[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M756PYDS6DPV9HMZ1CJCGX93":3,"skill-same:01M756PYDS6DPV9HMZ1CJCGX93":62},{"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":24,"commitSha":25,"skillDirectory":26,"forksCount":13,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M756PYDS6DPV9HMZ1CJCGX93","merge-request:fix",null,"Monitor an open pull\u002Fmerge request (by id) and, as feedback arrives, gather every actionable item — human + bot\u002FAI review comments and MR-attributable CI failures — deduplicated against a durable ledger so nothing is handled twice. Polls on an adaptive scheduled-wakeup cadence until the PR\u002FMR closes. First runs \u002Fdetect-source-control and hard-stops when the forge is unsupported. Use as \"watch the PR\", \"monitor the MR for feedback\", \"fix review feedback on \u003Cid>\", or the continuation of \u002Fmerge-request:create.","SoftwareAssassin2\u002FPlugins","SoftwareAssassin2","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F38842958?v=4","01M756NS55ARTH4KENH9HKVRNZ","01M756NS6B3QB1XR6QN6BPMR76",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Ffix\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: merge-request:fix\ndescription: Monitor an open pull\u002Fmerge request (by id) and, as feedback arrives, gather every actionable item — human + bot\u002FAI review comments and MR-attributable CI failures — deduplicated against a durable ledger so nothing is handled twice. Polls on an adaptive scheduled-wakeup cadence until the PR\u002FMR closes. First runs \u002Fdetect-source-control and hard-stops when the forge is unsupported. Use as \"watch the PR\", \"monitor the MR for feedback\", \"fix review feedback on \u003Cid>\", or the continuation of \u002Fmerge-request:create.\nargument-hint: \"\u003CID>  — the PR\u002FMR number to monitor (required)\"\n---\n\n# merge-request:fix\n\nWatch an open PR\u002FMR and keep the review loop moving: on an adaptive interval,\ngather freshly-arrived feedback (human comments, bot\u002FAI-reviewer comments, and\nCI failures), deduplicate it against a durable ledger, classify CI failures by\nattribution, and hand the unseen, actionable items to the triage-and-apply half\n(fn-10.2). Keep polling via scheduled wakeups until the PR\u002FMR is closed.\n\nThis is an **assistant-executed** skill: you (the assistant) follow the steps\nbelow. The deterministic forge work — fetching comments\u002Fthreads\u002FCI, computing\nstable dedupe keys, diffing against the `## Handled` ledger, and writing ledger\nrecords — lives in `scripts\u002Fgather-feedback.sh`. The poll loop itself is *your*\ncontinuation driven by the harness `ScheduleWakeup` tool; a script cannot\nschedule its own next wake.\n\n> **Scope note:** the **monitoring core** (fn-10.1) is the `\u003CID>` gate, the\n> wakeup loop, intent reload, feedback gathering + dedupe, CI attribution, and\n> the reported\u002Fnon-actionable CI write-back (Steps 0–4, 6, 7). The\n> **decide-and-apply** loop (fn-10.2) — implement criteria, the spec guardrail,\n> test-before-push, `Fixed`+resolve, and `## Declined` — is **Step 5** below.\n\n## Step 0 — require the `\u003CID>` argument\n\n`\u002Fmerge-request:fix` needs the PR\u002FMR **id** to monitor. If none was provided,\n**do nothing** beyond a clear message and stop:\n\n> `\u002Fmerge-request:fix` needs a PR\u002FMR id to monitor, e.g. `\u002Fmerge-request:fix 128`.\n> Nothing to do without one. (Run `\u002Fmerge-request:create` first to open one.)\n\nNever guess an id or pick \"the latest\" PR\u002FMR — absence of an id means there is\nnothing for this skill to do.\n\n## Step 1 — detect the forge and hard-stop if unsupported\n\nInvoke the shared **`\u002Fdetect-source-control`** skill and capture its stdout block\n**and** exit code. Apply the **fn-8 hard-stop contract** (see `..\u002F..\u002FREADME.md`),\nexactly as `\u002Fmerge-request:create` does:\n\n- **Non-zero exit** → operational failure (not a git repo, `git` missing). Stop\n  and report; nothing was changed.\n- **Exit `0` and `supported=false`** → a *successful* detection of an unsupported\n  forge. Stop, naming the detected `forge`\u002F`host`.\n- **Exit `0` and `supported=true`** → proceed. Remember `forge` (`github` or\n  `gitlab`) — you pass it to `gather-feedback.sh`.\n\nNever key the unsupported stop off a non-zero exit; `forge=unsupported` is a\nnormal exit-`0` result.\n\n## Step 2 — reload the intent \u002F spec context\n\nRead the shared artifact `.data\u002Fmerge\u002F\u003CID>.md` (written by `\u002Fmerge-request:create`).\nIts `## Intent` block is the author's pre-MR intent, and it survives a resumed\nsession — load it so your later judgement is anchored to what the change was\n*for*.\n\n- If `## Intent` is present (and not the `[TODO] Intent not provided` placeholder),\n  use it as the primary intent.\n- If it is **absent or a placeholder**, fall back to the change scope: the\n  flow-next epic when one is discoverable, plus the PR\u002FMR title\u002Fbody and the\n  branch diff (`## Change scope` is already stashed in the file). Just make sure\n  you have the best-available intent loaded before acting.