[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7ASMGQ491HWWAB930THZD84":3,"skill-same:01M7ASMGQ491HWWAB930THZD84":75},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":29,"updatedAt":30,"install":31,"resources":44},"01M7ASMGQ491HWWAB930THZD84","fix",null,"ozzycodes2\u002Fclaudellum","ozzycodes2","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F21285094?v=4","01M7ASJ57KTP519F96S4QGM7C5","Osniel Quintana","01M7ASJ589JS8S075AA92Y08S4",1,"MIT","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: fix\ndescription: Apply and commit fixes for the findings a prior \u002Freview:review recorded — refactor code smells using the named Fowler\u002FKerievsky refactoring in the language's own idiom, rewrite flagged documentation and comments for clarity, regenerate stale generated artifacts, and defer anything needing a human design decision rather than guessing. Then push the head branch once the suite is green and answer the review comments that raised them: reply with the fix and resolve the thread. Use after \u002Freview:review when you want the findings actually resolved on the PR, one commit per finding. Edits files, commits, pushes the PR branch, replies and resolves; never force-pushes, never merges. Also the fix engine behind monitor-pr.\nallowed-tools: Bash Read Write Edit Grep Glob Skill Task Agent\ndisable-model-invocation: false\n---\n\n# Fix Review Findings\n\n`\u002Freview:review` names the refactoring that resolves each finding and then posts\nit as a comment. This skill is the consumer that `required_actions` was written\nfor: it reads the durable findings ledger, applies each fix, commits one finding\nat a time, and answers the comment that asked.\n\nIt **edits, commits, pushes the PR's head branch once the suite is green, then\nreplies and resolves.** The push is what makes acknowledgement honest: a thread\nis resolved only when its commit is visible on the remote (Step 7.1), so the run\nthat pushes is the run that can close the loop.\n\nUnder `monitor-pr` the push stays with the caller, which passes `--no-push`.\n\n## Inputs\n\n```\n\u002Freview:fix [\u003Cpr-number|pr-url>] [--repo owner\u002Fname]\n            [--dry-run]            plan and partition only; write nothing\n            [--max-concurrent N]   parallel agents, default 4\n            [--force]              break a stale ledger lock\n            [--ack-only]           skip Steps 1–6.5; only answer review comments\n                                   for findings whose commits are already pushed\n            [--no-ack]             fix, commit, and push; reply to and resolve\n                                   nothing\n            [--post-deferred]      post one PR comment for deferrals that have\n                                   no thread to reply to (off by default)\n            [--no-push]            commit only; leave the branch ahead of origin\n                                   (what monitor-pr passes — it owns the push)\n```\n\nWith no argument, infer the PR from the current branch (`gh pr view --json\nnumber`). There is no `--only` or `--severity` flag: the scope is every open\nfinding. Selection is `status == \"open\"`, which also lets `monitor-pr` process\njust the findings appended since its last tick.\n\n## Step 0: Resolve the ledger\n\n```bash\nPLUGIN_ROOT=\"${CLAUDE_PLUGIN_ROOT:-}\"\nif [ ! -f \"$PLUGIN_ROOT\u002Fscripts\u002Ffindings_ledger.py\" ]; then\n  PLUGIN_ROOT=\"$(cd \"${CLAUDE_SKILL_DIR:-.}\u002F..\u002F..\" 2>\u002Fdev\u002Fnull && pwd)\"\nfi\nLEDGER_PY=\"$PLUGIN_ROOT\u002Fscripts\u002Ffindings_ledger.py\"\n\nPR_NUMBER=\"${1:-$(gh pr view --json number -q .number)}\"\nLEDGER=$(python3 \"$LEDGER_PY\" path \\\n  --repo \"$(gh repo view --json nameWithOwner -q .nameWithOwner)\" --pr \"$PR_NUMBER\")\nRUN_ID=\"fix-$(date -u +%Y%m%dT%H%M%SZ)-$$\"\n```\n\n`LEDGER_PY` is a path, and `python3` stays at the call site. Folding the\ninterpreter into the variable works in bash and silently breaks in zsh, which does\nnot word-split an unquoted expansion.\n\nIf the ledger is missing:\n\n> No findings ledger for PR #\\\u003CN>. Run `\u002Freview:review \u003Curl>` first.\n\n**Do not reconstruct one by scraping the posted review.** The inline-comment\nround-trip loses `remedy`, `refactoring`, `drift_class`, and `code_says` — which\nare exactly the fields that select a strategy.\n\n## Step 1: Lock, reconcile, and establish green\n\n### 1.1 Take the lock\n\n```bash\npython3 \"$LEDGER_PY\" lock --ledger \"$LEDGER\" --run \"$RUN_ID\"   # add --force to break a stale lock\n```\n\nOne `fix` run per ledger, for the same reason `monitor-pr` allows one monitor per\nPR: two runs on one branch means double commits and interleaved edits. A lock\nolder than two hours is stale — report it and offer `--force`.\n\n### 1.2 Reconcile against git, not against the ledger\n\nThe ledger records what a previous run *believed*. Verify it:\n\n- For each `committed` finding, `git cat-file -e \u003Csha>^{commit}`. A sha that no\n  longer exists (branch reset, worktree swap) demotes the finding to `open`.\n- For each `claimed` or `applied` finding, check `git status --porcelain` for its\n  recorded files — a previous run died mid-flight.\n- **Hard stop if the tree is dirty with edits `fix` cannot attribute** to a\n  finding in the ledger. Report the paths and ask. Never `git stash`, never\n  `git restore` a path you did not write.\n  - One exemption: `.claude\u002Fpr-monitor\u002Fpr-*.md`. That is `monitor-pr`'s tracker,\n    written by the caller that just invoked this skill, so it is dirty on every\n    delegated run. Leave it alone and never stage it — a monitor's scratch file\n    has no business in a fix commit.\n- For attributable in-flight edits, show the diff and ask: commit as-is, or\n  discard and redo that finding.\n\n### 1.3 Check the ledger is not stale\n\nCompare `pr.head_sha` against current `HEAD`. If the branch has moved since the\nreview, **every `line` in the ledger is suspect.** Warn, and treat agent rule 5\n(locate by content, not by line) as mandatory rather than advisory.\n\n### 1.4 Establish a green baseline\n\n```bash\n# only when context.test_setup.can_run_safely is true\n\u003Ccontext.test_setup.test_command>\n```\n\n**You cannot refactor on red.** If the baseline fails, nothing afterwards can\ndistinguish breakage this run caused from breakage it inherited. On a red\nbaseline, stop and report — unless the user asks to fix the baseline first, which\nis its own finding.\n\nWhere `can_run_safely` is false there is no baseline and no iteration loop; see\n`shared\u002Ffix-contract.md`.\n\n## Step 2: Route every open finding\n\n```bash\npython3 \"$LEDGER_PY\" show --ledger \"$LEDGER\" --status open --json\n```\n\nRead `references\u002Fstrategy-routing.md` and assign each finding exactly one\nstrategy: **S0** formatter, **S1** refactor, **S2** repair, **S3** prose, **S4**\ncomment, **S5** regenerate, **S6** defer. First match wins.