[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CQMD1MEVPCZ79K94685GFG":3,"skill-same:01M7CQMD1MEVPCZ79K94685GFG":83},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M7CQMD1MEVPCZ79K94685GFG","git-memory",null,"Classify commit\u002FPR memory before every git commit, gh pr create, or gh pr merge. Also use to recall why an old Git decision was made.","kouko\u002Floom-plugins","kouko","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1480254?v=4","01M7CQCPAJR9C92XDGR0QDCQ98","01M7CQCPB8FTVY1BT26ENK1YP4",0,"MIT","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: git-memory\ndescription: |\n  Classify commit\u002FPR memory before every git commit, gh pr create, or gh pr merge. Also use to recall why an old Git decision was made.\n---\n\n# Git Memory\n\nDurable lessons live in the repo's committed memory store\n(`docs\u002Floom\u002Fmemory\u002F` here), the authoritative carrier. Git artifacts provide\ncommit-bound capture: best-effort, secondary, and never the retrieval path a\ndurable lesson depends on.\n\n## Invocation policy\n\n> **This skill is an invocation gate, not a trailer gate.**\n\nTwo decisions must remain separate:\n\n| Decision | Owner | Rule |\n|---|---|---|\n| Invoke this skill? | Caller | Before `git commit` \u002F `gh pr create` \u002F `gh pr merge`: always yes in an agent session |\n| Add memory content? | This skill | Classify inside the skill; routine commits exit cleanly with no trailers |\n\nNever let the caller's pre-judgment replace the classification logic. The gate\nmust run even when its result is “no memory needed.” `gh pr merge`, especially\nwith `--squash`, is the last checkpoint before the branch closes and the final\nchance to catch an empty memory-worthy branch.\n\n## Carrier hierarchy\n\nUse the committed memory store for durable knowledge. Use git artifacts to\ncapture the why at the change boundary so any git-capable tool or human can\nrecover it without a vendor-specific database. These layers complement, rather\nthan replace, user-level agent memory.\n\n| Layer | Scope | Retrieval |\n|---|---|---|\n| Committed memory store | Durable repo lesson | Repository files and their index |\n| Commit messages + PR bodies | Change-bound capture | `git log`, PR history, and `scripts\u002Fmemory-grep.sh` |\n| Agent-native memory | User preferences across projects | Host-managed, session-loaded memory |\n\nDo not distill all git-memory into an always-loaded context file. Project\ndecisions are pulled on demand; user preferences can remain pushed by the host.\n\n## What to capture\n\nThe diff already shows what changed. Capture why it changed when at least one\nof these applies:\n\n- **Decision** — a non-obvious choice had a real alternative.\n- **Learning** — the work revealed a surprising constraint or behavior.\n- **Gotcha** — a concrete trap would otherwise catch a future contributor.\n\nUse the minimum earned set of structured trailers:\n\n- `Decision:` — why this approach won over the alternative.\n- `Learning:` — the specific discovery worth retaining.\n- `Gotcha:` — the misleading surface and the correct path.\n- `Related:` — an earlier PR whose context this change extends.\n- `Supersedes:` — the earlier live decision this change replaces.\n\nRoutine typo, formatting, version-bump, and obvious implementation commits\nproduce no trailers. Do not record restatements of the diff, generic advice,\nor user preferences unrelated to the repository. The full schema, wrapping,\nliveness rules, and examples live in `standards\u002Fmemory-conventions.md`.\n\n### Classification discipline\n\nAsk whether the work made a choice future readers could reasonably reopen,\ndiscovered a constraint they cannot infer from the code, or exposed a named\nfailure mode. Any yes makes the change memory-worthy. All no means the correct\noutput is empty. This classification happens separately for a commit and its\nPR: a sequence of ordinary commits can still produce a memory-worthy PR when\ntheir combined architecture or policy matters.\n\nDo not fill every field as a completeness ritual. One precise `Decision:` is\nbetter than three vague trailers. A useful entry remains actionable after the\nworking context disappears: it names the rejected alternative, the observed\nconstraint, or the recognizable trap. `Related:` links supporting context;\n`Supersedes:` declares replacement and lets recall hide the retired decision by\ndefault. Validate either relationship as required by the commit protocol so an\nimmutable pointer does not silently lead to a carrier with no memory.