-state file at\n`~\u002F.autospec\u002Floop\u002F\u003Crepo-slug>\u002Floop-state.json` via `loop-state.sh init`, writing\nthe frozen `goal`, `check` (the command string or `null`), and `mode`, plus the\nguardrail defaults (`max_iters`, `no_progress_K` — overridable by `--max-iters`\n\u002F `--no-progress`) and the zeroed counters (`iter`, `no_progress_count`,\n`progress`, `last_result`, `done`, `exit_reason`). Once frozen, the contract is\nauthoritative for the run; proceed to Phase 2.\n\n## Phase 2 — goal-conditioned loop\n\nState lives at `~\u002F.autospec\u002Floop\u002F\u003Crepo-slug>\u002Floop-state.json` (path-scoped slug\nto avoid cross-repo collision; atomic `tmp`+`mv` writes). Drive it **only**\nthrough the scripts that ship with this skill — never hand-edit the JSON:\n\n- `loop-state.sh init --repo \u003Cowner\u002Fname> --goal \u003CGOAL> --check \u003CCMD|null>\n  [--mode cumulative|polling] [--max-iters N] [--no-progress-k N]` — already run\n  by Phase 1 step 6 to freeze the contract.\n- `loop-state.sh read --repo \u003Cowner\u002Fname>` — print the current JSON; parse the\n  fields you need with `jq`.\n- `loop-state.sh update --repo \u003Cowner\u002Fname> --key \u003Cfield> --value \u003Cval>` —\n  atomic single-field write (the script picks the JSON type: `true`\u002F`false` →\n  boolean, all-digits → number, else string).\n- `loop-check.sh '\u003CCMD>'` — run the deterministic CHECK; prints `met` and exits\n  `0` when satisfied, `not-met` and exits `1` otherwise (a missing, empty, or\n  literal `null` command is always `not-met`).\n\n`progress[]` is an array. `loop-state.sh update --value` infers the JSON type\nfrom the value (`true`\u002F`false` → boolean, all-digits → number, **everything else\n→ string**), so it cannot write a real array — passing a serialized array string\nwould store `progress` as a quoted string and corrupt the schema. To append an\niteration summary, read the state, append with `jq`, and write the **whole**\nstate file atomically yourself (`jq '.progress += [\"\u003Csummary>\"]' \"$(loop-state.sh\npath --repo \u003Cowner\u002Fname>)\"` into a `mktemp` then `mv` over the path) — or, the\nsimpler default, keep a human-readable rolling log in `last_result` (a plain\nstring `update` handles correctly) and let Phase 1's `init` seed `progress` as\nthe empty array, treating it as the exit-report accumulator you only append to\nvia the `jq`+`mv` path. Resolve `\u003Cowner\u002Fname>` once at loop start (e.g.\n`gh repo view --json nameWithOwner --jq .nameWithOwner`) and reuse it for every\nscript call so all reads and writes target the same state file.\n\n**Pre-loop short-circuit:** read `check` from state; if it is non-null, run\n`loop-check.sh \"$check\"` once **before** the first iteration. If it prints `met`\n(exit 0), the goal already holds: `loop-state.sh update --key done --value true`,\n`--key exit_reason --value already-satisfied`, emit the exit report, and stop —\nno worker is ever dispatched. If `check` is `null`, skip the short-circuit and\nenter the loop (the in-loop verifier judges each iteration).\n\n**Each iteration** (stop-gates evaluated *before* doing work):\n\n1. **Halt gate** — before dispatching any worker, evaluate the guardrails (see\n   **Guardrails**) against current state. If `~\u002F.autospec\u002Fstop.flag` exists, or\n   `iter >= max_iters`, or `no_progress_count >= no_progress_K`, write the\n   matching `exit_reason` (`stopped-by-flag` \u002F `max-iters` \u002F `no-progress`) and\n   `done=true` via `loop-state.sh update`, emit the exit report, and stop\n   **without** dispatching another worker.\n2. **Work** — dispatch a **worker** subagent (Tier B; **inline on Codex**, which\n   has no subagents) with input = the refined prompt + GOAL + the carried\n   `progress[]` (cumulative mode) or the refined prompt + `last_result` only\n   (polling mode). Instruct it to make forward progress toward GOAL and report\n   concretely what changed (files touched, commands run, observable deltas).\n3. **Persist** — capture the worker's outcome. Set `last_result` to its summary\n   via `loop-state.sh update --key last_result --value \"\u003Csummary>\"`, and append a\n   one-line iteration summary to `progress[]` via the `jq`+`mv` whole-file path\n   above (never via `update`, which would stringify the array). Never carry the\n   raw worker transcript in your own context.