\n\n(The full **cumulative** spec guardrail — Intent + PR\u002FMR body + linked `.flow`\nspecs + branch diff, asking on conflict — is assembled in **Step 5a** before you\ndecide anything.)\n\n## Step 3 — gather feedback and dedupe against the ledger\n\nRun the gather helper for the resolved forge and id:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" gather \\\n  --forge \u003Cgithub|gitlab> --id \u003CID>\n```\n\nIt fetches, for the PR\u002FMR:\n\n- **plain issue comments** (`kind=comment`, non-resolvable),\n- **review threads** (`kind=thread`; `source_id` is the resolvable thread id —\n  GitLab discussion id \u002F GitHub review-thread node id — that fn-10.2 resolves),\n- **failing CI jobs** (`kind=ci-job`),\n\neach carrying a **stable id** and a **dedupe key** (`content_hash` for\nthread\u002Fcomment, `fingerprint` for ci-job), then **diffs them against the\n`## Handled` ledger** and prints only the **unseen-or-changed** items as JSONL,\nfollowed by a machine-readable trailer:\n\n```\nMR_STATE=open|closed|merged|unknown\nMR_UPDATED=\u003Ciso8601|unknown>\nMR_CADENCE=active|idle\nMR_POLL_SECONDS=\u003Cint>\nMR_FETCHED_COUNT=\u003Cint>\nMR_ACTIONABLE_COUNT=\u003Cint>\n```\n\nParse the JSONL lines (each begins with `{`) as your actionable feedback set, and\nthe `MR_*` trailer for the loop. **Dedupe is the script's job — never re-fetch or\nre-diff by hand.** Its guarantees:\n\n- **CI failures dedupe on the logical `fingerprint`** ({forge, job name, commit\n  SHA, normalized error signature}) — *not* the job\u002Fpipeline id, which changes on\n  every rerun. A rerun of the same failure is \"seen\"; a new commit re-surfaces it.\n- **thread\u002Fcomment dedupe on `source_id` + `content_hash`** — the same id with a\n  changed hash (an edited comment or a new reply) counts as *changed* and comes\n  back for reconsideration.\n- A standing `pending-user` ledger record therefore **suppresses re-asking** that\n  item until its source changes (Step 6).\n\n## Step 4 — classify CI failures by attribution (and record the ones you don't act on)\n\nFor each `kind=ci-job` item, decide **MR-attributable vs non-actionable** before\nanything downstream touches it:\n\n- **MR-attributable** (pass downstream to fn-10.2): the failing test\u002Ffile is\n  within — or exercises — the changed files, or the same job passes on the target\n  branch baseline. Use the item's `job_name`\u002F`signature`\u002F`commit` plus the logs\n  (pull them via `gh`\u002F`glab` when you need more than the signature) to judge.\n- **Non-actionable** (do **not** fix): network\u002Ftimeout, runner\u002Fimage\u002Fcache\u002Fsetup\n  failures, failures pre-existing on the target branch, and flaky\u002Fknown-intermittent\n  jobs. **Report** these (surface them to the user) but never auto-fix them.\n- **Ambiguous** attribution → treat as pending-user (Step 6): report and ask\n  rather than auto-fixing.\n\n**fn-10.1 owns the write path for the CI items you do NOT hand downstream.**\nImmediately after reporting a non-actionable (or reported-ambiguous) CI item,\nwrite its `## Handled` record so the next wakeup does not re-report it:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind ci-job --decision \u003Cnon-actionable|reported> \\\n  --fingerprint \u003Cfingerprint-from-the-item> \\\n  --commit \u003Ccommit-from-the-item> \\\n  --rationale \"\u003Cwhy it isn't MR-attributable>\"\n```\n\nThe `--fingerprint` **must** be the item's own `fingerprint` (that is the CI\ndedupe key). Only the **MR-attributable** failures go on to Step 5; fn-10.2 writes\ntheir terminal `## Handled` records when it implements or declines them.\n\n## Step 5 — decide and apply (the fn-10.2 loop)\n\nYou now hold the unseen items Step 3\u002F4 handed you: human comments, bot\u002FAI\ncomments, and **MR-attributable** CI failures. For each one, decide whether it\nwarrants a change, implement the worthy ones (tests first), acknowledge +\nresolve, record the declined ones, and write a terminal `## Handled` record.\n**Every** item you touch here ends with exactly one `## Handled` record written\nthrough `gather-feedback.sh record` — never hand-edit the ledger.\n\n### Step 5a — assemble the spec guardrail (cumulative, not a chain)\n\nBefore judging anything, load **all** available guardrail sources and treat them\ntogether — none replaces another:\n\n1. `## Intent` in `.data\u002Fmerge\u002F\u003CID>.md` (loaded in Step 2) — the author's pre-MR intent.\n2. The **PR\u002FMR title and body** (`gh pr view \u003CID>` \u002F `glab mr view \u003CID>`).\n3. Any **linked `.flow\u002Fspecs`\u002Ftasks** discoverable by id in the branch name, the\n   PR\u002FMR body, or commit messages (e.g. `fn-10`, `fn-10.2`) — read the epic\u002Ftask\n   with `flowctl cat \u003Cid>` when one is discoverable.\n4. The **branch diff \u002F commit range** `\u003Cdefault>..HEAD` (the `## Change scope`\n   stash already holds this) — it describes *scope* only.\n\nThe diff **never overrides a discoverable epic**: it tells you what this change\ntouches, not what the change is allowed to be. When the sources **conflict or are\nincomplete** — or when a feedback item's spec-impact is genuinely ambiguous —\n**ask the user once** and record `pending-user` (Step 6) rather than guessing.