\n\n## Compose at each git boundary\n\n### Commit\n\nBefore authoring or finalizing a commit, read\n`protocols\u002Fcompose-commit.md`. It owns the memory-worthiness filters, trailer\ncomposition, `Related:`\u002F`Supersedes:` validation, diagram venue, confirmation,\nand delegated close-out exception. Compose concise why-focused prose, then the\nminimum useful trailers.\n\n### PR create\n\nBefore `gh pr create`, read `protocols\u002Fcompose-pr.md`. For a memory-worthy PR,\nthe caller owns the body schema and publication consent. For Loom, contribute\nearned rationale inside Ship's contextual sections; never add `## Memory`, a\nsecond footer contract, or another confirmation.\n\n**For Loom, a durable `docs\u002Floom\u002Fmemory\u002F` entry must land before\n`closing-review`'s `finalize-review` generates the attestation for that\nchange, never after.** The attestation binds a `content_digest` computed over\nthe committed tree; a memory file (and its index entry) committed afterward\nis itself functional content, so it changes that digest and invalidates the\nattestation the branch already has — a push then fails on the mismatch, or\nthe addition has to be reverted to keep the digest the reviewers actually\nread. Fold the memory entry into whichever fix round is still open, before\n`finalize-review` runs; a lesson that surfaces only after it has run rides\nthe next change's branch instead of reopening this one.\n\nFor a non-Loom caller without an owning schema, both generic carriers are\nused:\n\n1. a rendered `## Memory` section after the standard test-plan content and\n   before the generated-attribution footer; and\n2. an unbolded raw `Decision:`\u002F`Learning:`\u002F`Gotcha:` footer as the absolute\n   last authored block.\n\nFor a routine non-Loom PR, omit both. Do not alter a caller-owned schema to\nsmuggle memory content into it.\n\nFor non-Loom callers, the rendered section and raw footer serve different readers. The section can\nexplain alternatives, learnings, gotchas, and architecture in readable prose.\nThe raw footer keeps short keys available to git retrieval. Memory-worthy means\nboth are required; neither substitutes for the other. Place Mermaid diagrams\nin PR prose when architecture, data flow, or state changes justify one. Commit\nmessages may use a small ASCII diagram, but routine changes need no diagram.\n\n### Merge and capture verification\n\nFor a non-Loom PR using the generic carriers, verify capture rather than\nassuming the authoring step worked:\n\n- Run `scripts\u002Fmemory-grep.sh --verify \u003Cref>` for the commit carrier. Exit `0`\n  means at least one memory key is text-retrievable; exit `4` means empty.\n- Confirm its PR `## Memory` section is present.\n- Use `--verify-merged \u003Cref>` after merge to catch a heading whose keys were\n  silently dropped, including a title-only squash result.\n- Use `--verify-strict \u003Cref>` only as a footer-parse diagnostic; it is not the\n  durable-lesson path.\n\nAn empty result is a flag for that non-Loom caller to fix **before** merge.\nThis verification proves commit-bound capture landed; it does not create a\nLoom lifecycle gate or mean that a durable lesson was filed.\n\nInterpret the checks narrowly. Plain `--verify` proves a key is visible in the\nmessage text, including the mid-body produced by squash. `--verify-merged`\nguards the merged carrier against a PR body or Memory section being discarded.\nStrict verification asks whether git still recognizes a real trailing footer;\nits failure can coexist with successful grep-level capture. None of these\nchecks promotes a broadly reusable lesson into `docs\u002Floom\u002Fmemory\u002F`; durable\nfiling is a separate repository workflow.\n\n## Squash-merge caveat\n\n> **On a squash-merged default branch, `%(trailers)` is unreliable.**\n\nA squash can relocate per-commit trailers into the merged commit's mid-body.\nGitHub may also append a divider and `Co-authored-by:` block, or hard-wrap a\nline, after the authored PR footer (live-observed on PR #576). Footer-only\nparsing through `%(trailers)` or\n`git interpret-trailers --parse` can therefore miss valid capture even when\nthe PR body followed the best-effort placement mandate.\n\n`git log --grep` and `memory-grep.sh --verify` still find keys that begin at a\nline start. A non-Loom PR's `## Memory` section remains human-readable on\nGitHub. Both\nare commit-bound evidence; the committed memory store remains authoritative.