[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77KB6EQTWDW0MNM3TXPP0BC":3,"skill-same:01M77KB6EQTWDW0MNM3TXPP0BC":99},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":29,"commitSha":30,"skillDirectory":31,"forksCount":32,"repositoryHtmlUrl":33,"repositoryHomepageUrl":6,"ownerHtmlUrl":34,"ownerBlogUrl":35,"ownerDescription":36,"updatedAt":37,"install":38,"resources":50},"01M77KB6EQTWDW0MNM3TXPP0BC","reduce",null,"Iteratively reduce coupling at any altitude (documents, code modules, applications, repositories): scan for change-transmitting dependencies typed against a coupling model, verify each finding, apply a budgeted batch of safe behavior-preserving reductions, and ledger structural candidates for design routing so repeated runs continue where the last stopped; after a value changes, find and classify every site stating it. Use when: 'reduce coupling', 'decouple', 'loosen coupling', 'too tightly coupled', 'high cohesion low coupling', 'break this dependency', 'dependency injection pass', 'externalize this config', 'connascence', 'coupling scan', 'these files always change together', 'stop copying between repos', 'I changed this value everywhere'. Skip when: reviewing a diff before merge (review tools), designing one chosen boundary (\u002Farchitecture:improve), general structural tidyings with no coupling focus (\u002Fcode-tidying:tidy), or a docs noise or dedup pass (docs-hygiene).","melodic-software\u002Fclaude-code-plugins","melodic-software","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F58273638?v=4","01M77JVQTXNGKCE4MDFY2BC936","Melodic Software","01M77JVQVN9QDWKAH6W03ANP76",30,"MIT","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Iteratively reduce coupling at any altitude (documents, code modules, applications, repositories): scan for change-transmitting dependencies typed against a coupling model, verify each finding, apply a budgeted batch of safe behavior-preserving reductions, and ledger structural candidates for design routing so repeated runs continue where the last stopped; after a value changes, find and classify every site stating it. Use when: 'reduce coupling', 'decouple', 'loosen coupling', 'too tightly coupled', 'high cohesion low coupling', 'break this dependency', 'dependency injection pass', 'externalize this config', 'connascence', 'coupling scan', 'these files always change together', 'stop copying between repos', 'I changed this value everywhere'. Skip when: reviewing a diff before merge (review tools), designing one chosen boundary (\u002Farchitecture:improve), general structural tidyings with no coupling focus (\u002Fcode-tidying:tidy), or a docs noise or dedup pass (docs-hygiene).\"\nargument-hint: \"[\u003Cscope>|dry-run [\u003Cscope>]|change [apply] \u003Cold> \u003Cnew> [\u003Cscope>]|status|help]\"\nuser-invocable: true\ndisable-model-invocation: false\nmetadata:\n  workflow-stage: anytime\n  summary: Scan for change-transmitting coupling, apply safe reductions in a budgeted batch, route the rest\n---\n\n## Repository context. Gather first\n\nCollect these with **individual** Bash calls, one command per call, never combined into a single\ninvocation:\n\n- Current branch, `git branch --show-current`\n- Recent commits, `git log --oneline -10`\n- Working tree status (empty = clean), `git status --porcelain | head -10`\n\nThe pipe is the bound and belongs in the command. A read-time cap (\"read only the first 10 entries\")\ncomes too late for whatever the command has already returned inline. What a result returns inline,\nand when it arrives as a saved file instead, is [Tools reference § Output\nlimits](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Ftools-reference#output-limits).\n\nTreat a failure (not a repository, git unavailable) as an unknown value and carry on. Keep these as\nseparate body Bash calls rather than pre-compute lines ([gather-block convention](https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fdocs\u002Fconventions\u002Fgather-block\u002FREADME.md#the-pre-compute-block-runs-as-one-shell-invocation),\n\"The pre-compute block runs as one shell invocation\").\n\n## Variables\n\nArguments: `$ARGUMENTS`\n\n## Purpose\n\nAnswer one question, repeatedly: **which dependency in this repository transmits the most\nunnecessary change, and what is the smallest mechanism that stops the transmission?** Each\ninvocation is one pass. Scan, verify, reduce what is safely reducible within a scope budget,\nsurface what is not, and record everything in a durable ledger so the next invocation resumes\ninstead of restarting. Coupling goes down monotonically across runs; no single run tries to\nfinish the job.\n\nTwo lanes, split by the nature of the finding, never by convenience:\n\n- **Apply lane.** Mechanical, contained, behavior-preserving reductions (weaken a\n  connascence form, de-duplicate a stated fact into a pointer, inject a hard-wired volatile\n  collaborator, name a magic value). Applied this run, verified, shipped as one\n  structure-only change set.\n- **Route lane.** Cross-file remediation and architectural judgment (move a boundary,\n  introduce a seam, split a shared database, merge two repos' duplicated logic). Coupling\n  findings of this shape exist to inform a human: they are surfaced, ranked, and routed,\n  never auto-applied.\n\nThis is not a diff reviewer, not a designer of one chosen boundary, and not a general tidy\npass. See \"What this skill does NOT do\".\n\n## Actions\n\n| Argument | Action |\n|----------|--------|\n| *(empty)* | Full pass over an inferred scope: scan → verify → apply lane → route lane → ledger |\n| `\u003Cscope>` | Same pass narrowed to a path, module, or altitude keyword (`docs`, `code`, `app`, `repo`) |\n| `dry-run [\u003Cscope>]` | Scan, verify, rank, and ledger only. Do NOT edit, branch, push, or file tracker items |\n| `change \u003Cold> \u003Cnew> [\u003Cscope>]` | Change mode, read-only: find every site that states `\u003Cold>`, classify each, and propose what happens to it. No edits |\n| `change apply \u003Cold> \u003Cnew> [\u003Cscope>]` | Change mode with edits: the same inventory, then, once the human confirms the site list, change the setup sites and convert them to a reference form |\n| `status` | Read the ledger and report open candidates, applied reductions, and what the next run should take |\n| `help` | Print this table with one worked example per action |\n\n## The model\n\nHub-and-spoke; read the spoke before the phase that needs it:\n\n- [`reference\u002Fcoupling-model.md`](reference\u002Fcoupling-model.md). What counts as coupling, the\n  strength ladder, connascence axes, volatility weighting, per-altitude mechanisms, and the\n  not-a-finding list. Read before scanning; findings are typed against it.