원클릭으로
check-requirements-against-code
Check requirements docs against implementation and report discrepancies
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Check requirements docs against implementation and report discrepancies
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Create a new bug ticket in Hot Sheet
Create a new feature ticket in Hot Sheet
Create a new investigation ticket in Hot Sheet
Create a new issue ticket in Hot Sheet
Create a new req change ticket in Hot Sheet
Create a new task ticket in Hot Sheet
| name | check-requirements-against-code |
| description | Check requirements docs against implementation and report discrepancies |
| allowed-tools | Read, Grep, Glob, Bash, Agent, Edit, Write |
Comprehensively compare the requirements documents in docs/ against the actual implementation. Also verify that the AI summary docs (docs/ai/code-summary.md, docs/ai/requirements-summary.md, docs/ai/usage-guide.md), the human-facing orientation one-pager (docs/orientation.md), and CLAUDE.md are in sync with both the requirements docs and the code. Generate a report with recommendations and questions about any discrepancies.
Read all requirements documents in docs/. Note every stated requirement, behavior, and constraint. The numbered set is 1-overview.md through 9-live-demo.md. If new numbered docs have been added since this skill was written, include them too. Also read README.md and llms.txt at repo root since both enumerate the public API and link to the docs — they drift the same way.
For each requirement, verify it against the implementation:
src/)Check for undocumented features: Scan src/ for significant functionality not covered by any requirements doc. These are public exports / observable behaviours that should either be documented in docs/ and docs/ai/usage-guide.md, or questioned.
Check for stale documentation: Requirements that describe behaviour that no longer exists or has changed.
Verify CLAUDE.md completeness:
docs/ appears in CLAUDE.md's "Requirements Documentation → Reading order" list. Report any docs present on disk but missing, or listed but missing from disk.src/ (including src/utils/*). Report any drift.src/index.ts. Report any export missing from CLAUDE.md or vice versa.npm run test* script that exists in package.json.vitest.config.ts.Synchronize docs/ai/code-summary.md: Open the file and confirm each section still matches the current codebase. Flag any inaccuracy, then update the file in place. Check specifically:
src/, tests/, docs/, and the repo root (use Glob/ls to verify).src/index.ts.npm run build actually emits (entries, chunks, source maps, .d.ts files).Synchronize docs/ai/requirements-summary.md: Open the file and confirm each entry still matches its source doc. Flag and update:
Synchronize docs/ai/usage-guide.md: Open the file and confirm:
import { … } from 'kerfjs' block lists every public export.Synchronize docs/orientation.md (KF-179): Open the file and confirm each section still matches the current codebase. This doc is hard-capped at 500 words (wc -w docs/orientation.md ≤ 500); preserve the cap when editing — trim elsewhere if you need to add. Check specifically:
src/ (and src/utils/). One bullet per primary export; add new files as they land in src/.docs/diagrams/render-pipeline.svg still reflects the actual pipeline (signal write → render fn → SafeHtml → morph + list reconciler → live DOM). If the pipeline shape changes — e.g. a new pre-morph stage, a new segment kind — the SVG needs an edit and the surrounding paragraph needs a rewrite.store.ts:REGISTRY and each.ts:context), the coverage thresholds, and the Hot Sheet KF- prefix convention. Coverage thresholds must match vitest.config.ts.wc -w docs/orientation.md and confirm ≤ 500 afterwards. If you cannot stay under the cap without losing essential information, surface that in the report instead of silently bloating the doc.Final consistency pass: Make sure CLAUDE.md, README.md, llms.txt, docs/orientation.md, docs/ai/code-summary.md, docs/ai/requirements-summary.md, and docs/ai/usage-guide.md agree with each other and with the source docs / code. Any disagreement gets resolved in favor of the source doc / code, and the summaries and CLAUDE.md are updated accordingly. The single most common drift in this project is when a new public export is added (e.g. isSafeHtml) and only src/index.ts + docs/8-api-reference.md are updated — the AI summaries and root-level files lag. Look for that pattern explicitly.
For each discrepancy:
missing (doc says X, code doesn't do X) | different (doc says X, code does Y) | undocumented (code does X, no doc mentions it) | stale (doc says X, feature was removed/changed)For doc-vs-doc drift (e.g., the same export listed in CLAUDE.md but missing from llms.txt), recommend updating both to match the canonical source — usually src/index.ts for the export list, the numbered docs for behaviour, and vitest.config.ts for coverage thresholds.
src/ but missing from CLAUDE.md "Source layout"src/index.ts but missing from CLAUDE.md import blocknpm run test* scripts in package.json but missing from CLAUDE.md "Testing" sectionvitest.config.ts that don't match the CLAUDE.md "Coverage thresholds" sentencedocs/ai/code-summary.md — list of sections edited and why (or "no changes needed")docs/ai/requirements-summary.md — list of entries edited and why (or "no changes needed")docs/ai/usage-guide.md — list of changes (or "no changes needed")docs/orientation.md — list of changes (or "no changes needed"). Always include the post-edit word count (wc -w) to confirm the 500-word cap is preserved.README.md / llms.txt — list of changes (these are usually quick one-liner updates to the API enumeration)List any ambiguous requirements where the implementation had to make a judgment call, and ask whether the current behavior is correct.