\nNever promise that a raw PR trailer footer guarantees structured parsing. Its\nguaranteed floor is grep-level retrieval. A merge commit is also valid when\npreserving each commit's real footer matters.\n\nKeep raw trailer lines short and consecutive, separated from preceding prose\nby one blank line, with no authored heading, comment, fence, or ordinary text\nafter them. That placement gives structured parsing its best chance before the\nhosting platform performs its own squash transformation. On feature branches,\nmerge-commit histories, and rebase-merge histories, a genuine footer remains\navailable to `%(trailers)`; on squash `main`, prefer the robust retrieval paths\nabove.\n\n## Privacy gate — fail-closed\n\nGit is public-by-default. After composing any commit message or PR body, follow\nthe exact two-layer privacy gate in its compose protocol:\n\n1. Run `scripts\u002Fprivacy-scan.py --text-file \u003Ccomposed>`.\n2. Layer 2 is conditional: known public identifiers do not dispatch. Only\n   ambiguous private-party identifying text follows the fresh-context judge\n   procedure in `protocols\u002Fprivacy-judge-spec.md` over the same text.\n3. Continue when the deterministic scan is clean and either no judge is\n   required or the dispatched judge returns a conforming `PASS`.\n\nAny deterministic finding, dispatched judge `BLOCK`, script error, required\ndispatch failure, or malformed required judge result makes the carrier\n**BLOCKED**. Surface the findings, stop, and escalate to the human. Delegated\nclose-out consent never overrides this stop. The commit protocol's optional\nquality note is advisory only; it cannot turn a privacy failure into a pass.\n\nRun the gate over the exact final text that would be committed or sent to\nGitHub, not an earlier excerpt. Do not manually waive a deterministic match or\nreplace the fresh judge with the composing agent's own confidence. When the\nprotocol is delegated from an already-authorized close-out, that authorization\ncan remove duplicate publication confirmation; it cannot authorize a blocked\ncarrier or broaden what sensitive material may leave the machine.\n\nNever write secrets or sensitive context into repository history. Use an\nappropriate ignored note or secret manager instead.\n\n- **Secrets or sensitive context** — the compose protocols enforce a\n  fail-closed privacy gate: `protocols\u002Fcompose-commit.md` and\n  `protocols\u002Fcompose-pr.md` always run `scripts\u002Fprivacy-scan.py` and invoke\n  the judge in `protocols\u002Fprivacy-judge-spec.md` only for ambiguous\n  private-party text. Any required gate failure stops publication.\n\n## Recall routing\n\nPast decisions are **pulled** on demand, not preloaded. Read\n`protocols\u002Frecall.md` when:\n\n- the user asks why an old decision, branch, or PR exists;\n- non-trivial work begins in a concrete file or module; or\n- planning weighs an alternative that may already have been rejected.\n\nRecall once with a scoped `--path` or `--match`, cap the result with `--top`,\nand accept empty output as normal. Default retrieval is live-only; add\n`--history` only to explain a superseded chain. Surface the answer in the\nuser's language, point-first, citing the PR or commit rather than dumping raw\ntrailers. If new work intentionally reverses a live decision, say so and treat\nit as a candidate `Supersedes:` relation.\n\nFor proactive work, use one scoped path recall before changing a non-trivial\narea. For alternative analysis, use one topic recall before recommending the\napproach. Do not broaden or loop because a query returned nothing; absence is a\nnormal result, not an operational failure. A live match should prevent silent\nre-litigation. A superseded match should be presented with its replacement so\nthe retired choice is not accidentally revived. Recall supports the current\ntask; it is not a transcript dump or session-start priming ritual.\n\n## Resource map\n\n- `protocols\u002Fcompose-commit.md` — commit classification, composition, privacy,\n  consent, and relation validation.\n- `protocols\u002Fcompose-pr.md` — PR memory section, footer placement, privacy, and\n  confirmation.\n- `protocols\u002Fprivacy-judge-spec.md` — fresh-context privacy judge contract.\n- `protocols\u002Frecall.md` — scoped pull triggers, commands, liveness, and user\n  reporting.\n- `standards\u002Fmemory-conventions.md` — schema, examples, supersession, line\n  length, diagram venues, and retrieval doctrine.