\n\nEvery finding gets a strategy. Nothing is dropped. S6 is the honest-deferral\npath, not a filter — if most findings land there, the routing is too lax and the\nreport must make that visible.\n\n### Findings with no `path`\n\nCommon and deliberate: architectural findings and CI-ingested findings carry no\nlocation. Do not guess a file and do not drop the finding.\n\n1. If the `body` or `remedy` names a symbol, `rg` for it and derive the file set.\n2. Confirm the symptom at the candidate before editing.\n3. Zero candidates, or more than three → **S6** `ambiguous-remedy`.\n\n**Never dispatch an agent with neither a file anchor nor a symbol.** That is how\nthis skill would rewrite half the repository.\n\n## Step 3: Order the work\n\nPer `references\u002Fstrategy-routing.md`:\n\n1. **S5 first** for any path S1\u002FS3\u002FS4 also touch — regeneration overwrites hand\n   edits.\n2. **S2** — correctness and security.\n3. **S3, S4, S0.**\n4. **S1 last** — where the suite could not run these are unverified, and at the\n   tip of the branch an unwind is one `git revert`.\n5. Within a sequential group, **descending line number.**\n\n## Step 4: Dispatch\n\nPartition open findings by **file set**. A rename's file set is every file that\nreferences the symbol — find them with LSP `findReferences`, or `rg -w`, before\npartitioning, because the rename tool rewrites all of them. Disjoint sets get one agent each, all\ndispatched in a single message. Findings sharing any file collapse into one\nagent that handles them sequentially, re-reading the file between findings. Cap\nconcurrency at `--max-concurrent` (default 4).\n\n**Agents edit; this skill commits and verifies.** There is one `.git\u002Findex` per\nworktree, so concurrent commits from several agents interleave and a commit ends\nup carrying another agent's half-finished work; and a test suite with shared\nfixtures — a database it resets, a port it binds — corrupts the run when four\nagents start it at once. The agent prompts forbid every `.git`-writing command\nand every suite invocation; the orchestrator stages the reported paths and runs\nthe suite itself.\n\nThe partition is therefore both the concurrency unit and the verification unit.\n\nPin **both** `subagent_type` and `model` on every dispatch — an omitted `model`\ninherits the session model and collapses the intended split:\n\n| Strategy | `subagent_type` | `model` |\n|---|---|---|\n| S1 design-level | `general-purpose` | `opus` |\n| S1 mechanical | `general-purpose` | `sonnet` |\n| S2 | `general-purpose` | `opus` |\n| S3, S4 | `general-purpose` | `haiku` |\n\nUse the verbatim prompts in `references\u002Fsubagent-prompts.md`, with the catalog\nrow, the idiom rows, the tool rows, and the rules **inlined** — a subagent runs\nin the reviewed repo and cannot resolve a path inside this skill.\n\n### Deterministic first\n\nRead `references\u002Frefactoring-tools.md`. A move a tool covers is applied by the\ntool, not rewritten by a model: the prompt carries the tool row, and the agent\nreports the command in `tool_applied`. A tool the table marks\norchestrator-only — it rewrites the whole project — you run yourself, like\nS0, before dispatching the partition it touches, and commit it as that\nfinding.\n\nS0, S5, and S6 need no agent: run the formatter, run the regeneration recipe, or\nrecord the deferral.\n\n### S5 — resolve the regeneration command\n\nIn order: a command named literally in `remedy`; an entry in\n`context.build_system.commands`; a task-runner recipe whose name matches the\nartifact. None resolves → **S6** `no-regen-command`. Commit only the generator's\noutput, and prove idempotence by running it twice (`git diff --exit-code` clean).\n\n### S0 — run the repo's formatter, not a reported diff\n\nA formatting finding may arrive with a diff attached — from a linter the security\nscan ran, or from a CI job `monitor-pr` ingested. Treat it as evidence that drift\nexists, discard the diff, and run `context.build_system.commands.lint` autofix:\nthe tool is authoritative and a review-time diff is stale by now. One batch\ncommit. No lint command → **S6** `no-lint-command`.\n\n## Step 5: Commit per finding, verify per partition\n\nAgents never run the suite. As each agent reports `applied`, commit its finding\nas reported. After a partition's last commit, run\n`context.test_setup.test_command` once.\n\nVerifying per finding costs one suite run per finding; verifying per partition\ncosts one per partition. On a suite that takes minutes — which is most of them —\nthat difference is the dominant term in how long a fix round takes, and the\ninformation it buys is recoverable: a red partition is a handful of commits, and\n`git bisect` finds the guilty one in a couple of runs.\n\n**When the partition goes red:**\n\n1. `git bisect` across the partition's commits, using the suite as the test.\n2. Re-dispatch that finding with the failure output attached. Adjust the change,\n   never the test's expectations unless the finding is about the test.\n   **Do not revert. Do not commit a WIP. Do not defer.**\n3. Re-run the suite. Repeat until green.\n\n**Never leave the branch red.** The claim each commit makes — that the branch is\ngreen at that commit — is written at commit time and made true at partition\ntime. A fix round that ends red has not ended.\n\n**When the loop stalls**, escalate rather than abandon: if the failure output\nstops changing, re-dispatch on a stronger model with the full failure history. If\nthat also stalls, **surface the finding to the user as a live question and pause\non it.** Never silently defer it and never move on as though it were handled.\n\nNo runnable suite (`context.test_setup.can_run_safely` is false) → commit as\nreported and say so in the Step 8 report; CI is the verification.\n\nCommit per `shared\u002Ffix-contract.md` — headline, body, preservation line, trailer\n— then record it:\n\n```bash\npython3 \"$LEDGER_PY\" resolve --ledger \"$LEDGER\" --id \u003CID> --sha \"$(git rev-parse HEAD)\" --strategy S1\n```\n\n## Step 6: Record deferrals\n\n```bash\npython3 \"$LEDGER_PY\" defer --ledger \"$LEDGER\" --id \u003CID> --reason \u003Creason> \\\n  --decision \"\u003Cthe one-sentence question a human must answer>\" --run \"$RUN_ID\"\n```\n\nA deferred finding gets **no edit and no commit**, keeps its original severity,\nand never gets a bare `TODO` as a consolation prize. This skill does not file\ntickets.\n\n## Step 6.5: Push the head branch\n\nEvery commit from Step 5 sits on a branch that is green and verified. Push it —\nthat is what lets Step 7 resolve the threads in this same run instead of handing\nthe user a second command.