\n- [`reference\u002Fremediations.md`](reference\u002Fremediations.md). Mechanism per finding kind, each\n  with its over-abstraction counterweight and the sequencing rule (smallest mechanism first).\n  Read before applying or routing anything.\n- [`reference\u002Fledger.md`](reference\u002Fledger.md). Ledger entry schema, status lifecycle, and\n  re-run semantics. Read at phase A and phase H.\n- [`reference\u002Fchange-mode.md`](reference\u002Fchange-mode.md). The `change` action: how the inventory\n  finds every form of a value, the site classes and what happens to each, reference forms, the\n  apply sequence, and its pitfalls. Read before any `change` run.\n\n## Workflow\n\nPhases run in order; each gate is hard. `dry-run` stops after D (ledger write included).\n\n**A. Orient.** Resolve scope from the argument, else infer from the conversation, else pick\nthe hottest area by commit frequency. The ledger path is the memory tier's constant slug `coupling`, default\n`.work\u002Fcoupling\u002Fcoupling-ledger.md` (`.work\u002F` unless the repo's instructions declare another working-docs root),\nnever committed, one ledger per repo regardless of scope. The slug is constant because coupling\nreduction is repo-scoped and spans many scopes and short-lived branches; a scope- or\nbranch-derived slug would fragment the one ledger successive runs must resume. On the session's\nfirst write, verify the memory root contains a `.gitignore` with `*`, creating it (announced) when\nabsent, and create the slice directory when absent. Create or resume the ledger. Discover the consuming repo's\nown review\u002Fengineering conventions (a review-criteria file such as `REVIEW.md`, a\n`docs\u002Fconventions\u002F` or standards directory, CLAUDE.md rules) and align finding vocabulary and\nseverity with them; the bundled model is the fallback, never an override of the consumer's\nown standards.\n\n**B. Scan.** Fan out fresh-context Explore subagents over the scope, each briefed with the\ncoupling model and returning findings in ledger-entry form (edge, kind, mechanism, strength,\ndegree, locality, evidence). Each is done when its slice of the scope has been read; it returns\nearly with what it has when a finding needs a file outside that slice. In parallel, mine co-change evidence from version-control\nhistory: file pairs that repeatedly change in the same commits without a declared dependency\nare coupled through a channel the import graph cannot see, and they outrank most statically\nvisible findings.\n\nBeside co-change, measure collision hotspots when the repository is hosted on GitHub: run\n`bash \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fcollision-hotspots.sh\" --prs \u003Cn>` from the repository root\n(add `--repo \u003Cowner>\u002F\u003Crepo>` when `gh`'s default repository is not the one that holds the pull\nrequests, `--exclude \u003Cglob>` for more files that change on every release, `--max-pairs \u003Cm>` to\ncap the replays, and `--dry-run` to see the pairs before fetching anything). The script lists\npull requests from that one repository and fetches their heads from `origin` when `origin` names\nit, otherwise from the URL `gh` reports for it. Size `--prs` by the\ndays of history wanted: count the merges into the default branch over the last few days with\n`git log --first-parent --since='\u003Cn> days ago' --format=%h \u003Cdefault-branch>`, one line per merge,\nand multiply the daily rate by the days. The script\npairs pull requests that were open at the same time and share a file, replays each pair with\n`git merge-tree`, and prints `conflicts\u003CTAB>pairs\u003CTAB>path` for every file at least 3 pairs\ntouched; read each ranked file as a candidate typed against the model's collision hotspot entry\nand verify it in phase C like any other finding. Carry its `gap:` lines into the report: a pull\nrequest whose file list stopped at the host's limit is undercounted, and a head that could not\nbe fetched was not replayed. The `known bump hotspots` line names release files that conflict\nfor version bumps, not design; report it once and do not ledger those files. When the script\nexits 2 (no `gh`, no `jq`, a `git` without `merge-tree --write-tree`, no host data), rank by\nco-change alone and record `gap: collision hotspots not measured: \u003Cthe script's message>` in\nthe report.\n\n- Pointer: the 100-file cap per pull request comes from\n  \u003Chttps:\u002F\u002Fdocs.github.com\u002Fen\u002Fgraphql\u002Foverview\u002Frate-limits-and-query-limits-for-the-graphql-api#node-limit>;\n  the exit codes the script reads are in \u003Chttps:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-merge-tree#OUTPUT> and\n  \u003Chttps:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-merge-tree#_exit_status>.\n- As of: 2026-11-10.\n- Recheck trigger: the forge's pull-request file list returns more than 100 files for one pull request,\n  or `git merge-tree --write-tree` changes its exit status meanings.\n\nThen check the ranked files for a second list. Ranked paths come from pull requests and are\nuntrusted: never type one into a command. Pipe the ranking straight in from the repository root,\n`bash \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fcollision-hotspots.sh\" --prs \u003Cn> | awk -F'\\t' 'NF == 3 && \\$1 ~\n\u002F^[0-9]+$\u002F { print \\$3 }' | python3 \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fderivable-list.py\" --paths-from -`\n(same flags as the ranking run), or write the paths one per line to a file with the Write tool\nand pass `--paths-from \u003Cfile>`; with no ranking, write the files co-change mining ranked highest\nthe same way. It prints `candidate`\nfor a file holding a list of at least 5 entries where at least 80% of the entries name a file or\nfolder git tracks, `below` or `no-list` otherwise, an `unnamed` row for each candidate entry that\nnames nothing (a stale entry is how such a list drifts), and `skip` for a path it did not read.