\n- `scripts\u002Fmemory-grep.sh` — scoped recall plus normal, merged, and strict\n  capture verification.\n- `scripts\u002Fprivacy-scan.py` — deterministic first privacy layer.\n\n## License\n\nMIT — see repository root `LICENSE`.\n","# Git Memory\n\nDurable lessons live in the repo's committed memory store\n(`docs\u002Floom\u002Fmemory\u002F` here), the authoritative carrier. Git artifacts provide\ncommit-bound capture: best-effort, secondary, and never the retrieval path a\ndurable lesson depends on.\n\n## Invocation policy\n\n> **This skill is an invocation gate, not a trailer gate.**\n\nTwo decisions must remain separate:\n\n| Decision | Owner | Rule |\n|---|---|---|\n| Invoke this skill? | Caller | Before `git commit` \u002F `gh pr create` \u002F `gh pr merge`: always yes in an agent session |\n| Add memory content? | This skill | Classify inside the skill; routine commits exit cleanly with no trailers |\n\nNever let the caller's pre-judgment replace the classification logic. The gate\nmust run even when its result is “no memory needed.” `gh pr merge`, especially\nwith `--squash`, is the last checkpoint before the branch closes and the final\nchance to catch an empty memory-worthy branch.\n\n## Carrier hierarchy\n\nUse the committed memory store for durable knowledge. Use git artifacts to\ncapture the why at the change boundary so any git-capable tool or human can\nrecover it without a vendor-specific database. These layers complement, rather\nthan replace, user-level agent memory.\n\n| Layer | Scope | Retrieval |\n|---|---|---|\n| Committed memory store | Durable repo lesson | Repository files and their index |\n| Commit messages + PR bodies | Change-bound capture | `git log`, PR history, and `scripts\u002Fmemory-grep.sh` |\n| Agent-native memory | User preferences across projects | Host-managed, session-loaded memory |\n\nDo not distill all git-memory into an always-loaded context file. Project\ndecisions are pulled on demand; user preferences can remain pushed by the host.\n\n## What to capture\n\nThe diff already shows what changed. Capture why it changed when at least one\nof these applies:\n\n- **Decision** — a non-obvious choice had a real alternative.\n- **Learning** — the work revealed a surprising constraint or behavior.\n- **Gotcha** — a concrete trap would otherwise catch a future contributor.\n\nUse the minimum earned set of structured trailers:\n\n- `Decision:` — why this approach won over the alternative.\n- `Learning:` — the specific discovery worth retaining.\n- `Gotcha:` — the misleading surface and the correct path.\n- `Related:` — an earlier PR whose context this change extends.\n- `Supersedes:` — the earlier live decision this change replaces.\n\nRoutine typo, formatting, version-bump, and obvious implementation commits\nproduce no trailers. Do not record restatements of the diff, generic advice,\nor user preferences unrelated to the repository. The full schema, wrapping,\nliveness rules, and examples live in `standards\u002Fmemory-conventions.md`.\n\n### Classification discipline\n\nAsk whether the work made a choice future readers could reasonably reopen,\ndiscovered a constraint they cannot infer from the code, or exposed a named\nfailure mode. Any yes makes the change memory-worthy. All no means the correct\noutput is empty. This classification happens separately for a commit and its\nPR: a sequence of ordinary commits can still produce a memory-worthy PR when\ntheir combined architecture or policy matters.\n\nDo not fill every field as a completeness ritual. One precise `Decision:` is\nbetter than three vague trailers. A useful entry remains actionable after the\nworking context disappears: it names the rejected alternative, the observed\nconstraint, or the recognizable trap. `Related:` links supporting context;\n`Supersedes:` declares replacement and lets recall hide the retired decision by\ndefault. Validate either relationship as required by the commit protocol so an\nimmutable pointer does not silently lead to a carrier with no memory.\n\n## Compose at each git boundary\n\n### Commit\n\nBefore authoring or finalizing a commit, read\n`protocols\u002Fcompose-commit.md`. It owns the memory-worthiness filters, trailer\ncomposition, `Related:`\u002F`Supersedes:` validation, diagram venue, confirmation,\nand delegated close-out exception. Compose concise why-focused prose, then the\nminimum useful trailers.