Forest-Project-Lab
UserStop stale specs from becoming code. A Markdown context-engineering blueprint for LLM coding agents: status, provenance, dependencies, and freshness.
Categories
Indexed Skills (9)
change-impact
Runs the 14-step change flow for a proposed change: traces dependencies to enumerate every document, implementation, and test affected, then drives the updates in the mandated order — decision (ADR) first, then current spec (SPEC), then implementation, then test, then LLM context, then deprecation cleanup — keeping "what changed" (CHANGE/git) separate from "why decided" (ADR). Grades the flow by whether the change contains a decision, never by its size: a change with no decision (typo, wording, formatting, link repair, projection re-render) keeps only the impact enumeration and the update order, and needs no CHANGE/IMPACT/ADR. Use this skill when the user wants to "assess the impact of a change", "what does changing X break", "plan a change", "run the change flow", "trace impact", "what needs updating if I change this spec/ICD", "影響範囲を洗い出して", "この変更で何が壊れるか", "変更フローを回して", "この仕様を変えたらどこを直すか", or "propose and roll out a change".
doc-author
Creates and updates typed governance documents — SPEC, REQ, ADR, DATA, API, IMPL, TEST, DECIDED, NONGOAL, WATCH, RESEARCH, ARCHIVE, and ICD — placing each in the correct location with correct frontmatter (id matching filename, type, domain, status, owner, dates, sources), and generating the domain folder and its layer subfolders the first time a document of that type is written. Use this skill when the user wants to "write a spec", "add an ADR", "create an ICD", "author a requirement", "add a doc", "document this decision", "record a data model", "start a new domain", "仕様を書いて", "ADR を起こして", "ICD を作って", "この決定を記録して", "要求を文書にして", or says "I need a SPEC/REQ/ADR/TEST for ...".
doc-review
Reviews a document's prose and positioning against the writing norms: runs the term checker alongside (unapproved terms, banned synonyms, banned calque expressions, undefined jargon), and judges calque the linter cannot catch using the back-translation tell (translate the suspect phrase back to English; if it lands on an English idiom, it is calque). On a defined cadence it also checks for missing canonical_for, off-dictionary translationese, and semantic duplication. Use this skill when the user wants to "review this doc", "check the writing", "proofread", "review terminology", "check for calque / 訳語臭", "is this prose clear", or runs the periodic "docs review / 定例レビュー".
docs-curate
Curates the document set one item at a time: inspects, then merges duplicate facts / demotes one step / or deletes, always checking reverse references first (zero current dependents before any demotion), re-rendering the projections (Overview, ICD-index, Context Map) from the model afterward, and shrinking the always-injected set whenever it exceeds the injection cap by consolidating facts and clearing expired review_by items. Use this skill when the user wants to "clean up the docs", "curate", "remove orphans", "merge duplicate facts", "the context is too big / over the cap", "shrink the always-set", "deprecate this doc", "re-render the projections", or runs the periodic "docs cleanup / 定例整理".
docs-system-init
Sets up the document governance system in a repository: creates the minimal _system layer (glossary, decided-facts, non-goals, overview projection) and the root entry points (CLAUDE.md, AGENTS.md), and wires up the guards and linter without overwriting anything that already exists. Use this skill when the user wants to "initialize docs", "set up the documentation system", "bootstrap docs governance", "set up doctrine", "scaffold _system", "文書統治を導入して", "docs を初期化して", "doctrine を設定して", "このリポジトリで使い始めたい", or asks "how do I start using this plugin" / "set up doc governance in this repo".
assurance-loop
Doctrine 継続保証キャンペーンの運転手順。保証ループを回す・DISCOVER/CHALLENGE を実行する・ 規範カタログや網羅台帳を進める・カオス試験や故障注入を計画する・事象を分析する・ 「保証キャンペーンを進めて」「保証ループを回して」「観点を創出して」「独立検証して」と 言われたときに使う。セッションを跨いだ再開は台帳と状態機械が持ち、会話の記憶に依存しない。 開発専用（assurance/ レーン）。配布物・配布 Skill には触れない。
system-map-draft
対象リポジトリから系の意味モデルの下書き(MODEL 型の .md。proposed 限定の候補)を統治木の中へ起草する。全ての値に出所(読んだ場所・確認日・判定)を付け、構造はリンタが、出所は map-draft-check.py が機械検証する。確定(confirmed への昇格と status の一押し)は人が行う。"draft a system map"・"generate a proposed system model"・「System Map の下書きを起草して」・「意味モデルの下書きを作って」と言われたときに使う。
llm-context-pack
Builds the minimal context for a specific task: gathers only the fewest documents whose coverage satisfies the task's requirements, excludes everything marked llm_context: never (RESEARCH, ARCHIVE), and shows the provenance (source document) of each fact so near-vocabulary documents are not confused. Use this skill when the user wants to "build context for this task", "pack the minimal context", "what docs do I need for X", "assemble LLM context", "give me the smallest set of docs to work on Y", "このタスクに要る文書だけ集めて", "最少の文脈を組んで", "どの文書を読めばよいか", or "prepare context for the agent".
regression-guard
Guards against regressions: prevents reviving deprecated approaches or re-adopting a withdrawn decision by cross-checking the proposed change against Decided Facts (DECIDED) and the Regression Watchlist (WATCH), flagging any change that contradicts a current decided fact or reintroduces something the watchlist says must not come back. Use this skill when the user wants to "check for regressions", "are we re-adding something we removed", "does this contradict a decided fact", "check the watchlist", "guard against backsliding", "退行していないか見て", "廃止した方針を復活させていないか", "決定と矛盾しないか確かめて", "戻してはならない一覧と突き合わせて", or before re-introducing a previously-rejected approach.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.