\nBoth numbers are judgment values, not settings: 80% leaves room for the stale entries a drifting\nlist carries while ruling out a list whose entries mostly hold data the tree does not, and below 5\nentries a list matches paths by chance and costs little to keep by hand. Verify each candidate in\nphase C (a list a generator already writes is already derived), then report it with the\nDesign-It-Twice comparison in [`reference\u002Fremediations.md`](reference\u002Fremediations.md) \"Second\nlists\". A second list is always a route-lane ledger entry, report-only, never part of the apply\nbatch.\n\nBrief the scan agents to list duplicate writers too: one durable key, path or table that more\nthan one owner writes (the model's common-coupling entry). Use `value-sites.py find` for a literal\nkey. A confirmed duplicate writer always goes to the route lane as a structural ledger entry and\nnever into the apply batch; phase H hands off its ownership rule per the remediation catalog's\n\"Duplicate writers\" section.\n\n**C. Verify (hard gate).** Scan agents have a demonstrated error rate. Reproduce every\nfinding against the actual artifacts before it reaches the ledger or the user. Confirm the\nedge exists, the mechanism is what the scan claims, and the depended-on side actually changes\n(volatility from history, not vibes). Drop or downgrade what does not reproduce, and record\nthat the ledger reflects verified state, not raw scan output.\n\n**D. Partition and rank.** Every verified finding gets a lane. Apply lane requires ALL of:\nmechanical, contained in scope, behavior-preserving, and reversible by revert. Anything\ncross-file in remediation or architectural in judgment is route lane. When in doubt, route.\nRank by `strength × degree × distance × volatility`; a weak-but-everywhere coupling on a hot\npath outranks a strong-but-local one in cold code. Write the full ranked set to the ledger.\n\n**E. Apply (apply lane only, scope-budgeted).** Work on a short-lived branch created from\nthe repository's default branch. Resolve that branch (remote HEAD or the repo's own\nconvention), never assume its name, and never base the batch on whatever feature branch the\nsession happens to be on: inherited unrelated commits would break the structure-only\ninvariant. Never commit directly on the default branch. Before editing any target, require\nit clean in `git status --porcelain`; a target carrying pre-existing local modifications\ndefers its finding with the reason recorded. Foreign edits are never mixed into the batch.\nBudget per run: the target and hard cap in\n[`${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fpr-scope-budget.md`](..\u002F..\u002Freference\u002Fpr-scope-budget.md); overflow stays\n`proposed` in the ledger for the next run. A confirmed `change apply` site set is the one\nexception (see the Apply sequence in [`reference\u002Fchange-mode.md`](reference\u002Fchange-mode.md)). Use the Edit tool; one atomic commit per logical\nreduction; stage listed paths only and inspect the staged diff before each commit. Never\ntouch CI workflow files, hook or settings files, lint configs, database migrations, or\nany published contract surface (API shapes, message schemas, tool schemas). Those are route\nlane by definition.\n\n**F. Verify the batch.** Invoke `\u002Ftoolchain:check` via the Skill tool for the affected\necosystems when the `toolchain` plugin is installed; otherwise run the consuming project's\nown build, test, and lint commands (its CLAUDE.md or rules may name them). A reduction that\nbreaks a test was secretly behavioral: revert that reduction, reclassify it as route lane,\nand continue. Never ship red.\n\n**G. Ship.** Present the verified diff, or when the session should open a pull request,\ninvoke `\u002Fsource-control:pull-request create` via the Skill tool when that plugin is installed; otherwise use the\nrepo's own PR convention with a body listing each reduction as `edge → mechanism → why safe`.\nOne coupling pass per PR; structure-only, no behavioral riders. A human merges. This skill\nnever auto-merges.\n\n**H. Ledger update and report.** Update statuses (`applied`, `deferred`, `routed`,\n`rejected`). Schema in [`reference\u002Fledger.md`](reference\u002Fledger.md). For route-lane\ncandidates: hand the top one to `\u002Farchitecture:improve`, invoked via the Skill tool (if that plugin is installed), for\ndesign exploration, and file the rest by invoking `\u002Fwork-items:track add` via the Skill tool when that plugin is\ninstalled, else the repo's own tracker, else present the list to the user. For each confirmed\nduplicate writer, write the ownership-rule findings file under `\u003Cmemory_dir>\u002Fcoupling\u002F\u003Cbranch-slug>\u002F`\nper [`reference\u002Fremediations.md`](reference\u002Fremediations.md) \"Duplicate writers\", name it in the\nledger entry's `outcome`, and offer `\u002Fcorrections:codify findings \u003Cfile>` when that skill is\navailable; never invoke it unasked. Close by reporting\nwhat waits on the human first (the PR to merge, any candidate needing a decision), then the ledger\npath, what was applied, what was routed where, and the recommended next-run scope.\n\n## Change mode\n\nThe `change` action starts from a value, not a scan: `\u003Cold>` changed or must become `\u003Cnew>`, and\nevery place that states it has to be found and handled by what kind of place it is. It runs its own\npath, not phases B to D. Read [`reference\u002Fchange-mode.md`](reference\u002Fchange-mode.md) first.\n\n1. **Inventory.** Run `python3 \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fvalue-sites.py\" find --old '\u003Cold>'\n   --format summary` from the repository root (add scope paths after the flags), then the same\n   with `--format tsv` for the rows: `class, path, line, col, form, reason, anchor, text`, plus a\n   `skip, path, reason` row for each file it did not read. The script finds every spelling of the\n   value and classifies each site; do not repeat that work by hand.\n2. **Review the classes.** Read each non-setup row and any setup row whose file says otherwise;\n   reclassify with a stated reason. Handle each class per the spoke's Site classes table.\n3. **Propose.** For each setup site, name the reference form it converts to. Report the counts,\n   every non-setup row, and the proposals. `change` without `apply` stops here.