\n\n### PR create\n\nBefore `gh pr create`, read `protocols\u002Fcompose-pr.md`. For a memory-worthy PR,\nthe caller owns the body schema and publication consent. For Loom, contribute\nearned rationale inside Ship's contextual sections; never add `## Memory`, a\nsecond footer contract, or another confirmation.\n\n**For Loom, a durable `docs\u002Floom\u002Fmemory\u002F` entry must land before\n`closing-review`'s `finalize-review` generates the attestation for that\nchange, never after.** The attestation binds a `content_digest` computed over\nthe committed tree; a memory file (and its index entry) committed afterward\nis itself functional content, so it changes that digest and invalidates the\nattestation the branch already has — a push then fails on the mismatch, or\nthe addition has to be reverted to keep the digest the reviewers actually\nread. Fold the memory entry into whichever fix round is still open, before\n`finalize-review` runs; a lesson that surfaces only after it has run rides\nthe next change's branch instead of reopening this one.\n\nFor a non-Loom caller without an owning schema, both generic carriers are\nused:\n\n1. a rendered `## Memory` section after the standard test-plan content and\n   before the generated-attribution footer; and\n2. an unbolded raw `Decision:`\u002F`Learning:`\u002F`Gotcha:` footer as the absolute\n   last authored block.\n\nFor a routine non-Loom PR, omit both. Do not alter a caller-owned schema to\nsmuggle memory content into it.\n\nFor non-Loom callers, the rendered section and raw footer serve different readers. The section can\nexplain alternatives, learnings, gotchas, and architecture in readable prose.\nThe raw footer keeps short keys available to git retrieval. Memory-worthy means\nboth are required; neither substitutes for the other. Place Mermaid diagrams\nin PR prose when architecture, data flow, or state changes justify one. Commit\nmessages may use a small ASCII diagram, but routine changes need no diagram.\n\n### Merge and capture verification\n\nFor a non-Loom PR using the generic carriers, verify capture rather than\nassuming the authoring step worked:\n\n- Run `scripts\u002Fmemory-grep.sh --verify \u003Cref>` for the commit carrier. Exit `0`\n  means at least one memory key is text-retrievable; exit `4` means empty.\n- Confirm its PR `## Memory` section is present.\n- Use `--verify-merged \u003Cref>` after merge to catch a heading whose keys were\n  silently dropped, including a title-only squash result.\n- Use `--verify-strict \u003Cref>` only as a footer-parse diagnostic; it is not the\n  durable-lesson path.\n\nAn empty result is a flag for that non-Loom caller to fix **before** merge.\nThis verification proves commit-bound capture landed; it does not create a\nLoom lifecycle gate or mean that a durable lesson was filed.\n\nInterpret the checks narrowly. Plain `--verify` proves a key is visible in the\nmessage text, including the mid-body produced by squash. `--verify-merged`\nguards the merged carrier against a PR body or Memory section being discarded.\nStrict verification asks whether git still recognizes a real trailing footer;\nits failure can coexist with successful grep-level capture. None of these\nchecks promotes a broadly reusable lesson into `docs\u002Floom\u002Fmemory\u002F`; durable\nfiling is a separate repository workflow.\n\n## Squash-merge caveat\n\n> **On a squash-merged default branch, `%(trailers)` is unreliable.**\n\nA squash can relocate per-commit trailers into the merged commit's mid-body.\nGitHub may also append a divider and `Co-authored-by:` block, or hard-wrap a\nline, after the authored PR footer (live-observed on PR #576). Footer-only\nparsing through `%(trailers)` or\n`git interpret-trailers --parse` can therefore miss valid capture even when\nthe PR body followed the best-effort placement mandate.\n\n`git log --grep` and `memory-grep.sh --verify` still find keys that begin at a\nline start. A non-Loom PR's `## Memory` section remains human-readable on\nGitHub. Both\nare commit-bound evidence; the committed memory store remains authoritative.\nNever promise that a raw PR trailer footer guarantees structured parsing. Its\nguaranteed floor is grep-level retrieval. A merge commit is also valid when\npreserving each commit's real footer matters.\n\nKeep raw trailer lines short and consecutive, separated from preceding prose\nby one blank line, with no authored heading, comment, fence, or ordinary text\nafter them. That placement gives structured parsing its best chance before the\nhosting platform performs its own squash transformation. On feature branches,\nmerge-commit histories, and rebase-merge histories, a genuine footer remains\navailable to `%(trailers)`; on squash `main`, prefer the robust retrieval paths\nabove.