\n\nSkip entirely under `--no-push`, `--dry-run`, or `--ack-only`, and skip when\nStep 5 made no commits.\n\n### 6.5.1 Gate on green\n\nPush only when **all** of these hold:\n\n- Every partition's final suite run passed. A partition still red, or a finding\n  Step 5 paused on, blocks the push for the whole run — do not push \"the green\n  part\". The branch tip is what CI and reviewers see.\n- No finding is in `claimed` or `applied` state. Everything is `committed` or\n  `deferred`.\n- The tree is clean: `git status --porcelain` is empty apart from the\n  `.claude\u002Fpr-monitor\u002Fpr-*.md` exemption from Step 1.2.\n\nDeferrals do **not** block the push. A deferral produced no commit, so the\npushed branch is exactly the verified work.\n\nWhere `context.test_setup.can_run_safely` is false there is no local green\nsignal and CI is the verification — push, and mark every commit `Unverified` in\nthe Step 8 report. Withholding the push there withholds the only signal\navailable.\n\n### 6.5.2 Push\n\n```bash\nHEAD_BRANCH=$(gh pr view \"$PR_NUMBER\" --json headRefName -q .headRefName)\nCROSS_REPO=$(gh pr view \"$PR_NUMBER\" --json isCrossRepository -q .isCrossRepository)\n\ntest \"$(git rev-parse --abbrev-ref HEAD)\" = \"$HEAD_BRANCH\"   # refuse otherwise\ntest \"$CROSS_REPO\" = \"false\"                                  # refuse otherwise\n\ngit push origin \"HEAD:refs\u002Fheads\u002F$HEAD_BRANCH\"\n```\n\nChecked out somewhere other than the PR's head branch, or the PR comes from a\nfork whose head repo is not `origin` → **do not push.** Report it and let Step 7\nfall back to acknowledging nothing. Guessing a remote pushes someone's work to\nthe wrong place.\n\n**A rejected push is reported, never forced around.** `non-fast-forward` means\nthe remote moved while this run was working. Do not force-push, do not rebase,\ndo not pull — all three are on the deny-list and all three can discard a\ncollaborator's commit. Say the branch diverged, name the commits that stayed\nlocal, and stop; the user re-runs after reconciling.\n\nPush once, after the last partition. Pushing per commit re-triggers CI on every\none of them, which costs a full pipeline per finding and leaves a trail of runs\nnobody reads.\n\n## Step 7: Answer the review comments\n\n`\u002Freview:review` asked each of these questions as an inline comment. A finding\nresolved in silence leaves the reviewer with an open thread and no idea it was\nhandled, so close the loop: reply to the comment that raised it, and resolve the\nthread when the work is actually on the PR.\n\n### 7.1 Only acknowledge what the PR can show\n\n```bash\ngit fetch origin \"$HEAD_BRANCH\" --quiet\ngit merge-base --is-ancestor \u003Csha> \"origin\u002F$HEAD_BRANCH\"   # per committed finding\n```\n\n**A thread is resolved only when its commit is on the remote.** Run the check\neven on a run that just pushed — it is what catches a partial push, a rejected\npush, and a sha the ledger recorded from an earlier run that no longer exists.\nResolving a thread for a commit that exists only on this machine tells a\nreviewer to go look at something they cannot see.\n\nAfter a successful Step 6.5 the check passes for everything this run committed,\nand the loop closes in one command. When Step 6.5 was skipped or refused, it\nfails for those commits — that is the correct outcome, not a bug — and the work\nfinishes later:\n\n```\n\u002Freview:fix 123 --ack-only      # acknowledge pushed work only\n```\n\nUnder `monitor-pr` this resolves itself — it passes `--no-push` and calls `fix`\nbefore pushing, so nothing is acknowledged here, and `monitor-pr` acknowledges\nin its Phase 3 after its push. One acknowledgement per thread, from whoever can\nsee the commit.\n\n### 7.2 Map findings to threads\n\n```bash\ngh api graphql -f query='\nquery($owner:String!,$repo:String!,$pr:Int!){\n  repository(owner:$owner,name:$repo){\n    pullRequest(number:$pr){\n      reviewThreads(first:100){\n        nodes{\n          id isResolved isOutdated\n          comments(first:1){ nodes{ databaseId author{login} path line originalLine body } }\n        }\n      }\n    }\n  }\n}' -F owner=\"$OWNER\" -F repo=\"$NAME\" -F pr=\"$PR_NUMBER\" \\\n  | python3 \"$PLUGIN_ROOT\u002Fscripts\u002Fpr_threads.py\" match --ledger \"$LEDGER\" --threads - --json\n```\n\nThe script attributes a thread to `\u002Freview:review` by the Conventional Comments\nmarker its comments lead with (`**issue (blocking):** …`), so it works whichever\naccount posted the review. It pairs a thread with a finding on same-file plus\nsame-line or same-title, and returns `matched`, `ambiguous`,\n`unmatched_findings`, `unmatched_threads`.\n\n**Act only on `matched`.** An `ambiguous` entry means two findings fit one thread\nequally well — report it and leave the thread alone. A wrong match answers a\nreviewer's question with the resolution of a different question, which is worse\nthan leaving it open.\n\nA `low` confidence match is still a match, but say so in the report so it can be\nspot-checked. Findings with `reason: no-path` never had an inline comment — CI\nfailures and architectural findings — and are not a gap.\n\n### 7.3 Reply, then resolve\n\nPer `matched` entry, reply first. If the reply fails, do not resolve: a resolved\nthread with no explanation is worse than an unresolved one.\n\n```bash\ngh api graphql -f query='mutation($t:ID!,$b:String!){\n  addPullRequestReviewThreadReply(input:{pullRequestReviewThreadId:$t, body:$b}){\n    comment{ url }\n  }\n}' -F t=\"\u003Cthread_id>\" -F b=\"\u003Cbody>\"\n```\n\n**Committed** — one line of what changed, and the sha:\n\n> Fixed in `\u003Csha>` — extracted the retry loop into `Transport::send_with_retry`\n> so the backoff lives with the transport. `\u003Ctest_command>` passes.\n\n**Deferred** — the question, not an apology. The thread stays open:\n\n> Not fixed — this needs a decision first: outbox table, CDC, or accept the\n> window? Each changes the failure mode, so picking one here would be a guess.\n\nThen resolve, for `action: reply+resolve` only:\n\n```bash\ngh api graphql -f query='mutation($id:ID!){\n  resolveReviewThread(input:{threadId:$id}){ thread{ isResolved } }\n}' -F id=\"\u003Cthread_id>\"\n```\n\nNever resolve a `reply` action. A deferral is the one finding that most needs a\nhuman to see it, and resolving it hides it.\n\nWrite the reply from the diff and the finding, not from this run's shape. No\nstrategy codes, no finding IDs, no \"as requested\" — the reviewer is reading a PR,\nnot this skill's output. `F012` means nothing to them.