\n4. **Apply** (`change apply` only). Wait for the human to confirm the site list and each\n   conversion, then follow the spoke's Apply sequence, which runs phases E and F. `apply` takes\n   each confirmed site as `path:line:col:anchor` from its row and refuses the run if that line\n   changed since `find`.\n5. **Ledger and report.** Record converted and flagged sites in the ledger and report what waits\n   on the human first.\n\n## Fresh-eyes note\n\nPhase B\u002FC findings are judged against artifacts this session did not author, and phase F's\nverdict is a deterministic build\u002Ftest\u002Flint pass. The fresh-context scan agents plus the\nobjective gate carry the independence; a separate fresh-context reviewer is owed only if this\nskill is ever extended to judge work its own session produced outside those gates.\n\n## What this skill does NOT do\n\n- **Does not review a diff.** Pre-merge review is a review tool's job; this hunts standing\n  coupling in existing artifacts.\n- **Does not deep-design one boundary.** A route-lane candidate needing interface\n  exploration goes to `\u002Farchitecture:improve` (when installed) or a design session.\n- **Does not apply general tidyings.** Rename\u002Finline\u002Fextract without a coupling edge is\n  `\u002Fcode-tidying:tidy` territory.\n- **Does not enforce.** Blocking a new literal for a value that already has an owner is a\n  write-time guard's job; `change` cleans up after a change, it does not prevent the next copy.\n- **Does not chase decoupling for its own sake.** A dependency on something stable and owned\n  is not a finding; see the not-a-finding list in the model.\n\n## Composition\n\nGraceful degradation: where a named collaborator is not installed, inline the equivalent\nwork or present the handoff manually; never skip silently. Every skill named in the `Then` column\nis invoked via the Skill tool.\n\n| When | Then |\n|------|------|\n| Route-lane candidate needs design exploration | `\u002Farchitecture:improve` when installed; else summarize the candidate for a design session |\n| Deferred or route-lane items need tracking | `\u002Fwork-items:track add` when installed; else the repo's own tracker; else present the list |\n| Batch needs build\u002Ftest verification | `\u002Ftoolchain:check` when installed; else the project's own commands |\n| Shipping a PR | `\u002Fsource-control:pull-request create` when installed; else the repo's own PR convention |\n| A docs finding is pure prose dedup | `\u002Fdocs-hygiene:extract-ssot` when installed owns the extraction; else apply per the remediation catalog |\n| A `change` is an identifier rename, not a value | `\u002Fdocs-hygiene:rename-references` when installed owns the sweep; else run `change` and review each row; records still stay |\n\n## Next\n\n\u002Fverification:confirm\n\nIt checks the batch against the repo's own build, test and lint gates.\n\n## Gotchas\n\nThe counterweights this skill exists to hold. Add here when a new one surfaces.\n\n- **Over-abstraction is decoupling's own disease.** An interface with one implementation, an\n  event bus for a one-to-one call, a config knob nothing varies. Each adds indirection while\n  the coupling remains. Every remediation entry carries a *not when*; honor it. The deletion\n  test: if removing the new indirection tomorrow would change nothing but line count, it earned\n  nothing.\n- **Cross-file and architectural findings are never auto-applied.** They are designed to\n  inform a human; surfacing them ranked is the success state, not a failure to finish.\n- **Identical text encoding different knowledge is coincidence, not duplication.** Two\n  documents (or functions) that happen to read the same but would change for different\n  reasons must not be consolidated. Consolidation actively harms. Test what changes\n  together, not what looks alike.\n- **Unverified scan claims do not ship.** A scan claim inherits the report's authority, so\n  phase C reproduces every finding against the artifacts before it reaches the ledger or the\n  user.\n- **A reduction that breaks a test was secretly behavioral.** Revert it and reclassify;\n  never patch the test to keep the reduction.\n- **A record that states the old value is correct.** Changelogs, evidence, and decision logs\n  record what was true then; rewriting them to the new value falsifies them. Setup sites change,\n  fixtures only when the human chooses, and contracts only through a proposal.\n- **The ledger records what a re-scan currently finds; it never replays.** Re-emitting stale\n  findings re-injects problems that may already be fixed. Statuses advance, evidence gets\n  re-checked, and a finding that no longer reproduces is closed, not repeated.\n","## Repository context. Gather first\n\nCollect these with **individual** Bash calls, one command per call, never combined into a single\ninvocation:\n\n- Current branch, `git branch --show-current`\n- Recent commits, `git log --oneline -10`\n- Working tree status (empty = clean), `git status --porcelain | head -10`\n\nThe pipe is the bound and belongs in the command. A read-time cap (\"read only the first 10 entries\")\ncomes too late for whatever the command has already returned inline. What a result returns inline,\nand when it arrives as a saved file instead, is [Tools reference § Output\nlimits](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Ftools-reference#output-limits).\n\nTreat a failure (not a repository, git unavailable) as an unknown value and carry on. Keep these as\nseparate body Bash calls rather than pre-compute lines ([gather-block convention](https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fdocs\u002Fconventions\u002Fgather-block\u002FREADME.md#the-pre-compute-block-runs-as-one-shell-invocation),\n\"The pre-compute block runs as one shell invocation\").\n\n## Variables\n\nArguments: `$ARGUMENTS`\n\n## Purpose\n\nAnswer one question, repeatedly: **which dependency in this repository transmits the most\nunnecessary change, and what is the smallest mechanism that stops the transmission?** Each\ninvocation is one pass. Scan, verify, reduce what is safely reducible within a scope budget,\nsurface what is not, and record everything in a durable ledger so the next invocation resumes\ninstead of restarting. Coupling goes down monotonically across runs; no single run tries to\nfinish the job.\n\nTwo lanes, split by the nature of the finding, never by convenience:\n\n- **Apply lane.