\n\n## Privacy gate — fail-closed\n\nGit is public-by-default. After composing any commit message or PR body, follow\nthe exact two-layer privacy gate in its compose protocol:\n\n1. Run `scripts\u002Fprivacy-scan.py --text-file \u003Ccomposed>`.\n2. Layer 2 is conditional: known public identifiers do not dispatch. Only\n   ambiguous private-party identifying text follows the fresh-context judge\n   procedure in `protocols\u002Fprivacy-judge-spec.md` over the same text.\n3. Continue when the deterministic scan is clean and either no judge is\n   required or the dispatched judge returns a conforming `PASS`.\n\nAny deterministic finding, dispatched judge `BLOCK`, script error, required\ndispatch failure, or malformed required judge result makes the carrier\n**BLOCKED**. Surface the findings, stop, and escalate to the human. Delegated\nclose-out consent never overrides this stop. The commit protocol's optional\nquality note is advisory only; it cannot turn a privacy failure into a pass.\n\nRun the gate over the exact final text that would be committed or sent to\nGitHub, not an earlier excerpt. Do not manually waive a deterministic match or\nreplace the fresh judge with the composing agent's own confidence. When the\nprotocol is delegated from an already-authorized close-out, that authorization\ncan remove duplicate publication confirmation; it cannot authorize a blocked\ncarrier or broaden what sensitive material may leave the machine.\n\nNever write secrets or sensitive context into repository history. Use an\nappropriate ignored note or secret manager instead.\n\n- **Secrets or sensitive context** — the compose protocols enforce a\n  fail-closed privacy gate: `protocols\u002Fcompose-commit.md` and\n  `protocols\u002Fcompose-pr.md` always run `scripts\u002Fprivacy-scan.py` and invoke\n  the judge in `protocols\u002Fprivacy-judge-spec.md` only for ambiguous\n  private-party text. Any required gate failure stops publication.\n\n## Recall routing\n\nPast decisions are **pulled** on demand, not preloaded. Read\n`protocols\u002Frecall.md` when:\n\n- the user asks why an old decision, branch, or PR exists;\n- non-trivial work begins in a concrete file or module; or\n- planning weighs an alternative that may already have been rejected.\n\nRecall once with a scoped `--path` or `--match`, cap the result with `--top`,\nand accept empty output as normal. Default retrieval is live-only; add\n`--history` only to explain a superseded chain. Surface the answer in the\nuser's language, point-first, citing the PR or commit rather than dumping raw\ntrailers. If new work intentionally reverses a live decision, say so and treat\nit as a candidate `Supersedes:` relation.\n\nFor proactive work, use one scoped path recall before changing a non-trivial\narea. For alternative analysis, use one topic recall before recommending the\napproach. Do not broaden or loop because a query returned nothing; absence is a\nnormal result, not an operational failure. A live match should prevent silent\nre-litigation. A superseded match should be presented with its replacement so\nthe retired choice is not accidentally revived. Recall supports the current\ntask; it is not a transcript dump or session-start priming ritual.\n\n## Resource map\n\n- `protocols\u002Fcompose-commit.md` — commit classification, composition, privacy,\n  consent, and relation validation.\n- `protocols\u002Fcompose-pr.md` — PR memory section, footer placement, privacy, and\n  confirmation.\n- `protocols\u002Fprivacy-judge-spec.md` — fresh-context privacy judge contract.\n- `protocols\u002Frecall.md` — scoped pull triggers, commands, liveness, and user\n  reporting.\n- `standards\u002Fmemory-conventions.md` — schema, examples, supersession, line\n  length, diagram venues, and retrieval doctrine.\n- `scripts\u002Fmemory-grep.sh` — scoped recall plus normal, merged, and strict\n  capture verification.\n- `scripts\u002Fprivacy-scan.py` — deterministic first privacy layer.\n\n## License\n\nMIT — see repository root `LICENSE`.