\n\n### Step 5b — resolve the repo's published standards\n\nRead the repo's technical standards in this precedence: **`CLAUDE.md` \u002F\n`AGENTS.md` first, then `docs\u002F`.** These are what \"consistent with repo\nstandards\" (criterion 2 below) is measured against. Absence of any standard is\nnot license to invent one — fall back to the surrounding code's conventions.\n\n### Step 5c — the three-criteria implement decision\n\nImplement a feedback item **only when all three hold**:\n\n1. **It does not change the spec** — the change stays inside the Step-5a\n   guardrail (fixes\u002Fclarifies the existing scope; never adds or redefines it).\n2. **It is consistent with the repo's published standards** (Step 5b).\n3. **It flags a legitimate shortcoming** in the MR\u002FPR code (a real bug, gap, or\n   quality issue — not taste\u002Fpreference, not already-correct code).\n\nOutcomes:\n\n- **All three hold → implement** (Step 5d), then acknowledge + record (Step 5e).\n- **Any one fails → decline** (Step 5f).\n- **Spec-impact ambiguous** (you cannot tell whether it would change the spec) →\n  **ask once**, record `pending-user` (Step 6); do not implement or decline yet.\n\n### Step 5d — implement, then the test-before-push gate\n\nFor each worthy item, make the change following the surrounding code + Step-5b\nstandards. Then, **before committing**:\n\n1. **Discover the test command** — in this order: `CLAUDE.md`\u002F`AGENTS.md`,\n   package scripts (`package.json`, `pyproject.toml`, …), a `Makefile`\u002F`justfile`\n   target, or the CI config. Run the **relevant** automated tests for what you\n   changed.\n2. **Commit + push only on pass.** On a green run, commit the change and push to\n   the PR\u002FMR branch. Capture the commit SHA — you record it in Step 5e.\n3. **No test command found → BLOCKER, not a warning.** Stop, tell the user no\n   automated test command could be found, and do **not** commit or push without\n   their explicit approval — never an untested push. Record the item as\n   `pending-user` with rationale `\"no automated test command found\"` (via the\n   Step 6 command) so wakeups stop re-processing it until the user answers or the\n   source changes.\n4. **Tests fail** → the change isn't ready: fix it and re-run, or (if the item\n   can't be satisfied within the guardrail) decline it (Step 5f). Never push red.\n\n### Step 5e — acknowledge, resolve, and record an implemented item\n\nOnce the change is committed + pushed, post exactly `Fixed` and resolve the\noriginating thread through the forge-specific helper:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Freply-resolve.sh\" \\\n  --forge \u003Cgithub|gitlab> --id \u003CID> \\\n  --kind \u003Cthread|comment> --source-id \u003Citem.source_id>\n```\n\n- `kind=thread` → it replies `Fixed` **and** resolves (GitLab resolve-discussion \u002F\n  GitHub `resolveReviewThread`). A `RESOLVED=0` in its output \u002F a non-zero exit\n  means the thread stayed open — surface it.\n- `kind=comment` → non-resolvable plain comment: it posts `Fixed` as a general\n  note and reports `RESOLVED=skipped` (the documented fallback).\n- **`kind=ci-job` has no thread** — do **not** call `reply-resolve.sh`; the pushed\n  commit *is* the acknowledgement. Go straight to the ledger record.\n\nThen write the terminal `## Handled` record — carrying the item's dedupe key\n**unchanged** from the gather JSONL (`--content-hash` + `--source-id` for\nthread\u002Fcomment, `--fingerprint` for ci-job):\n\n```bash\n# thread \u002F comment:\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind \u003Cthread|comment> --decision implemented \\\n  --source-id \u003Citem.source_id> --content-hash \u003Citem.content_hash> \\\n  --commit \u003Ccommit-sha>\n\n# ci-job:\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind ci-job --decision implemented \\\n  --fingerprint \u003Citem.fingerprint> --commit \u003Ccommit-sha>\n```\n\n### Step 5f — record a declined item\n\nWhen an item fails any of the three criteria, **do not change code**. Append the\nitem and a short rationale to the `## Declined` section of `.data\u002Fmerge\u002F\u003CID>.md`\n(create the heading if absent; it is append-only prose), e.g.:\n\n```markdown\n## Declined\n\n- **\u003Cthread\u002Fcomment\u002FCI ref>** — \u003Cone-line why>: \u003Cdeclined because it would change\n  the spec \u002F conflicts with \u003Cstandard> \u002F isn't a legitimate shortcoming>.\n```\n\nThen write the terminal `## Handled` record with `--decision declined` and the\nrationale, carrying the same per-kind dedupe key:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind \u003Cthread|comment> --decision declined \\\n  --source-id \u003Citem.source_id> --content-hash \u003Citem.content_hash> \\\n  --rationale \"\u003Cwhy it was declined>\"\n# (ci-job declines use --fingerprint \u003Citem.fingerprint> instead of --source-id\u002F--content-hash)\n```\n\nThe `## Declined` prose is the human-readable record (per the user's request);\nthe `## Handled` record is the machine dedupe key so the next wakeup never\nre-surfaces the same declined item.