** Mechanical, contained, behavior-preserving reductions (weaken a\n  connascence form, de-duplicate a stated fact into a pointer, inject a hard-wired volatile\n  collaborator, name a magic value). Applied this run, verified, shipped as one\n  structure-only change set.\n- **Route lane.** Cross-file remediation and architectural judgment (move a boundary,\n  introduce a seam, split a shared database, merge two repos' duplicated logic). Coupling\n  findings of this shape exist to inform a human: they are surfaced, ranked, and routed,\n  never auto-applied.\n\nThis is not a diff reviewer, not a designer of one chosen boundary, and not a general tidy\npass. See \"What this skill does NOT do\".\n\n## Actions\n\n| Argument | Action |\n|----------|--------|\n| *(empty)* | Full pass over an inferred scope: scan → verify → apply lane → route lane → ledger |\n| `\u003Cscope>` | Same pass narrowed to a path, module, or altitude keyword (`docs`, `code`, `app`, `repo`) |\n| `dry-run [\u003Cscope>]` | Scan, verify, rank, and ledger only. Do NOT edit, branch, push, or file tracker items |\n| `change \u003Cold> \u003Cnew> [\u003Cscope>]` | Change mode, read-only: find every site that states `\u003Cold>`, classify each, and propose what happens to it. No edits |\n| `change apply \u003Cold> \u003Cnew> [\u003Cscope>]` | Change mode with edits: the same inventory, then, once the human confirms the site list, change the setup sites and convert them to a reference form |\n| `status` | Read the ledger and report open candidates, applied reductions, and what the next run should take |\n| `help` | Print this table with one worked example per action |\n\n## The model\n\nHub-and-spoke; read the spoke before the phase that needs it:\n\n- [`reference\u002Fcoupling-model.md`](reference\u002Fcoupling-model.md). What counts as coupling, the\n  strength ladder, connascence axes, volatility weighting, per-altitude mechanisms, and the\n  not-a-finding list. Read before scanning; findings are typed against it.\n- [`reference\u002Fremediations.md`](reference\u002Fremediations.md). Mechanism per finding kind, each\n  with its over-abstraction counterweight and the sequencing rule (smallest mechanism first).\n  Read before applying or routing anything.\n- [`reference\u002Fledger.md`](reference\u002Fledger.md). Ledger entry schema, status lifecycle, and\n  re-run semantics. Read at phase A and phase H.\n- [`reference\u002Fchange-mode.md`](reference\u002Fchange-mode.md). The `change` action: how the inventory\n  finds every form of a value, the site classes and what happens to each, reference forms, the\n  apply sequence, and its pitfalls. Read before any `change` run.\n\n## Workflow\n\nPhases run in order; each gate is hard. `dry-run` stops after D (ledger write included).\n\n**A. Orient.** Resolve scope from the argument, else infer from the conversation, else pick\nthe hottest area by commit frequency. The ledger path is the memory tier's constant slug `coupling`, default\n`.work\u002Fcoupling\u002Fcoupling-ledger.md` (`.work\u002F` unless the repo's instructions declare another working-docs root),\nnever committed, one ledger per repo regardless of scope. The slug is constant because coupling\nreduction is repo-scoped and spans many scopes and short-lived branches; a scope- or\nbranch-derived slug would fragment the one ledger successive runs must resume. On the session's\nfirst write, verify the memory root contains a `.gitignore` with `*`, creating it (announced) when\nabsent, and create the slice directory when absent. Create or resume the ledger. Discover the consuming repo's\nown review\u002Fengineering conventions (a review-criteria file such as `REVIEW.md`, a\n`docs\u002Fconventions\u002F` or standards directory, CLAUDE.md rules) and align finding vocabulary and\nseverity with them; the bundled model is the fallback, never an override of the consumer's\nown standards.\n\n**B. Scan.** Fan out fresh-context Explore subagents over the scope, each briefed with the\ncoupling model and returning findings in ledger-entry form (edge, kind, mechanism, strength,\ndegree, locality, evidence). Each is done when its slice of the scope has been read; it returns\nearly with what it has when a finding needs a file outside that slice. In parallel, mine co-change evidence from version-control\nhistory: file pairs that repeatedly change in the same commits without a declared dependency\nare coupled through a channel the import graph cannot see, and they outrank most statically\nvisible findings.\n\nBeside co-change, measure collision hotspots when the repository is hosted on GitHub: run\n`bash \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fcollision-hotspots.sh\" --prs \u003Cn>` from the repository root\n(add `--repo \u003Cowner>\u002F\u003Crepo>` when `gh`'s default repository is not the one that holds the pull\nrequests, `--exclude \u003Cglob>` for more files that change on every release, `--max-pairs \u003Cm>` to\ncap the replays, and `--dry-run` to see the pairs before fetching anything). The script lists\npull requests from that one repository and fetches their heads from `origin` when `origin` names\nit, otherwise from the URL `gh` reports for it. Size `--prs` by the\ndays of history wanted: count the merges into the default branch over the last few days with\n`git log --first-parent --since='\u003Cn> days ago' --format=%h \u003Cdefault-branch>`, one line per merge,\nand multiply the daily rate by the days. The script\npairs pull requests that were open at the same time and share a file, replays each pair with\n`git merge-tree`, and prints `conflicts\u003CTAB>pairs\u003CTAB>path` for every file at least 3 pairs\ntouched; read each ranked file as a candidate typed against the model's collision hotspot entry\nand verify it in phase C like any other finding. Carry its `gap:` lines into the report: a pull\nrequest whose file list stopped at the host's limit is undercounted, and a head that could not\nbe fetched was not replayed. The `known bump hotspots` line names release files that conflict\nfor version bumps, not design; report it once and do not ledger those files. When the script\nexits 2 (no `gh`, no `jq`, a `git` without `merge-tree --write-tree`, no host data), rank by\nco-change alone and record `gap: collision hotspots not measured: \u003Cthe script's message>` in\nthe report.