\n",{"name":5,"description":7},"main","3b1a9972644517911454530dc1876f88525b1f37","loom-workflow\u002Fskills\u002Fgit-memory","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins","https:\u002F\u002Fgithub.com\u002Fkouko","2026-11-15 00:45:02",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add kouko\u002Floom-plugins --skill git-memory",{"tool":35,"command":36},"bunx","bunx skills add kouko\u002Floom-plugins --skill git-memory",{"tool":38,"command":39},"pnpm","pnpm dlx skills add kouko\u002Floom-plugins --skill git-memory","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002FSKILL.md",true,[43,49,53,57,61,65,71,75,79],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"protocols\u002Fcompose-commit.md","REFERENCE","参考文档",8075,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fprotocols\u002Fcompose-commit.md",{"path":50,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":51,"sourceHtmlUrl":52},"protocols\u002Fcompose-pr.md",11800,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fprotocols\u002Fcompose-pr.md",{"path":54,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":55,"sourceHtmlUrl":56},"protocols\u002Fprivacy-judge-spec.md",6421,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fprotocols\u002Fprivacy-judge-spec.md",{"path":58,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":59,"sourceHtmlUrl":60},"protocols\u002Frecall.md",3815,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fprotocols\u002Frecall.md",{"path":62,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":63,"sourceHtmlUrl":64},"README.md",1805,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002FREADME.md",{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Fmemory-grep.sh","SCRIPT","脚本",47464,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fscripts\u002Fmemory-grep.sh",{"path":72,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":73,"sourceHtmlUrl":74},"scripts\u002Fprivacy-scan.py",7534,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fscripts\u002Fprivacy-scan.py",{"path":76,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":77,"sourceHtmlUrl":78},"standards\u002Fmemory-conventions.md",15730,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Fstandards\u002Fmemory-conventions.md",{"path":80,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":81,"sourceHtmlUrl":82},"test-prompts.json",2294,"https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fgit-memory\u002Ftest-prompts.json",[84,91,98,105,112,119],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":18},"01M7CQNVD2W5MP7PKX3MD0YSQE","using-loom-workflow","Route broad Loom workflow-tool requests to the right skill. Use when the request spans tools or the right workflow skill is unclear.","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fusing-loom-workflow\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":18},"01M7CQNNTPVG52CS2C6Q9HJ2E2","skill-retro","Review selected past or current Claude Code or Codex sessions against readable target skill sources and related integration. Produce source-linked proposals without applying changes.","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fskill-retro\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":18},"01M7CQNHR1BDKJ63CC4R1RY9GE","recap-state","Give an in-session recap of current progress for 'where were we', '我們剛剛在幹嘛', '剛剛講到哪', '我跟丟了', 'ちょっと振り返って', '今どこだっけ', or '振り返り'. The built-in \u002Frecap is an away-summary.","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Frecap-state\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":18},"01M7CQNE1A9272WPRE1MM3JRFK","project-retro","Review one current or specified coding session for source-backed project\ndevelopment improvements. Use for a requested retrospective, including\n振り返り seeking improvements; not for current-progress recaps or recurring skill mining.","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Fproject-retro\u002FSKILL.md",[],[],{"skillNo":113,"name":114,"nameZh":6,"description":115,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":116,"categories":117,"keywords":118,"publishedAt":18},"01M7CQNACJB36T0A039HT89P38","loom-visualization","Show comparisons, flows, decisions, states or reasoning chains as tables, ASCII or Mermaid in coding chat, including when a Loom station reports to the user (intent restatement, choices, acceptance test results); not Obsidian notes.","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Floom-visualization\u002FSKILL.md",[],[],{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":18},"01M7CQN7TVS0XF44JXQBWA32Z1","loom-memory","Record repo-wide lessons when saving to your own memory or asked to remember; recall, reconcile, or forget them when a task needs it.","https:\u002F\u002Fgithub.com\u002Fkouko\u002Floom-plugins\u002Fblob\u002Fmain\u002Floom-workflow\u002Fskills\u002Floom-memory\u002FSKILL.md",[],[]]