\n4. **Goal eval** — if `check` is non-null, run `loop-check.sh \"$check\"`: `met`\n   (exit 0) means the goal holds, `not-met` (exit 1) means keep going. If `check`\n   is `null`, dispatch a **separate verifier** subagent (Tier B; Tier A for\n   high-stakes goals) — **never the worker subagent that produced this result**,\n   to honour the no-self-approval rule — to judge `last_result` \u002F current repo\n   state against GOAL and return met \u002F not-met plus a one-line reason.\n5. **No-progress detection** — did this iteration produce a measurable change\n   toward the goal (CHECK output flipped or moved, files changed, a non-empty\n   `progress[]` delta)? **No** → `no_progress_count++` (read, increment, write).\n   **Yes** → reset `no_progress_count` to `0`.\n6. **Converge** — goal met (`loop-check.sh` printed `met`, or the verifier\n   returned met) → `loop-state.sh update --key done --value true`,\n   `--key exit_reason --value goal-met`, break. Otherwise increment `iter`\n   (`loop-state.sh update --key iter --value \u003Citer+1>`) and continue to the next\n   iteration's halt gate.\n\n**Exit report.** On any loop exit, read the final state once\n(`loop-state.sh read`) and report to the operator, in plain language:\n\n- the **final status** drawn from `exit_reason` — one of `goal-met`,\n  `max-iters`, `no-progress`, `stopped-by-flag`, or `already-satisfied`;\n- the **iteration count** (`iter`) and, when relevant, the `no_progress_count`;\n- the **progress log** — the `progress[]` entries in order, so the operator can\n  see what each iteration accomplished and why the loop stopped.\n\nMake the status unambiguous: distinguish a true `goal-met` convergence from a\nguardrail halt (`max-iters` \u002F `no-progress` \u002F `stopped-by-flag`) so the operator\nknows whether to re-run, raise `--max-iters`, or intervene.\n\n## Guardrails\n\nConservative caps so the loop never spins unbounded (small-model target):\n\n- **stop.flag** — `~\u002F.autospec\u002Fstop.flag` exists (shared with `\u002Fautospec-stop`)\n  → halt with `exit_reason=\"stopped-by-flag\"`.\n- **max-iters** — `iter >= max_iters` (default 25, `--max-iters N`) →\n  `exit_reason=\"max-iters\"`.\n- **no-progress** — `no_progress_count >= no_progress_K` (default 3,\n  `--no-progress K`) → `exit_reason=\"no-progress\"`.\n- **pre-loop short-circuit** — CHECK already satisfied before iterating →\n  `exit_reason=\"already-satisfied\"`, no wasted iteration.\n- **no self-approval** — when CHECK is `null`, goal judgement is always done by\n  a **separate** verifier subagent, never the worker that produced the result.\n\nThe halt gate is evaluated *before* each iteration's work, so a set flag or an\nexhausted cap stops the loop without dispatching another worker.\n",{"name":5,"description":7},"main","9e7f04357ff46ab731d4f56a4ac5ca5449c13bb0","skills\u002Fautospec-loop",1,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec","https:\u002F\u002Fgithub.com\u002Fberlinguyinca","Love to lead, code, cook and contribute to open source and believe that you can learn anything you set your mind too.\r\n","2026-11-11 07:09:35",{"source":8,"skillName":5,"commands":58,"prompt":68,"zipAvailable":69},[59,62,65],{"tool":60,"command":61},"npx","npx skills add berlinguyinca\u002Fautospec --skill autospec-loop",{"tool":63,"command":64},"bunx","bunx skills add berlinguyinca\u002Fautospec --skill autospec-loop",{"tool":66,"command":67},"pnpm","pnpm dlx skills add berlinguyinca\u002Fautospec --skill autospec-loop","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002FSKILL.md",true,[71,77,83,87,91,95,99,103],{"path":72,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":75,"sourceHtmlUrl":76},"codex\u002Fprompt.md","REFERENCE","参考文档",19078,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Fcodex\u002Fprompt.md",{"path":78,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":81,"sourceHtmlUrl":82},"install.sh","SCRIPT","脚本",10787,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Finstall.sh",{"path":84,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":85,"sourceHtmlUrl":86},"opencode\u002Fagent.md",19519,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Fopencode\u002Fagent.md",{"path":88,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":89,"sourceHtmlUrl":90},"README.md",4214,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002FREADME.md",{"path":92,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":93,"sourceHtmlUrl":94},"scripts\u002Floop-check.sh",966,