\n\n- Pointer: the 100-file cap per pull request comes from\n  \u003Chttps:\u002F\u002Fdocs.github.com\u002Fen\u002Fgraphql\u002Foverview\u002Frate-limits-and-query-limits-for-the-graphql-api#node-limit>;\n  the exit codes the script reads are in \u003Chttps:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-merge-tree#OUTPUT> and\n  \u003Chttps:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-merge-tree#_exit_status>.\n- As of: 2026-11-10.\n- Recheck trigger: the forge's pull-request file list returns more than 100 files for one pull request,\n  or `git merge-tree --write-tree` changes its exit status meanings.\n\nThen check the ranked files for a second list. Ranked paths come from pull requests and are\nuntrusted: never type one into a command. Pipe the ranking straight in from the repository root,\n`bash \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fcollision-hotspots.sh\" --prs \u003Cn> | awk -F'\\t' 'NF == 3 && \\$1 ~\n\u002F^[0-9]+$\u002F { print \\$3 }' | python3 \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fderivable-list.py\" --paths-from -`\n(same flags as the ranking run), or write the paths one per line to a file with the Write tool\nand pass `--paths-from \u003Cfile>`; with no ranking, write the files co-change mining ranked highest\nthe same way. It prints `candidate`\nfor a file holding a list of at least 5 entries where at least 80% of the entries name a file or\nfolder git tracks, `below` or `no-list` otherwise, an `unnamed` row for each candidate entry that\nnames nothing (a stale entry is how such a list drifts), and `skip` for a path it did not read.\nBoth numbers are judgment values, not settings: 80% leaves room for the stale entries a drifting\nlist carries while ruling out a list whose entries mostly hold data the tree does not, and below 5\nentries a list matches paths by chance and costs little to keep by hand. Verify each candidate in\nphase C (a list a generator already writes is already derived), then report it with the\nDesign-It-Twice comparison in [`reference\u002Fremediations.md`](reference\u002Fremediations.md) \"Second\nlists\". A second list is always a route-lane ledger entry, report-only, never part of the apply\nbatch.\n\nBrief the scan agents to list duplicate writers too: one durable key, path or table that more\nthan one owner writes (the model's common-coupling entry). Use `value-sites.py find` for a literal\nkey. A confirmed duplicate writer always goes to the route lane as a structural ledger entry and\nnever into the apply batch; phase H hands off its ownership rule per the remediation catalog's\n\"Duplicate writers\" section.\n\n**C. Verify (hard gate).** Scan agents have a demonstrated error rate. Reproduce every\nfinding against the actual artifacts before it reaches the ledger or the user. Confirm the\nedge exists, the mechanism is what the scan claims, and the depended-on side actually changes\n(volatility from history, not vibes). Drop or downgrade what does not reproduce, and record\nthat the ledger reflects verified state, not raw scan output.\n\n**D. Partition and rank.** Every verified finding gets a lane. Apply lane requires ALL of:\nmechanical, contained in scope, behavior-preserving, and reversible by revert. Anything\ncross-file in remediation or architectural in judgment is route lane. When in doubt, route.\nRank by `strength × degree × distance × volatility`; a weak-but-everywhere coupling on a hot\npath outranks a strong-but-local one in cold code. Write the full ranked set to the ledger.\n\n**E. Apply (apply lane only, scope-budgeted).** Work on a short-lived branch created from\nthe repository's default branch. Resolve that branch (remote HEAD or the repo's own\nconvention), never assume its name, and never base the batch on whatever feature branch the\nsession happens to be on: inherited unrelated commits would break the structure-only\ninvariant. Never commit directly on the default branch. Before editing any target, require\nit clean in `git status --porcelain`; a target carrying pre-existing local modifications\ndefers its finding with the reason recorded. Foreign edits are never mixed into the batch.\nBudget per run: the target and hard cap in\n[`${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fpr-scope-budget.md`](..\u002F..\u002Freference\u002Fpr-scope-budget.md); overflow stays\n`proposed` in the ledger for the next run. A confirmed `change apply` site set is the one\nexception (see the Apply sequence in [`reference\u002Fchange-mode.md`](reference\u002Fchange-mode.md)). Use the Edit tool; one atomic commit per logical\nreduction; stage listed paths only and inspect the staged diff before each commit. Never\ntouch CI workflow files, hook or settings files, lint configs, database migrations, or\nany published contract surface (API shapes, message schemas, tool schemas). Those are route\nlane by definition.\n\n**F. Verify the batch.** Invoke `\u002Ftoolchain:check` via the Skill tool for the affected\necosystems when the `toolchain` plugin is installed; otherwise run the consuming project's\nown build, test, and lint commands (its CLAUDE.md or rules may name them). A reduction that\nbreaks a test was secretly behavioral: revert that reduction, reclassify it as route lane,\nand continue. Never ship red.\n\n**G. Ship.** Present the verified diff, or when the session should open a pull request,\ninvoke `\u002Fsource-control:pull-request create` via the Skill tool when that plugin is installed; otherwise use the\nrepo's own PR convention with a body listing each reduction as `edge → mechanism → why safe`.\nOne coupling pass per PR; structure-only, no behavioral riders. A human merges. This skill\nnever auto-merges.\n\n**H. Ledger update and report.** Update statuses (`applied`, `deferred`, `routed`,\n`rejected`). Schema in [`reference\u002Fledger.md`](reference\u002Fledger.md). For route-lane\ncandidates: hand the top one to `\u002Farchitecture:improve`, invoked via the Skill tool (if that plugin is installed), for\ndesign exploration, and file the rest by invoking `\u002Fwork-items:track add` via the Skill tool when that plugin is\ninstalled, else the repo's own tracker, else present the list to the user. For each confirmed\nduplicate writer, write the ownership-rule findings file under `\u003Cmemory_dir>\u002Fcoupling\u002F\u003Cbranch-slug>\u002F`\nper [`reference\u002Fremediations.md`](reference\u002Fremediations.md) \"Duplicate writers\", name it in the\nledger entry's `outcome`, and offer `\u002Fcorrections:codify findings \u003Cfile>` when that skill is\navailable; never invoke it unasked. Close by reporting\nwhat waits on the human first (the PR to merge, any candidate needing a decision), then the ledger\npath, what was applied, what was routed where, and the recommended next-run scope.