\n\n## Step 6 — ambiguous items: ask once, record `pending-user`\n\nWhen an item's spec-impact or CI attribution is genuinely ambiguous, **ask the\nuser once**, then record it so wakeups stop re-asking:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind \u003Cthread|comment|ci-job> --decision pending-user \\\n  \u003Cdedupe-key flags: --content-hash \u003Chash> --source-id \u003Cid>  | --fingerprint \u003Cfp>> \\\n  --rationale \"\u003Cthe question you asked>\"\n```\n\nWhile a `pending-user` record stands, the Step-3 dedupe **suppresses re-asking**\nthat item — until either (a) its **source changes** (a new commit \u002F `fingerprint`\nfor CI, or a new `content_hash` for a thread\u002Fcomment) or (b) the **user answers**.\nThe user's answer **supersedes** the `pending-user` record: reprocess that one\nitem **exactly once** (implement or decline it via fn-10.2, which writes the\nterminal record), rather than waiting for the next fetch.\n\n## Step 7 — schedule the next wakeup (or terminate)\n\nRead `MR_STATE` from the trailer:\n\n- **`MR_STATE=closed` or `merged`** → the PR\u002FMR is closed. **Terminate the loop**\n  and report a short summary (what was handled this session). Do not schedule\n  another wakeup.\n- **otherwise** → schedule the next wakeup with the harness **`ScheduleWakeup`**\n  tool, after `MR_POLL_SECONDS` (the adaptive cadence hint: ~2–3 min while active,\n  ~15–20 min when idle — you may adjust within that band). The wakeup **payload\n  must carry enough state to run stateless**, because each wake re-enters this\n  skill fresh:\n\n  | payload field | value |\n  |---------------|-------|\n  | `id`          | the PR\u002FMR `\u003CID>` |\n  | `forge`       | `github` \u002F `gitlab` (from Step 1) |\n  | `stash`       | `.data\u002Fmerge\u002F\u003CID>.md` (holds `## Intent`\u002F`## Declined`\u002F`## Handled`) |\n  | `last_activity` | `MR_UPDATED` from the trailer |\n  | `cadence`     | `MR_CADENCE` (`active`\u002F`idle`) |\n  | `terminate_when` | PR\u002FMR closed |\n\n  The wakeup prompt re-invokes `\u002Fmerge-request:fix \u003CID>` — which re-runs Steps 1–7\n  from the top. Because gathering is deduped against the durable `## Handled`\n  ledger, a fresh wake never re-handles prior feedback.\n\n## Notes\n\n- **Session-bound.** Scheduled wakeups live only for the current session; keep\n  active-poll intervals under ~5 min (the prompt-cache window). If the session\n  ends before the PR\u002FMR closes, monitoring stops — resume by simply rerunning\n  `\u002Fmerge-request:fix \u003CID>`. The `## Handled` ledger makes resume idempotent.\n- **The ledger is the source of truth for \"already handled\".** Every terminal\n  outcome writes one record: reported \u002F non-actionable CI here (fn-10.1);\n  implemented \u002F declined and `pending-user` in fn-10.2. Never re-derive dedupe by\n  hand — always go through `gather-feedback.sh`.\n- **Read-then-write ordering.** Record a CI item as reported\u002Fnon-actionable\n  *immediately after* you report it, before scheduling the next wakeup, so a\n  quick next wake can't re-surface it.\n","# merge-request:fix\n\nWatch an open PR\u002FMR and keep the review loop moving: on an adaptive interval,\ngather freshly-arrived feedback (human comments, bot\u002FAI-reviewer comments, and\nCI failures), deduplicate it against a durable ledger, classify CI failures by\nattribution, and hand the unseen, actionable items to the triage-and-apply half\n(fn-10.2). Keep polling via scheduled wakeups until the PR\u002FMR is closed.\n\nThis is an **assistant-executed** skill: you (the assistant) follow the steps\nbelow. The deterministic forge work — fetching comments\u002Fthreads\u002FCI, computing\nstable dedupe keys, diffing against the `## Handled` ledger, and writing ledger\nrecords — lives in `scripts\u002Fgather-feedback.sh`. The poll loop itself is *your*\ncontinuation driven by the harness `ScheduleWakeup` tool; a script cannot\nschedule its own next wake.\n\n> **Scope note:** the **monitoring core** (fn-10.1) is the `\u003CID>` gate, the\n> wakeup loop, intent reload, feedback gathering + dedupe, CI attribution, and\n> the reported\u002Fnon-actionable CI write-back (Steps 0–4, 6, 7). The\n> **decide-and-apply** loop (fn-10.2) — implement criteria, the spec guardrail,\n> test-before-push, `Fixed`+resolve, and `## Declined` — is **Step 5** below.\n\n## Step 0 — require the `\u003CID>` argument\n\n`\u002Fmerge-request:fix` needs the PR\u002FMR **id** to monitor. If none was provided,\n**do nothing** beyond a clear message and stop:\n\n> `\u002Fmerge-request:fix` needs a PR\u002FMR id to monitor, e.g. `\u002Fmerge-request:fix 128`.\n> Nothing to do without one. (Run `\u002Fmerge-request:create` first to open one.)\n\nNever guess an id or pick \"the latest\" PR\u002FMR — absence of an id means there is\nnothing for this skill to do.\n\n## Step 1 — detect the forge and hard-stop if unsupported\n\nInvoke the shared **`\u002Fdetect-source-control`** skill and capture its stdout block\n**and** exit code. Apply the **fn-8 hard-stop contract** (see `..