\n\n### 7.4 Record it\n\n```bash\npython3 \"$LEDGER_PY\" ack --ledger \"$LEDGER\" --id \u003CID> --thread \u003Cthread_id> \\\n  --action reply+resolve --url \"\u003Creply url>\" --run \"$RUN_ID\"\n```\n\nIdempotent per `(finding, thread)`, and the matcher skips anything already\nrecorded. That is what stops a re-run — or a `monitor-pr` tick after a manual\n`fix` — from replying to the same reviewer twice.\n\n`--no-ack` skips this whole step.\n\n## Step 8: Report\n\n```\nFix run complete — PR #\u003CN>\n\nResolved (\u003Cn>):\n  \u003Csha>  refactor: move rate limiting into the transport   F003\n  \u003Csha>  docs: tighten the auth overview                   F007\n\nDeferred (\u003Cn>):\n  F012 · src\u002Foutbox.rs:88 · dual-write with no outbox —\n         decision: outbox table, CDC, or accept the window?\n\nUnverified (\u003Cn>):   # only when the suite could not run\n  \u003Csha>  F005 — suite not run (\u003Csafety_notes>); revert with `git revert \u003Csha>`\n\nAcknowledged (\u003Cn>):\n  F003  replied + resolved   PRRT_x\n  F012  replied, left open   PRRT_y   (deferred)\n\nNot acknowledged (\u003Cn>):\n  F009  ambiguous: PRRT_z fits F009 and F011 equally well\n  F005  no inline comment to answer (CI finding)\n\nPushed \u003Cn> commit(s) to origin\u002F\u003Cbranch> (\u003Csha>..\u003Csha>). CI is re-running.\n```\n\nThe last line states what reached the remote. When Step 6.5 was skipped or\nrefused, say which and name the command that finishes the job:\n\n```\nNot pushed — \u003Creason: --no-push | suite red on partition 2 | branch diverged\nfrom origin\u002F\u003Cbranch>>. \u003Cn> commit(s) on \u003Cbranch>, ahead of origin.\nPush, then `\u002Freview:fix \u003CN> --ack-only`.\n```\n\nThen `python3 \"$LEDGER_PY\" unlock --ledger \"$LEDGER\"`.\n\nMake the deferral count prominent. A run that defers most of its findings has\nmis-routed them, and that has to be visible rather than buried.\n\nMake **Not acknowledged** equally prominent. An unanswered thread is a question\nthe reviewer is still waiting on. On a run that pushed, the only entries left\nare genuine gaps — an ambiguous match, a finding with no inline comment — and\neach needs a human. On a run that did not push, the whole list is one push away\nfrom closing.\n\nAnd `unlock` runs on every exit path, including a refused or rejected push. A\nrun that dies holding the lock blocks the next one for two hours.\n\n`--post-deferred` posts one PR-level comment listing deferrals that had **no\nthread to reply to**. Deferrals with a thread are answered in Step 7.3 instead,\nwhere the reviewer who raised them is actually looking. Off by default.\n\n## Deny-list\n\nFrom `shared\u002Ffix-contract.md`, which is the single copy: never `gh pr merge`,\nnever force-push, never touch the default branch, never `git reset --hard`,\nnever `git rebase`, never `--amend` a pushed commit, never `git stash`, never\n`--no-verify`. A pre-commit hook rejection is a finding to report, not an\nobstacle to route around.\n\nThe one permitted `git push` is Step 6.5's: fast-forward, to the PR's own head\nbranch on `origin`, from a green tree. Every other push — a different branch, a\nfork remote, a force, a push to recover from a rejection — stays forbidden.\n\nOn GitHub, Step 7 is the whole permitted surface: reply in a review thread, and\nresolve a thread whose commit is on the remote. Never resolve a thread this run\ndid not fix, never resolve a deferral, never dismiss a review, never edit or\ndelete someone else's comment, and never post a PR-level comment except the one\n`--post-deferred` asks for.\n\nOne narrow exception, orchestrator-only: `git restore --source=HEAD -- \u003Cexplicit\npaths>` to discard a dead run's in-flight edits after the user confirms (Step\n1.2), restricted to paths the ledger attributes to that finding.\n\n## Notes\n\n- Repo-agnostic by construction. Every command comes from the ledger's `context`\n  block (`test_setup.test_command`, `build_system.commands.*`,\n  `languages.primary`); nothing about a toolchain, a ticket format, or a\n  directory layout is hardcoded.\n- Resumable. State lives in the ledger, outside the repository, so it survives\n  compaction, `git clean`, and worktree switches.\n- Idempotent. A second run with no new findings makes no commits, pushes\n  nothing, replies to nobody twice, and says so.\n- `monitor-pr` drives this skill for live CI failures and review threads; it\n  keeps the polling and the push, and passes `--no-push` to say so.\n  Acknowledgement belongs to whichever of the two can see the commit on the\n  remote — under `monitor-pr` that is always `monitor-pr`, because `--no-push`\n  leaves this skill's commits local.\n","# Fix Review Findings\n\n`\u002Freview:review` names the refactoring that resolves each finding and then posts\nit as a comment. This skill is the consumer that `required_actions` was written\nfor: it reads the durable findings ledger, applies each fix, commits one finding\nat a time, and answers the comment that asked.\n\nIt **edits, commits, pushes the PR's head branch once the suite is green, then\nreplies and resolves.** The push is what makes acknowledgement honest: a thread\nis resolved only when its commit is visible on the remote (Step 7.1), so the run\nthat pushes is the run that can close the loop.\n\nUnder `monitor-pr` the push stays with the caller, which passes `--no-push`.\n\n## Inputs\n\n```\n\u002Freview:fix [\u003Cpr-number|pr-url>] [--repo owner\u002Fname]\n            [--dry-run]            plan and partition only; write nothing\n            [--max-concurrent N]   parallel agents, default 4\n            [--force]              break a stale ledger lock\n            [--ack-only]           skip Steps 1–6.5; only answer review comments\n                                   for findings whose commits are already pushed\n            [--no-ack]             fix, commit, and push; reply to and resolve\n                                   nothing\n            [--post-deferred]      post one PR comment for deferrals that have\n                                   no thread to reply to (off by default)\n            [--no-push]            commit only; leave the branch ahead of origin\n                                   (what monitor-pr passes — it owns the push)\n```\n\nWith no argument, infer the PR from the current branch (`gh pr view --json\nnumber`). There is no `--only` or `--severity` flag: the scope is every open\nfinding. Selection is `status == \"open\"`, which also lets `monitor-pr` process\njust the findings appended since its last tick.\n\n## Step 0: Resolve the ledger\n\n```bash\nPLUGIN_ROOT=\"${CLAUDE_PLUGIN_ROOT:-}\"\nif [ ! -f \"$PLUGIN_ROOT\u002Fscripts\u002Ffindings_ledger.py\" ]; then\n  PLUGIN_ROOT=\"$(cd \"${CLAUDE_SKILL_DIR:-.}\u002F..\u002F..