\n\n## Change mode\n\nThe `change` action starts from a value, not a scan: `\u003Cold>` changed or must become `\u003Cnew>`, and\nevery place that states it has to be found and handled by what kind of place it is. It runs its own\npath, not phases B to D. Read [`reference\u002Fchange-mode.md`](reference\u002Fchange-mode.md) first.\n\n1. **Inventory.** Run `python3 \"${CLAUDE_SKILL_DIR}\u002Fscripts\u002Fvalue-sites.py\" find --old '\u003Cold>'\n   --format summary` from the repository root (add scope paths after the flags), then the same\n   with `--format tsv` for the rows: `class, path, line, col, form, reason, anchor, text`, plus a\n   `skip, path, reason` row for each file it did not read. The script finds every spelling of the\n   value and classifies each site; do not repeat that work by hand.\n2. **Review the classes.** Read each non-setup row and any setup row whose file says otherwise;\n   reclassify with a stated reason. Handle each class per the spoke's Site classes table.\n3. **Propose.** For each setup site, name the reference form it converts to. Report the counts,\n   every non-setup row, and the proposals. `change` without `apply` stops here.\n4. **Apply** (`change apply` only). Wait for the human to confirm the site list and each\n   conversion, then follow the spoke's Apply sequence, which runs phases E and F. `apply` takes\n   each confirmed site as `path:line:col:anchor` from its row and refuses the run if that line\n   changed since `find`.\n5. **Ledger and report.** Record converted and flagged sites in the ledger and report what waits\n   on the human first.\n\n## Fresh-eyes note\n\nPhase B\u002FC findings are judged against artifacts this session did not author, and phase F's\nverdict is a deterministic build\u002Ftest\u002Flint pass. The fresh-context scan agents plus the\nobjective gate carry the independence; a separate fresh-context reviewer is owed only if this\nskill is ever extended to judge work its own session produced outside those gates.\n\n## What this skill does NOT do\n\n- **Does not review a diff.** Pre-merge review is a review tool's job; this hunts standing\n  coupling in existing artifacts.\n- **Does not deep-design one boundary.** A route-lane candidate needing interface\n  exploration goes to `\u002Farchitecture:improve` (when installed) or a design session.\n- **Does not apply general tidyings.** Rename\u002Finline\u002Fextract without a coupling edge is\n  `\u002Fcode-tidying:tidy` territory.\n- **Does not enforce.** Blocking a new literal for a value that already has an owner is a\n  write-time guard's job; `change` cleans up after a change, it does not prevent the next copy.\n- **Does not chase decoupling for its own sake.** A dependency on something stable and owned\n  is not a finding; see the not-a-finding list in the model.\n\n## Composition\n\nGraceful degradation: where a named collaborator is not installed, inline the equivalent\nwork or present the handoff manually; never skip silently. Every skill named in the `Then` column\nis invoked via the Skill tool.\n\n| When | Then |\n|------|------|\n| Route-lane candidate needs design exploration | `\u002Farchitecture:improve` when installed; else summarize the candidate for a design session |\n| Deferred or route-lane items need tracking | `\u002Fwork-items:track add` when installed; else the repo's own tracker; else present the list |\n| Batch needs build\u002Ftest verification | `\u002Ftoolchain:check` when installed; else the project's own commands |\n| Shipping a PR | `\u002Fsource-control:pull-request create` when installed; else the repo's own PR convention |\n| A docs finding is pure prose dedup | `\u002Fdocs-hygiene:extract-ssot` when installed owns the extraction; else apply per the remediation catalog |\n| A `change` is an identifier rename, not a value | `\u002Fdocs-hygiene:rename-references` when installed owns the sweep; else run `change` and review each row; records still stay |\n\n## Next\n\n\u002Fverification:confirm\n\nIt checks the batch against the repo's own build, test and lint gates.\n\n## Gotchas\n\nThe counterweights this skill exists to hold. Add here when a new one surfaces.\n\n- **Over-abstraction is decoupling's own disease.** An interface with one implementation, an\n  event bus for a one-to-one call, a config knob nothing varies. Each adds indirection while\n  the coupling remains. Every remediation entry carries a *not when*; honor it. The deletion\n  test: if removing the new indirection tomorrow would change nothing but line count, it earned\n  nothing.\n- **Cross-file and architectural findings are never auto-applied.** They are designed to\n  inform a human; surfacing them ranked is the success state, not a failure to finish.\n- **Identical text encoding different knowledge is coincidence, not duplication.** Two\n  documents (or functions) that happen to read the same but would change for different\n  reasons must not be consolidated. Consolidation actively harms. Test what changes\n  together, not what looks alike.\n- **Unverified scan claims do not ship.** A scan claim inherits the report's authority, so\n  phase C reproduces every finding against the artifacts before it reaches the ledger or the\n  user.\n- **A reduction that breaks a test was secretly behavioral.** Revert it and reclassify;\n  never patch the test to keep the reduction.\n- **A record that states the old value is correct.** Changelogs, evidence, and decision logs\n  record what was true then; rewriting them to the new value falsifies them. Setup sites change,\n  fixtures only when the human chooses, and contracts only through a proposal.\n- **The ledger records what a re-scan currently finds; it never replays.** Re-emitting stale\n  findings re-injects problems that may already be fixed. Statuses advance, evidence gets\n  re-checked, and a finding that no longer reproduces is closed, not repeated.