\u002F..\u002FREADME.md`),\nexactly as `\u002Fmerge-request:create` does:\n\n- **Non-zero exit** → operational failure (not a git repo, `git` missing). Stop\n  and report; nothing was changed.\n- **Exit `0` and `supported=false`** → a *successful* detection of an unsupported\n  forge. Stop, naming the detected `forge`\u002F`host`.\n- **Exit `0` and `supported=true`** → proceed. Remember `forge` (`github` or\n  `gitlab`) — you pass it to `gather-feedback.sh`.\n\nNever key the unsupported stop off a non-zero exit; `forge=unsupported` is a\nnormal exit-`0` result.\n\n## Step 2 — reload the intent \u002F spec context\n\nRead the shared artifact `.data\u002Fmerge\u002F\u003CID>.md` (written by `\u002Fmerge-request:create`).\nIts `## Intent` block is the author's pre-MR intent, and it survives a resumed\nsession — load it so your later judgement is anchored to what the change was\n*for*.\n\n- If `## Intent` is present (and not the `[TODO] Intent not provided` placeholder),\n  use it as the primary intent.\n- If it is **absent or a placeholder**, fall back to the change scope: the\n  flow-next epic when one is discoverable, plus the PR\u002FMR title\u002Fbody and the\n  branch diff (`## Change scope` is already stashed in the file). Just make sure\n  you have the best-available intent loaded before acting.\n\n(The full **cumulative** spec guardrail — Intent + PR\u002FMR body + linked `.flow`\nspecs + branch diff, asking on conflict — is assembled in **Step 5a** before you\ndecide anything.)\n\n## Step 3 — gather feedback and dedupe against the ledger\n\nRun the gather helper for the resolved forge and id:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" gather \\\n  --forge \u003Cgithub|gitlab> --id \u003CID>\n```\n\nIt fetches, for the PR\u002FMR:\n\n- **plain issue comments** (`kind=comment`, non-resolvable),\n- **review threads** (`kind=thread`; `source_id` is the resolvable thread id —\n  GitLab discussion id \u002F GitHub review-thread node id — that fn-10.2 resolves),\n- **failing CI jobs** (`kind=ci-job`),\n\neach carrying a **stable id** and a **dedupe key** (`content_hash` for\nthread\u002Fcomment, `fingerprint` for ci-job), then **diffs them against the\n`## Handled` ledger** and prints only the **unseen-or-changed** items as JSONL,\nfollowed by a machine-readable trailer:\n\n```\nMR_STATE=open|closed|merged|unknown\nMR_UPDATED=\u003Ciso8601|unknown>\nMR_CADENCE=active|idle\nMR_POLL_SECONDS=\u003Cint>\nMR_FETCHED_COUNT=\u003Cint>\nMR_ACTIONABLE_COUNT=\u003Cint>\n```\n\nParse the JSONL lines (each begins with `{`) as your actionable feedback set, and\nthe `MR_*` trailer for the loop. **Dedupe is the script's job — never re-fetch or\nre-diff by hand.** Its guarantees:\n\n- **CI failures dedupe on the logical `fingerprint`** ({forge, job name, commit\n  SHA, normalized error signature}) — *not* the job\u002Fpipeline id, which changes on\n  every rerun. A rerun of the same failure is \"seen\"; a new commit re-surfaces it.\n- **thread\u002Fcomment dedupe on `source_id` + `content_hash`** — the same id with a\n  changed hash (an edited comment or a new reply) counts as *changed* and comes\n  back for reconsideration.\n- A standing `pending-user` ledger record therefore **suppresses re-asking** that\n  item until its source changes (Step 6).\n\n## Step 4 — classify CI failures by attribution (and record the ones you don't act on)\n\nFor each `kind=ci-job` item, decide **MR-attributable vs non-actionable** before\nanything downstream touches it:\n\n- **MR-attributable** (pass downstream to fn-10.2): the failing test\u002Ffile is\n  within — or exercises — the changed files, or the same job passes on the target\n  branch baseline. Use the item's `job_name`\u002F`signature`\u002F`commit` plus the logs\n  (pull them via `gh`\u002F`glab` when you need more than the signature) to judge.\n- **Non-actionable** (do **not** fix): network\u002Ftimeout, runner\u002Fimage\u002Fcache\u002Fsetup\n  failures, failures pre-existing on the target branch, and flaky\u002Fknown-intermittent\n  jobs. **Report** these (surface them to the user) but never auto-fix them.\n- **Ambiguous** attribution → treat as pending-user (Step 6): report and ask\n  rather than auto-fixing.\n\n**fn-10.1 owns the write path for the CI items you do NOT hand downstream.**\nImmediately after reporting a non-actionable (or reported-ambiguous) CI item,\nwrite its `## Handled` record so the next wakeup does not re-report it:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind ci-job --decision \u003Cnon-actionable|reported> \\\n  --fingerprint \u003Cfingerprint-from-the-item> \\\n  --commit \u003Ccommit-from-the-item> \\\n  --rationale \"\u003Cwhy it isn't MR-attributable>\"\n```\n\nThe `--fingerprint` **must** be the item's own `fingerprint` (that is the CI\ndedupe key). Only the **MR-attributable** failures go on to Step 5; fn-10.2 writes\ntheir terminal `## Handled` records when it implements or declines them.