\" 2>\u002Fdev\u002Fnull && pwd)\"\nfi\nLEDGER_PY=\"$PLUGIN_ROOT\u002Fscripts\u002Ffindings_ledger.py\"\n\nPR_NUMBER=\"${1:-$(gh pr view --json number -q .number)}\"\nLEDGER=$(python3 \"$LEDGER_PY\" path \\\n  --repo \"$(gh repo view --json nameWithOwner -q .nameWithOwner)\" --pr \"$PR_NUMBER\")\nRUN_ID=\"fix-$(date -u +%Y%m%dT%H%M%SZ)-$$\"\n```\n\n`LEDGER_PY` is a path, and `python3` stays at the call site. Folding the\ninterpreter into the variable works in bash and silently breaks in zsh, which does\nnot word-split an unquoted expansion.\n\nIf the ledger is missing:\n\n> No findings ledger for PR #\\\u003CN>. Run `\u002Freview:review \u003Curl>` first.\n\n**Do not reconstruct one by scraping the posted review.** The inline-comment\nround-trip loses `remedy`, `refactoring`, `drift_class`, and `code_says` — which\nare exactly the fields that select a strategy.\n\n## Step 1: Lock, reconcile, and establish green\n\n### 1.1 Take the lock\n\n```bash\npython3 \"$LEDGER_PY\" lock --ledger \"$LEDGER\" --run \"$RUN_ID\"   # add --force to break a stale lock\n```\n\nOne `fix` run per ledger, for the same reason `monitor-pr` allows one monitor per\nPR: two runs on one branch means double commits and interleaved edits. A lock\nolder than two hours is stale — report it and offer `--force`.\n\n### 1.2 Reconcile against git, not against the ledger\n\nThe ledger records what a previous run *believed*. Verify it:\n\n- For each `committed` finding, `git cat-file -e \u003Csha>^{commit}`. A sha that no\n  longer exists (branch reset, worktree swap) demotes the finding to `open`.\n- For each `claimed` or `applied` finding, check `git status --porcelain` for its\n  recorded files — a previous run died mid-flight.\n- **Hard stop if the tree is dirty with edits `fix` cannot attribute** to a\n  finding in the ledger. Report the paths and ask. Never `git stash`, never\n  `git restore` a path you did not write.\n  - One exemption: `.claude\u002Fpr-monitor\u002Fpr-*.md`. That is `monitor-pr`'s tracker,\n    written by the caller that just invoked this skill, so it is dirty on every\n    delegated run. Leave it alone and never stage it — a monitor's scratch file\n    has no business in a fix commit.\n- For attributable in-flight edits, show the diff and ask: commit as-is, or\n  discard and redo that finding.\n\n### 1.3 Check the ledger is not stale\n\nCompare `pr.head_sha` against current `HEAD`. If the branch has moved since the\nreview, **every `line` in the ledger is suspect.** Warn, and treat agent rule 5\n(locate by content, not by line) as mandatory rather than advisory.\n\n### 1.4 Establish a green baseline\n\n```bash\n# only when context.test_setup.can_run_safely is true\n\u003Ccontext.test_setup.test_command>\n```\n\n**You cannot refactor on red.** If the baseline fails, nothing afterwards can\ndistinguish breakage this run caused from breakage it inherited. On a red\nbaseline, stop and report — unless the user asks to fix the baseline first, which\nis its own finding.\n\nWhere `can_run_safely` is false there is no baseline and no iteration loop; see\n`shared\u002Ffix-contract.md`.\n\n## Step 2: Route every open finding\n\n```bash\npython3 \"$LEDGER_PY\" show --ledger \"$LEDGER\" --status open --json\n```\n\nRead `references\u002Fstrategy-routing.md` and assign each finding exactly one\nstrategy: **S0** formatter, **S1** refactor, **S2** repair, **S3** prose, **S4**\ncomment, **S5** regenerate, **S6** defer. First match wins.\n\nEvery finding gets a strategy. Nothing is dropped. S6 is the honest-deferral\npath, not a filter — if most findings land there, the routing is too lax and the\nreport must make that visible.\n\n### Findings with no `path`\n\nCommon and deliberate: architectural findings and CI-ingested findings carry no\nlocation. Do not guess a file and do not drop the finding.\n\n1. If the `body` or `remedy` names a symbol, `rg` for it and derive the file set.\n2. Confirm the symptom at the candidate before editing.\n3. Zero candidates, or more than three → **S6** `ambiguous-remedy`.\n\n**Never dispatch an agent with neither a file anchor nor a symbol.** That is how\nthis skill would rewrite half the repository.\n\n## Step 3: Order the work\n\nPer `references\u002Fstrategy-routing.md`:\n\n1. **S5 first** for any path S1\u002FS3\u002FS4 also touch — regeneration overwrites hand\n   edits.\n2. **S2** — correctness and security.\n3. **S3, S4, S0.**\n4. **S1 last** — where the suite could not run these are unverified, and at the\n   tip of the branch an unwind is one `git revert`.\n5. Within a sequential group, **descending line number.**\n\n## Step 4: Dispatch\n\nPartition open findings by **file set**. A rename's file set is every file that\nreferences the symbol — find them with LSP `findReferences`, or `rg -w`, before\npartitioning, because the rename tool rewrites all of them. Disjoint sets get one agent each, all\ndispatched in a single message. Findings sharing any file collapse into one\nagent that handles them sequentially, re-reading the file between findings. Cap\nconcurrency at `--max-concurrent` (default 4).\n\n**Agents edit; this skill commits and verifies.** There is one `.git\u002Findex` per\nworktree, so concurrent commits from several agents interleave and a commit ends\nup carrying another agent's half-finished work; and a test suite with shared\nfixtures — a database it resets, a port it binds — corrupts the run when four\nagents start it at once. The agent prompts forbid every `.git`-writing command\nand every suite invocation; the orchestrator stages the reported paths and runs\nthe suite itself.\n\nThe partition is therefore both the concurrency unit and the verification unit.\n\nPin **both** `subagent_type` and `model` on every dispatch — an omitted `model`\ninherits the session model and collapses the intended split:\n\n| Strategy | `subagent_type` | `model` |\n|---|---|---|\n| S1 design-level | `general-purpose` | `opus` |\n| S1 mechanical | `general-purpose` | `sonnet` |\n| S2 | `general-purpose` | `opus` |\n| S3, S4 | `general-purpose` | `haiku` |\n\nUse the verbatim prompts in `references\u002Fsubagent-prompts.md`, with the catalog\nrow, the idiom rows, the tool rows, and the rules **inlined** — a subagent runs\nin the reviewed repo and cannot resolve a path inside this skill.\n\n### Deterministic first\n\nRead `references\u002Frefactoring-tools.md`. A move a tool covers is applied by the\ntool, not rewritten by a model: the prompt carries the tool row, and the agent\nreports the command in `tool_applied`. A tool the table marks\norchestrator-only — it rewrites the whole project — you run yourself, like\nS0, before dispatching the partition it touches, and commit it as that\nfinding.\n\nS0, S5, and S6 need no agent: run the formatter, run the regeneration recipe, or\nrecord the deferral.