\n",{"metadata":24,"description":7,"argument-hint":27,"user-invocable":28,"disable-model-invocation":20},{"summary":25,"workflow-stage":26},"Scan for change-transmitting coupling, apply safe reductions in a budgeted batch, route the rest","anytime","[\u003Cscope>|dry-run [\u003Cscope>]|change [apply] \u003Cold> \u003Cnew> [\u003Cscope>]|status|help]",true,"main","d0e68c3f36b92a1d6381f0ba0060d7a0b8b886c5","plugins\u002Fcoupling\u002Fskills\u002Freduce",7,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins","https:\u002F\u002Fgithub.com\u002Fmelodic-software","https:\u002F\u002Fmelodicsoftware.com","Independent software studio building developer tools, automation, and application services, with roots in music.","2026-11-15 05:15:06",{"source":8,"skillName":5,"commands":39,"prompt":49,"zipAvailable":28},[40,43,46],{"tool":41,"command":42},"npx","npx skills add melodic-software\u002Fclaude-code-plugins --skill reduce",{"tool":44,"command":45},"bunx","bunx skills add melodic-software\u002Fclaude-code-plugins --skill reduce",{"tool":47,"command":48},"pnpm","pnpm dlx skills add melodic-software\u002Fclaude-code-plugins --skill reduce","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002FSKILL.md",[51,57,61,65,69,75,79,83,87,91,95],{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"reference\u002Fchange-mode.md","REFERENCE","参考文档",15237,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Freference\u002Fchange-mode.md",{"path":58,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":59,"sourceHtmlUrl":60},"reference\u002Fcoupling-model.md",10098,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Freference\u002Fcoupling-model.md",{"path":62,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":63,"sourceHtmlUrl":64},"reference\u002Fledger.md",3345,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Freference\u002Fledger.md",{"path":66,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":67,"sourceHtmlUrl":68},"reference\u002Fremediations.md",13455,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Freference\u002Fremediations.md",{"path":70,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":73,"sourceHtmlUrl":74},"scripts\u002Fcollision-hotspots.sh","SCRIPT","脚本",12464,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Fcollision-hotspots.sh",{"path":76,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":77,"sourceHtmlUrl":78},"scripts\u002Fcollision-hotspots.test.sh",14895,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Fcollision-hotspots.test.sh",{"path":80,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":81,"sourceHtmlUrl":82},"scripts\u002Fderivable-list.py",7951,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Fderivable-list.py",{"path":84,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":85,"sourceHtmlUrl":86},"scripts\u002Fderivable-list.test.sh",11160,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Fderivable-list.test.sh",{"path":88,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":89,"sourceHtmlUrl":90},"scripts\u002Ftest-helpers.sh",1270,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Ftest-helpers.sh",{"path":92,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":93,"sourceHtmlUrl":94},"scripts\u002Fvalue-sites.py",24310,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Fvalue-sites.py",{"path":96,"resourceType":71,"resourceTypeDesc":72,"sizeBytes":97,"sourceHtmlUrl":98},"scripts\u002Fvalue-sites.test.sh",40059,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcoupling\u002Fskills\u002Freduce\u002Fscripts\u002Fvalue-sites.test.sh",[100,107,114,121,126,133],{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":19},"01M7DYYDG31B9YX7N9MB9QW2GW","track","Backlog reads through the bound tracker: search open and closed items, view one item with its comments and state history. Use when another skill searches or reads tracker items.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fevals\u002Frecall-still-reports-reopened-bug\u002Fwork-items-stub\u002Fskills\u002Ftrack\u002FSKILL.md",[],[],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":19},"01M7DYXVBAVG3NV30PA7GKW9Q6","reviewer-media","Reviewer media for one pull request: decides whether it needs a narrated walkthrough, builds the storyboard from \u002Freview:explain-change's digest, has \u002Fexplainer-video:produce render it and hands the manifest to \u002Fsource-control:pull-request media. Use when asked for 'reviewer media', 'a walkthrough video for this PR', 'a produced video of this change', or when pull-request ready or monitor hands a PR over.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Freview\u002Fskills\u002Freviewer-media\u002FSKILL.md",[],[],{"skillNo":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":19},"01M7DYRZ5YD44AMYH26YR25GZ8","route","Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-pin-without-override\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":122,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":19},"01M7DYRTVT4FWTGDJY0HCCFJMZ","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-override-from-route\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":127,"name":128,"nameZh":6,"description":129,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":130,"categories":131,"keywords":132,"publishedAt":19},"01M7DYNZC8PNW3T27V2ADKECKJ","principles","Module-design vocabulary for architecture findings, and the checks built on it. Use when: 'what is a deep module', 'what is a seam', 'deep vs shallow module', 'what is the deletion test', 'leverage and locality', 'module design vocabulary', 'which word do we use for X in architecture findings'. Scanning a codebase for friction is \u002Farchitecture:improve.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fskills\u002Fprinciples\u002FSKILL.md",[],[],{"skillNo":134,"name":135,"nameZh":6,"description":136,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":137,"categories":138,"keywords":139,"publishedAt":19},"01M7DYNMCZNEWBFHV9KTSS6H60","write","Structured five-field bug report (title, steps to reproduce, expected vs actual, justified severity, suggested fix location) from an informal description, read-only. Use when the user names a defect they observed, or asks for one written up ('report a bug', 'file a bug', 'write this up as a bug').","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fevals\u002Fdeepening-defect-goes-to-bugs-write\u002Fbugs-stub\u002Fskills\u002Fwrite\u002FSKILL.md",[],[]]