\n\n## Step 5 — decide and apply (the fn-10.2 loop)\n\nYou now hold the unseen items Step 3\u002F4 handed you: human comments, bot\u002FAI\ncomments, and **MR-attributable** CI failures. For each one, decide whether it\nwarrants a change, implement the worthy ones (tests first), acknowledge +\nresolve, record the declined ones, and write a terminal `## Handled` record.\n**Every** item you touch here ends with exactly one `## Handled` record written\nthrough `gather-feedback.sh record` — never hand-edit the ledger.\n\n### Step 5a — assemble the spec guardrail (cumulative, not a chain)\n\nBefore judging anything, load **all** available guardrail sources and treat them\ntogether — none replaces another:\n\n1. `## Intent` in `.data\u002Fmerge\u002F\u003CID>.md` (loaded in Step 2) — the author's pre-MR intent.\n2. The **PR\u002FMR title and body** (`gh pr view \u003CID>` \u002F `glab mr view \u003CID>`).\n3. Any **linked `.flow\u002Fspecs`\u002Ftasks** discoverable by id in the branch name, the\n   PR\u002FMR body, or commit messages (e.g. `fn-10`, `fn-10.2`) — read the epic\u002Ftask\n   with `flowctl cat \u003Cid>` when one is discoverable.\n4. The **branch diff \u002F commit range** `\u003Cdefault>..HEAD` (the `## Change scope`\n   stash already holds this) — it describes *scope* only.\n\nThe diff **never overrides a discoverable epic**: it tells you what this change\ntouches, not what the change is allowed to be. When the sources **conflict or are\nincomplete** — or when a feedback item's spec-impact is genuinely ambiguous —\n**ask the user once** and record `pending-user` (Step 6) rather than guessing.\n\n### Step 5b — resolve the repo's published standards\n\nRead the repo's technical standards in this precedence: **`CLAUDE.md` \u002F\n`AGENTS.md` first, then `docs\u002F`.** These are what \"consistent with repo\nstandards\" (criterion 2 below) is measured against. Absence of any standard is\nnot license to invent one — fall back to the surrounding code's conventions.\n\n### Step 5c — the three-criteria implement decision\n\nImplement a feedback item **only when all three hold**:\n\n1. **It does not change the spec** — the change stays inside the Step-5a\n   guardrail (fixes\u002Fclarifies the existing scope; never adds or redefines it).\n2. **It is consistent with the repo's published standards** (Step 5b).\n3. **It flags a legitimate shortcoming** in the MR\u002FPR code (a real bug, gap, or\n   quality issue — not taste\u002Fpreference, not already-correct code).\n\nOutcomes:\n\n- **All three hold → implement** (Step 5d), then acknowledge + record (Step 5e).\n- **Any one fails → decline** (Step 5f).\n- **Spec-impact ambiguous** (you cannot tell whether it would change the spec) →\n  **ask once**, record `pending-user` (Step 6); do not implement or decline yet.\n\n### Step 5d — implement, then the test-before-push gate\n\nFor each worthy item, make the change following the surrounding code + Step-5b\nstandards. Then, **before committing**:\n\n1. **Discover the test command** — in this order: `CLAUDE.md`\u002F`AGENTS.md`,\n   package scripts (`package.json`, `pyproject.toml`, …), a `Makefile`\u002F`justfile`\n   target, or the CI config. Run the **relevant** automated tests for what you\n   changed.\n2. **Commit + push only on pass.** On a green run, commit the change and push to\n   the PR\u002FMR branch. Capture the commit SHA — you record it in Step 5e.\n3. **No test command found → BLOCKER, not a warning.** Stop, tell the user no\n   automated test command could be found, and do **not** commit or push without\n   their explicit approval — never an untested push. Record the item as\n   `pending-user` with rationale `\"no automated test command found\"` (via the\n   Step 6 command) so wakeups stop re-processing it until the user answers or the\n   source changes.\n4. **Tests fail** → the change isn't ready: fix it and re-run, or (if the item\n   can't be satisfied within the guardrail) decline it (Step 5f). Never push red.\n\n### Step 5e — acknowledge, resolve, and record an implemented item\n\nOnce the change is committed + pushed, post exactly `Fixed` and resolve the\noriginating thread through the forge-specific helper:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Freply-resolve.sh\" \\\n  --forge \u003Cgithub|gitlab> --id \u003CID> \\\n  --kind \u003Cthread|comment> --source-id \u003Citem.source_id>\n```\n\n- `kind=thread` → it replies `Fixed` **and** resolves (GitLab resolve-discussion \u002F\n  GitHub `resolveReviewThread`). A `RESOLVED=0` in its output \u002F a non-zero exit\n  means the thread stayed open — surface it.\n- `kind=comment` → non-resolvable plain comment: it posts `Fixed` as a general\n  note and reports `RESOLVED=skipped` (the documented fallback).\n- **`kind=ci-job` has no thread** — do **not** call `reply-resolve.sh`; the pushed\n  commit *is* the acknowledgement. Go straight to the ledger record.