\n\n### S5 — resolve the regeneration command\n\nIn order: a command named literally in `remedy`; an entry in\n`context.build_system.commands`; a task-runner recipe whose name matches the\nartifact. None resolves → **S6** `no-regen-command`. Commit only the generator's\noutput, and prove idempotence by running it twice (`git diff --exit-code` clean).\n\n### S0 — run the repo's formatter, not a reported diff\n\nA formatting finding may arrive with a diff attached — from a linter the security\nscan ran, or from a CI job `monitor-pr` ingested. Treat it as evidence that drift\nexists, discard the diff, and run `context.build_system.commands.lint` autofix:\nthe tool is authoritative and a review-time diff is stale by now. One batch\ncommit. No lint command → **S6** `no-lint-command`.\n\n## Step 5: Commit per finding, verify per partition\n\nAgents never run the suite. As each agent reports `applied`, commit its finding\nas reported. After a partition's last commit, run\n`context.test_setup.test_command` once.\n\nVerifying per finding costs one suite run per finding; verifying per partition\ncosts one per partition. On a suite that takes minutes — which is most of them —\nthat difference is the dominant term in how long a fix round takes, and the\ninformation it buys is recoverable: a red partition is a handful of commits, and\n`git bisect` finds the guilty one in a couple of runs.\n\n**When the partition goes red:**\n\n1. `git bisect` across the partition's commits, using the suite as the test.\n2. Re-dispatch that finding with the failure output attached. Adjust the change,\n   never the test's expectations unless the finding is about the test.\n   **Do not revert. Do not commit a WIP. Do not defer.**\n3. Re-run the suite. Repeat until green.\n\n**Never leave the branch red.** The claim each commit makes — that the branch is\ngreen at that commit — is written at commit time and made true at partition\ntime. A fix round that ends red has not ended.\n\n**When the loop stalls**, escalate rather than abandon: if the failure output\nstops changing, re-dispatch on a stronger model with the full failure history. If\nthat also stalls, **surface the finding to the user as a live question and pause\non it.** Never silently defer it and never move on as though it were handled.\n\nNo runnable suite (`context.test_setup.can_run_safely` is false) → commit as\nreported and say so in the Step 8 report; CI is the verification.\n\nCommit per `shared\u002Ffix-contract.md` — headline, body, preservation line, trailer\n— then record it:\n\n```bash\npython3 \"$LEDGER_PY\" resolve --ledger \"$LEDGER\" --id \u003CID> --sha \"$(git rev-parse HEAD)\" --strategy S1\n```\n\n## Step 6: Record deferrals\n\n```bash\npython3 \"$LEDGER_PY\" defer --ledger \"$LEDGER\" --id \u003CID> --reason \u003Creason> \\\n  --decision \"\u003Cthe one-sentence question a human must answer>\" --run \"$RUN_ID\"\n```\n\nA deferred finding gets **no edit and no commit**, keeps its original severity,\nand never gets a bare `TODO` as a consolation prize. This skill does not file\ntickets.\n\n## Step 6.5: Push the head branch\n\nEvery commit from Step 5 sits on a branch that is green and verified. Push it —\nthat is what lets Step 7 resolve the threads in this same run instead of handing\nthe user a second command.\n\nSkip entirely under `--no-push`, `--dry-run`, or `--ack-only`, and skip when\nStep 5 made no commits.\n\n### 6.5.1 Gate on green\n\nPush only when **all** of these hold:\n\n- Every partition's final suite run passed. A partition still red, or a finding\n  Step 5 paused on, blocks the push for the whole run — do not push \"the green\n  part\". The branch tip is what CI and reviewers see.\n- No finding is in `claimed` or `applied` state. Everything is `committed` or\n  `deferred`.\n- The tree is clean: `git status --porcelain` is empty apart from the\n  `.claude\u002Fpr-monitor\u002Fpr-*.md` exemption from Step 1.2.\n\nDeferrals do **not** block the push. A deferral produced no commit, so the\npushed branch is exactly the verified work.\n\nWhere `context.test_setup.can_run_safely` is false there is no local green\nsignal and CI is the verification — push, and mark every commit `Unverified` in\nthe Step 8 report. Withholding the push there withholds the only signal\navailable.\n\n### 6.5.2 Push\n\n```bash\nHEAD_BRANCH=$(gh pr view \"$PR_NUMBER\" --json headRefName -q .headRefName)\nCROSS_REPO=$(gh pr view \"$PR_NUMBER\" --json isCrossRepository -q .isCrossRepository)\n\ntest \"$(git rev-parse --abbrev-ref HEAD)\" = \"$HEAD_BRANCH\"   # refuse otherwise\ntest \"$CROSS_REPO\" = \"false\"                                  # refuse otherwise\n\ngit push origin \"HEAD:refs\u002Fheads\u002F$HEAD_BRANCH\"\n```\n\nChecked out somewhere other than the PR's head branch, or the PR comes from a\nfork whose head repo is not `origin` → **do not push.** Report it and let Step 7\nfall back to acknowledging nothing. Guessing a remote pushes someone's work to\nthe wrong place.\n\n**A rejected push is reported, never forced around.** `non-fast-forward` means\nthe remote moved while this run was working. Do not force-push, do not rebase,\ndo not pull — all three are on the deny-list and all three can discard a\ncollaborator's commit. Say the branch diverged, name the commits that stayed\nlocal, and stop; the user re-runs after reconciling.\n\nPush once, after the last partition. Pushing per commit re-triggers CI on every\none of them, which costs a full pipeline per finding and leaves a trail of runs\nnobody reads.\n\n## Step 7: Answer the review comments\n\n`\u002Freview:review` asked each of these questions as an inline comment. A finding\nresolved in silence leaves the reviewer with an open thread and no idea it was\nhandled, so close the loop: reply to the comment that raised it, and resolve the\nthread when the work is actually on the PR.\n\n### 7.1 Only acknowledge what the PR can show\n\n```bash\ngit fetch origin \"$HEAD_BRANCH\" --quiet\ngit merge-base --is-ancestor \u003Csha> \"origin\u002F$HEAD_BRANCH\"   # per committed finding\n```\n\n**A thread is resolved only when its commit is on the remote.** Run the check\neven on a run that just pushed — it is what catches a partial push, a rejected\npush, and a sha the ledger recorded from an earlier run that no longer exists.\nResolving a thread for a commit that exists only on this machine tells a\nreviewer to go look at something they cannot see.\n\nAfter a successful Step 6.5 the check passes for everything this run committed,\nand the loop closes in one command. When Step 6.5 was skipped or refused, it\nfails for those commits — that is the correct outcome, not a bug — and the work\nfinishes later:\n\n```\n\u002Freview:fix 123 --ack-only      # acknowledge pushed work only\n```\n\nUnder `monitor-pr` this resolves itself — it passes `--no-push` and calls `fix`\nbefore pushing, so nothing is acknowledged here, and `monitor-pr` acknowledges\nin its Phase 3 after its push. One acknowledgement per thread, from whoever can\nsee the commit.