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Fscripts\u002Floop-check.sh",{"path":96,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":97,"sourceHtmlUrl":98},"scripts\u002Floop-state.sh",8717,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Fscripts\u002Floop-state.sh",{"path":100,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":101,"sourceHtmlUrl":102},"uninstall.sh",4031,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Funinstall.sh",{"path":104,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":105,"sourceHtmlUrl":106},"validate.sh",8788,"https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-loop\u002Fvalidate.sh",[108,121,130,139,164,187],{"skillNo":109,"name":110,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":118,"publishedAt":44},"01M71RJHYMKPTC9AAPXWQ6F4RT","trio-pass-skill","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Flockstep-trio-pass\u002FSKILL.md",[113],{"categoryNo":114,"categoryName":115,"taxonomyType":116,"taxonomyTypeDesc":117},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[119,120],"trio","trio pass skill",{"skillNo":122,"name":123,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":124,"categories":125,"keywords":127,"publishedAt":44},"01M71RJFBQ1KAF9XM3T3Y4VMSK","duo-pass-skill","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Flockstep-duo-pass\u002FSKILL.md",[126],{"categoryNo":114,"categoryName":115,"taxonomyType":116,"taxonomyTypeDesc":117},[128,129],"duo pass skill","duo",{"skillNo":131,"name":132,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":133,"categories":134,"keywords":136,"publishedAt":44},"01M71RJDT1TB3P9TRWR6A534FB","duo-divergence-skill","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Flockstep-duo-divergence\u002FSKILL.md",[135],{"categoryNo":114,"categoryName":115,"taxonomyType":116,"taxonomyTypeDesc":117},[137,129,138],"duo divergence skill","divergence",{"skillNo":140,"name":24,"nameZh":6,"description":141,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":142,"categories":143,"keywords":147,"publishedAt":44},"01M71RJAC4NNEQSYBDSY3NZS73","Use when the user wants to ship a feature end-to-end across multiple commits — bootstraps repo if missing, brainstorms a design spec, decomposes into linked GitHub issues, or splits an existing spec into issues, then runs an autonomous implementation loop with auto-merge.","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec\u002FSKILL.md",[144],{"categoryNo":145,"categoryName":146,"taxonomyType":116,"taxonomyTypeDesc":117},"01JBSEEDCAT000000000000015","视觉设计",[24,148,149,150,151,152,153,25,38,154,155,156,157,158,159,160,161,162,163,39],"ship a feature","end","issues","design","autonomous","auto","across","design spec","merge","ship","implementation loop","github issues","feature","commits","repo","missing",{"skillNo":165,"name":166,"nameZh":6,"description":167,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":168,"categories":169,"keywords":170,"publishedAt":44},"01M71RJ4F9HT8PT26CKNM9XE1M","autospec-upgrade","Use when the user wants to upgrade an Angular, Next.js, or React project to the latest official versions safely — by locking observable behavior before touching versions, upgrading one major at a time with official codemods, and gating completion on mutation score rather than coverage theater.","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-upgrade\u002FSKILL.md",[],[171,24,172,173,174,175,176,177,178,38,179,180,181,182,39,183,184,34,185,186],"autospec upgrade","upgrade","mutation","coverage","official","project","time","gating","angular","react","next","at","Next.js","one","behavior","latest",{"skillNo":188,"name":189,"nameZh":6,"description":190,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":191,"categories":192,"keywords":196,"publishedAt":44},"01M71RJ28FJ4PM6DXHB7T5CF8H","autospec-ui-audit","Use when a React Router application needs a deterministic whole-site route inventory before UI, UX, navigation, or evidence auditing begins.","https:\u002F\u002Fgithub.com\u002Fberlinguyinca\u002Fautospec\u002Fblob\u002Fmain\u002Fskills\u002Fautospec-ui-audit\u002FSKILL.md",[193],{"categoryNo":194,"categoryName":195,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000032","设计师",[197,24,198,199,200,39,201,202,203,204,205,206,207,208,209,180,210,38,211],"ui audit","audit","autospec ui audit","ui","inventory","evidence","deterministic","route","ux","site","react router","navigation","application","needs","router"]