\n\nThen write the terminal `## Handled` record — carrying the item's dedupe key\n**unchanged** from the gather JSONL (`--content-hash` + `--source-id` for\nthread\u002Fcomment, `--fingerprint` for ci-job):\n\n```bash\n# thread \u002F comment:\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind \u003Cthread|comment> --decision implemented \\\n  --source-id \u003Citem.source_id> --content-hash \u003Citem.content_hash> \\\n  --commit \u003Ccommit-sha>\n\n# ci-job:\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind ci-job --decision implemented \\\n  --fingerprint \u003Citem.fingerprint> --commit \u003Ccommit-sha>\n```\n\n### Step 5f — record a declined item\n\nWhen an item fails any of the three criteria, **do not change code**. Append the\nitem and a short rationale to the `## Declined` section of `.data\u002Fmerge\u002F\u003CID>.md`\n(create the heading if absent; it is append-only prose), e.g.:\n\n```markdown\n## Declined\n\n- **\u003Cthread\u002Fcomment\u002FCI ref>** — \u003Cone-line why>: \u003Cdeclined because it would change\n  the spec \u002F conflicts with \u003Cstandard> \u002F isn't a legitimate shortcoming>.\n```\n\nThen write the terminal `## Handled` record with `--decision declined` and the\nrationale, carrying the same per-kind dedupe key:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind \u003Cthread|comment> --decision declined \\\n  --source-id \u003Citem.source_id> --content-hash \u003Citem.content_hash> \\\n  --rationale \"\u003Cwhy it was declined>\"\n# (ci-job declines use --fingerprint \u003Citem.fingerprint> instead of --source-id\u002F--content-hash)\n```\n\nThe `## Declined` prose is the human-readable record (per the user's request);\nthe `## Handled` record is the machine dedupe key so the next wakeup never\nre-surfaces the same declined item.\n\n## Step 6 — ambiguous items: ask once, record `pending-user`\n\nWhen an item's spec-impact or CI attribution is genuinely ambiguous, **ask the\nuser once**, then record it so wakeups stop re-asking:\n\n```bash\nbash \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh\" record \\\n  --id \u003CID> --kind \u003Cthread|comment|ci-job> --decision pending-user \\\n  \u003Cdedupe-key flags: --content-hash \u003Chash> --source-id \u003Cid>  | --fingerprint \u003Cfp>> \\\n  --rationale \"\u003Cthe question you asked>\"\n```\n\nWhile a `pending-user` record stands, the Step-3 dedupe **suppresses re-asking**\nthat item — until either (a) its **source changes** (a new commit \u002F `fingerprint`\nfor CI, or a new `content_hash` for a thread\u002Fcomment) or (b) the **user answers**.\nThe user's answer **supersedes** the `pending-user` record: reprocess that one\nitem **exactly once** (implement or decline it via fn-10.2, which writes the\nterminal record), rather than waiting for the next fetch.\n\n## Step 7 — schedule the next wakeup (or terminate)\n\nRead `MR_STATE` from the trailer:\n\n- **`MR_STATE=closed` or `merged`** → the PR\u002FMR is closed. **Terminate the loop**\n  and report a short summary (what was handled this session). Do not schedule\n  another wakeup.\n- **otherwise** → schedule the next wakeup with the harness **`ScheduleWakeup`**\n  tool, after `MR_POLL_SECONDS` (the adaptive cadence hint: ~2–3 min while active,\n  ~15–20 min when idle — you may adjust within that band). The wakeup **payload\n  must carry enough state to run stateless**, because each wake re-enters this\n  skill fresh:\n\n  | payload field | value |\n  |---------------|-------|\n  | `id`          | the PR\u002FMR `\u003CID>` |\n  | `forge`       | `github` \u002F `gitlab` (from Step 1) |\n  | `stash`       | `.data\u002Fmerge\u002F\u003CID>.md` (holds `## Intent`\u002F`## Declined`\u002F`## Handled`) |\n  | `last_activity` | `MR_UPDATED` from the trailer |\n  | `cadence`     | `MR_CADENCE` (`active`\u002F`idle`) |\n  | `terminate_when` | PR\u002FMR closed |\n\n  The wakeup prompt re-invokes `\u002Fmerge-request:fix \u003CID>` — which re-runs Steps 1–7\n  from the top. Because gathering is deduped against the durable `## Handled`\n  ledger, a fresh wake never re-handles prior feedback.\n\n## Notes\n\n- **Session-bound.** Scheduled wakeups live only for the current session; keep\n  active-poll intervals under ~5 min (the prompt-cache window). If the session\n  ends before the PR\u002FMR closes, monitoring stops — resume by simply rerunning\n  `\u002Fmerge-request:fix \u003CID>`. The `## Handled` ledger makes resume idempotent.\n- **The ledger is the source of truth for \"already handled\".** Every terminal\n  outcome writes one record: reported \u002F non-actionable CI here (fn-10.1);\n  implemented \u002F declined and `pending-user` in fn-10.2. Never re-derive dedupe by\n  hand — always go through `gather-feedback.sh`.\n- **Read-then-write ordering.** Record a CI item as reported\u002Fnon-actionable\n  *immediately after* you report it, before scheduling the next wakeup, so a\n  quick next wake can't re-surface it.