\n\n### 7.2 Map findings to threads\n\n```bash\ngh api graphql -f query='\nquery($owner:String!,$repo:String!,$pr:Int!){\n  repository(owner:$owner,name:$repo){\n    pullRequest(number:$pr){\n      reviewThreads(first:100){\n        nodes{\n          id isResolved isOutdated\n          comments(first:1){ nodes{ databaseId author{login} path line originalLine body } }\n        }\n      }\n    }\n  }\n}' -F owner=\"$OWNER\" -F repo=\"$NAME\" -F pr=\"$PR_NUMBER\" \\\n  | python3 \"$PLUGIN_ROOT\u002Fscripts\u002Fpr_threads.py\" match --ledger \"$LEDGER\" --threads - --json\n```\n\nThe script attributes a thread to `\u002Freview:review` by the Conventional Comments\nmarker its comments lead with (`**issue (blocking):** …`), so it works whichever\naccount posted the review. It pairs a thread with a finding on same-file plus\nsame-line or same-title, and returns `matched`, `ambiguous`,\n`unmatched_findings`, `unmatched_threads`.\n\n**Act only on `matched`.** An `ambiguous` entry means two findings fit one thread\nequally well — report it and leave the thread alone. A wrong match answers a\nreviewer's question with the resolution of a different question, which is worse\nthan leaving it open.\n\nA `low` confidence match is still a match, but say so in the report so it can be\nspot-checked. Findings with `reason: no-path` never had an inline comment — CI\nfailures and architectural findings — and are not a gap.\n\n### 7.3 Reply, then resolve\n\nPer `matched` entry, reply first. If the reply fails, do not resolve: a resolved\nthread with no explanation is worse than an unresolved one.\n\n```bash\ngh api graphql -f query='mutation($t:ID!,$b:String!){\n  addPullRequestReviewThreadReply(input:{pullRequestReviewThreadId:$t, body:$b}){\n    comment{ url }\n  }\n}' -F t=\"\u003Cthread_id>\" -F b=\"\u003Cbody>\"\n```\n\n**Committed** — one line of what changed, and the sha:\n\n> Fixed in `\u003Csha>` — extracted the retry loop into `Transport::send_with_retry`\n> so the backoff lives with the transport. `\u003Ctest_command>` passes.\n\n**Deferred** — the question, not an apology. The thread stays open:\n\n> Not fixed — this needs a decision first: outbox table, CDC, or accept the\n> window? Each changes the failure mode, so picking one here would be a guess.\n\nThen resolve, for `action: reply+resolve` only:\n\n```bash\ngh api graphql -f query='mutation($id:ID!){\n  resolveReviewThread(input:{threadId:$id}){ thread{ isResolved } }\n}' -F id=\"\u003Cthread_id>\"\n```\n\nNever resolve a `reply` action. A deferral is the one finding that most needs a\nhuman to see it, and resolving it hides it.\n\nWrite the reply from the diff and the finding, not from this run's shape. No\nstrategy codes, no finding IDs, no \"as requested\" — the reviewer is reading a PR,\nnot this skill's output. `F012` means nothing to them.\n\n### 7.4 Record it\n\n```bash\npython3 \"$LEDGER_PY\" ack --ledger \"$LEDGER\" --id \u003CID> --thread \u003Cthread_id> \\\n  --action reply+resolve --url \"\u003Creply url>\" --run \"$RUN_ID\"\n```\n\nIdempotent per `(finding, thread)`, and the matcher skips anything already\nrecorded. That is what stops a re-run — or a `monitor-pr` tick after a manual\n`fix` — from replying to the same reviewer twice.\n\n`--no-ack` skips this whole step.\n\n## Step 8: Report\n\n```\nFix run complete — PR #\u003CN>\n\nResolved (\u003Cn>):\n  \u003Csha>  refactor: move rate limiting into the transport   F003\n  \u003Csha>  docs: tighten the auth overview                   F007\n\nDeferred (\u003Cn>):\n  F012 · src\u002Foutbox.rs:88 · dual-write with no outbox —\n         decision: outbox table, CDC, or accept the window?\n\nUnverified (\u003Cn>):   # only when the suite could not run\n  \u003Csha>  F005 — suite not run (\u003Csafety_notes>); revert with `git revert \u003Csha>`\n\nAcknowledged (\u003Cn>):\n  F003  replied + resolved   PRRT_x\n  F012  replied, left open   PRRT_y   (deferred)\n\nNot acknowledged (\u003Cn>):\n  F009  ambiguous: PRRT_z fits F009 and F011 equally well\n  F005  no inline comment to answer (CI finding)\n\nPushed \u003Cn> commit(s) to origin\u002F\u003Cbranch> (\u003Csha>..\u003Csha>). CI is re-running.\n```\n\nThe last line states what reached the remote. When Step 6.5 was skipped or\nrefused, say which and name the command that finishes the job:\n\n```\nNot pushed — \u003Creason: --no-push | suite red on partition 2 | branch diverged\nfrom origin\u002F\u003Cbranch>>. \u003Cn> commit(s) on \u003Cbranch>, ahead of origin.\nPush, then `\u002Freview:fix \u003CN> --ack-only`.\n```\n\nThen `python3 \"$LEDGER_PY\" unlock --ledger \"$LEDGER\"`.\n\nMake the deferral count prominent. A run that defers most of its findings has\nmis-routed them, and that has to be visible rather than buried.\n\nMake **Not acknowledged** equally prominent. An unanswered thread is a question\nthe reviewer is still waiting on. On a run that pushed, the only entries left\nare genuine gaps — an ambiguous match, a finding with no inline comment — and\neach needs a human. On a run that did not push, the whole list is one push away\nfrom closing.\n\nAnd `unlock` runs on every exit path, including a refused or rejected push. A\nrun that dies holding the lock blocks the next one for two hours.\n\n`--post-deferred` posts one PR-level comment listing deferrals that had **no\nthread to reply to**. Deferrals with a thread are answered in Step 7.3 instead,\nwhere the reviewer who raised them is actually looking. Off by default.\n\n## Deny-list\n\nFrom `shared\u002Ffix-contract.md`, which is the single copy: never `gh pr merge`,\nnever force-push, never touch the default branch, never `git reset --hard`,\nnever `git rebase`, never `--amend` a pushed commit, never `git stash`, never\n`--no-verify`. A pre-commit hook rejection is a finding to report, not an\nobstacle to route around.\n\nThe one permitted `git push` is Step 6.5's: fast-forward, to the PR's own head\nbranch on `origin`, from a green tree. Every other push — a different branch, a\nfork remote, a force, a push to recover from a rejection — stays forbidden.\n\nOn GitHub, Step 7 is the whole permitted surface: reply in a review thread, and\nresolve a thread whose commit is on the remote. Never resolve a thread this run\ndid not fix, never resolve a deferral, never dismiss a review, never edit or\ndelete someone else's comment, and never post a PR-level comment except the one\n`--post-deferred` asks for.\n\nOne narrow exception, orchestrator-only: `git restore --source=HEAD -- \u003Cexplicit\npaths>` to discard a dead run's in-flight edits after the user confirms (Step\n1.2), restricted to paths the ledger attributes to that finding.