\n",{"name":5,"description":7,"argument-hint":23},"\u003CID>  — the PR\u002FMR number to monitor (required)","main","d01d9bbc815bb1c77bf61f24152dd84fca29ca41","plugins\u002Fmerge-request\u002Fskills\u002Ffix","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2","2026-07-23 21:04:05",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add SoftwareAssassin2\u002FPlugins --skill \"merge-request:fix\"",{"tool":36,"command":37},"bunx","bunx skills add SoftwareAssassin2\u002FPlugins --skill \"merge-request:fix\"",{"tool":39,"command":40},"pnpm","pnpm dlx skills add SoftwareAssassin2\u002FPlugins --skill \"merge-request:fix\"","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Ffix\u002FSKILL.md",true,[44,50,54,58],{"path":45,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":48,"sourceHtmlUrl":49},"scripts\u002Fgather-feedback.sh","SCRIPT","脚本",26527,"https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Ffix\u002Fscripts\u002Fgather-feedback.sh",{"path":51,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":52,"sourceHtmlUrl":53},"scripts\u002Freply-resolve.sh",8238,"https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Ffix\u002Fscripts\u002Freply-resolve.sh",{"path":55,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":56,"sourceHtmlUrl":57},"tests\u002Fgather-feedback_test.sh",13669,"https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Ffix\u002Ftests\u002Fgather-feedback_test.sh",{"path":59,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":60,"sourceHtmlUrl":61},"tests\u002Freply-resolve_test.sh",6941,"https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Ffix\u002Ftests\u002Freply-resolve_test.sh",[63,70,77,84,91,98],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":18},"01M756R6NM9JCPKPDDEYYQ8756","wendy","Summon Wendy, a performance coach modeled on Wendy Rhoades from Billions, when the user invokes \u002Fwendy or asks for Wendy by name, or explicitly asks to get his head right — tilting, stalling, spiraling, avoiding, flat, or needing to lock in before a hard push. Wendy works the operator's state, never the plan. Do NOT use for feedback on strategies, plans, or decisions, and do not trigger uninvited.","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fwendy\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":18},"01M756R3CHWZP6F4XEYZAHD0E7","ubiquitous-language","Extract a DDD-style ubiquitous language glossary from the current conversation, flagging ambiguities and proposing canonical terms. Saves to .\u002Fdocs\u002Fubiquitous-language.md. Use when user wants to define domain terms, build a glossary, harden terminology, create a ubiquitous language, or mentions \"domain model\" or \"DDD\". Forked from Matt Pocock's \"ubiquitous-language\" skill.","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fubiquitous-language\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":18},"01M756QYHCNZKH4EW9CGJPB4GT","tdd","Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs using TDD, mentions \"red-green-refactor\", wants integration tests, or asks for test-first development. Forked from Matt Pocock's \"tdd\" skill.","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Ftdd\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":18},"01M756QSQJ0WCP4HPW3SHBAFD5","voice","Speak Claude's responses aloud through the native OS voice (macOS `say`) — on demand via \u002Fspeak:speak (last response) or \u002Fspeak:specify (pick one), or automatically on every response via a toggleable Stop hook. Works locally on a Mac and, through a manually started host listener, from inside a Dev Container. Use when the user wants responses read aloud \u002F spoken \u002F voice output, wants auto-speak turned on or off, or asks whether the speak install is healthy.","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fspeak\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":18},"01M756QGEMD1DF8YQ22FJ5DCAE","merge-request:review","Review a pull\u002Fmerge request (by id) or batch-review every open one against Chris's own engineering standards, staging high-signal findings to .data\u002Fmerge\u002F\u003CID>.md for later posting. Fetches head SHAs and applies the \"Reviewed at commit\" skip check BEFORE any checkout, resolves the head fork-safely, builds\u002Ftests it in a dedicated worktree, and treats a failing build\u002Fcheckout as the top blocking finding. First runs \u002Fdetect-source-control and hard-stops when the forge is unsupported. Use as \"review the PR\", \"review the open MRs\", \"any PRs need my eyes\", \"review \u003Cid>\".","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Freview\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":18},"01M756Q7S9ABZSR6AT45GEDA9T","merge-request:post-findings","Take the findings staged by \u002Fmerge-request:review in .data\u002Fmerge\u002F\u003CID>.md and, with per-finding approval, post the approved ones to the forge as inline comments (general-comment fallback when a finding has no valid diff location) — or formally sign off a clean PR\u002FMR with a single \"Looks good.\" First runs \u002Fdetect-source-control and hard-stops when the forge is unsupported. Nothing reaches the forge except through the bundled scripts. Also runs a propose-then-confirm learned-preferences loop that teaches \u002Fmerge-request:review what you value (Don't-raise \u002F Wording \u002F Confirmed-valued), and records skipped findings to the artifact's","https:\u002F\u002Fgithub.com\u002FSoftwareAssassin2\u002FPlugins\u002Fblob\u002Fmain\u002Fplugins\u002Fmerge-request\u002Fskills\u002Fpost-findings\u002FSKILL.md",[],[]]