\n\n## Notes\n\n- Repo-agnostic by construction. Every command comes from the ledger's `context`\n  block (`test_setup.test_command`, `build_system.commands.*`,\n  `languages.primary`); nothing about a toolchain, a ticket format, or a\n  directory layout is hardcoded.\n- Resumable. State lives in the ledger, outside the repository, so it survives\n  compaction, `git clean`, and worktree switches.\n- Idempotent. A second run with no new findings makes no commits, pushes\n  nothing, replies to nobody twice, and says so.\n- `monitor-pr` drives this skill for live CI failures and review threads; it\n  keeps the polling and the push, and passes `--no-push` to say so.\n  Acknowledgement belongs to whichever of the two can see the commit on the\n  remote — under `monitor-pr` that is always `monitor-pr`, because `--no-push`\n  leaves this skill's commits local.\n",{},"main","a3f19bcc313571bff79dca391a83f9c4d38bb335","review\u002Fskills\u002Ffix",0,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum","https:\u002F\u002Fgithub.com\u002Fozzycodes2","Software Engineer | Engineering Manager | AI whisperer | ☕Enjoyer","2026-11-14 04:50:11",{"source":7,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add ozzycodes2\u002Fclaudellum --skill fix",{"tool":37,"command":38},"bunx","bunx skills add ozzycodes2\u002Fclaudellum --skill fix",{"tool":40,"command":41},"pnpm","pnpm dlx skills add ozzycodes2\u002Fclaudellum --skill fix","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002FSKILL.md",true,[45,51,55,59,63,67,71],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"evals\u002Fevals.json","REFERENCE","参考文档",9110,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Fevals\u002Fevals.json",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Flanguage-idioms.md",5811,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Freferences\u002Flanguage-idioms.md",{"path":56,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":57,"sourceHtmlUrl":58},"references\u002Fprose-rules.md",6801,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Freferences\u002Fprose-rules.md",{"path":60,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Frefactoring-catalog.md",5459,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Freferences\u002Frefactoring-catalog.md",{"path":64,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Frefactoring-tools.md",4544,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Freferences\u002Frefactoring-tools.md",{"path":68,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":69,"sourceHtmlUrl":70},"references\u002Fstrategy-routing.md",6460,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Freferences\u002Fstrategy-routing.md",{"path":72,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":73,"sourceHtmlUrl":74},"references\u002Fsubagent-prompts.md",9499,"https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Freview\u002Fskills\u002Ffix\u002Freferences\u002Fsubagent-prompts.md",[76,83,90,97,104,111],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":18},"01M7ASQB3CTKMKDNAXD9DCF1YC","writing-design-docs","Use when the user wants to write a problem-statement design doc, says \"let's plan a feature properly\" \u002F \"write a design doc\" \u002F \"I'm thinking about how to do X\" for a non-trivial X, or before committing to a non-obvious technical direction","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Ftechnical-writing\u002Fskills\u002Fwriting-design-docs\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":18},"01M7ASQ6JD4QRV5ZKK2QME401Y","write-doc","Use when the user asks to write or draft a README, ADR, runbook, tutorial, how-to, explanation, or reference doc, or a caller passes \"type ADR\" or \"type README\". Picks the type, fills a template, adds diagrams. RFCs and design docs belong to writing-design-docs.","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Ftechnical-writing\u002Fskills\u002Fwrite-doc\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":18},"01M7ASQ1VWNVP1RE7M5D4G2QMG","review-doc","Use when the user asks to review, critique, score, grade, or \"how good is this doc\" for a Markdown doc or pasted text. Read-only. Scores it on the 20 Practical Prose dimensions and emits a findings table. Does not edit; fix-doc applies the findings next.","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Ftechnical-writing\u002Fskills\u002Freview-doc\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":18},"01M7ASPWST208JZNR7HF7J6KMV","fix-doc","Use when the user asks to fix, edit for prose quality, polish, tighten, humanize, or de-slop a Markdown doc or pasted text, or says \"make this read less AI\". Applies review-doc findings, generates flagged diagrams, lists substantive issues for the author, then runs the AI-tell pass last. Modifies the target.","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Ftechnical-writing\u002Fskills\u002Ffix-doc\u002FSKILL.md",[],[],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":18},"01M7ASPS6Y9ZGK9KCDHNPDR75A","create-slides","Use when the user asks for a slide deck, presentation, talk, or slides from a document, report, PR, notes, or topic. Builds one self-contained Machinify-branded HTML deck in a minimalist style, with headline-plus-visual slides, speaker notes, and every diagram drawn by architecture-design:create-diagram.","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Ftechnical-writing\u002Fskills\u002Fcreate-slides\u002FSKILL.md",[],[],{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"publishedAt":18},"01M7ASPAKFY045KHR92Y4QFTWV","start-session","Open a Claude Code working session in a fresh, isolated git worktree, so the user can start a task with a single \u002Fstart-session \u003Cname>. Use at the START of a session when the user says \"start session\", \"start a worktree\", \"spin up a worktree for ABC-123\", \"begin work on \u003Cticket>\", or passes a name\u002Fticket to work under. Creates and ENTERS a worktree via the EnterWorktree tool (never `git worktree add`) under .claude\u002Fworktrees\u002F, named exactly after the argument, on a local branch `worktree-\u003Cargument>`. Reports what the worktree inherited (`.worktreeinclude` copies, symlinked node_modules) and whether its `.env` still points at the main checkout's stack. If the argument is a JIRA key (e.g. ABC-123, PROJ-1505), also loads that story with the Atlassian CLI (`acli`) when available and summarizes the implementation ask. Also renames the session to the argument (the same effect as `\u002Frename \u003CNAME>`) so it's easy to find in `\u002Fresume`. Then stops and waits — it sets up the workspace and context, it does not implement.","https:\u002F\u002Fgithub.com\u002Fozzycodes2\u002Fclaudellum\u002Fblob\u002Fmain\u002Fsession-mgmt\u002Fskills\u002Fstart-session\u002FSKILL.md",[],[]]