compliance-compiler

name

compliance-compiler

description

Install a per-repo compliance catalog + PRP-plan validator into the current repository. Parallel agents distil GDPR/DSGVO, SOC 2, and ISO 27001 into a tracked catalog of atomic constraints ("features"); a PostToolUse hook then checks every PRP plan (.claude/PRPs/plans/*.plan.md) against it. Trigger when the user says "compliance compiler", "install compliance compiler", "compliance catalog", "GDPR/SOC2/ISO catalog", "compliance einrichten", "validate plans against compliance", or wants a repo-local, self-checking compliance baseline.

Compliance Compiler — Install Skill

Installs the compliance-compiler engine into the current repo. Two halves:

1. Extraction — scripts/extract.py fans out ~30 parallel Claude Agent SDK agents (one per framework shard) that distil GDPR/DSGVO, SOC 2, and ISO 27001 into <catalog_dir>/catalog/{gdpr,soc2,iso27001}.json — atomic constraints with an applies_when predicate and a check.
2. Validation — a PostToolUse hook checks each PRP plan file (.claude/PRPs/plans/*.plan.md) as it is written: a fast structural precheck inline, plus a detached deep LLM report in <catalog_dir>/reports/.

The catalog always lives inside the repo, never under .claude/.

Authentication

Extraction and deep validation use the Claude Agent SDK, which needs ANTHROPIC_API_KEY (or CLAUDE_CODE_OAUTH_TOKEN). Subscription credentials are NOT sanctioned for third-party plugins — public/customer installs must set an API key. Install, scaffolding, hooks, and the inline structural precheck work without it.

Copyright note

ISO 27001 and SOC 2 standard text is copyrighted. The catalog stores official control identifiers + short titles + a paraphrased requirement, never verbatim text. Supply source text locally if you want higher-fidelity extraction.

Naming / collision note

Invoke this as neurawork-cc-harness:compliance-compiler. Its hooks are co- prefixed and it uses the PostToolUse event, so it coexists with the harness's knowledge-compiler and claudemd-lerner in one .claude/settings.json.

Phase A — Recon (read-only)

python3 "${CLAUDE_PLUGIN_ROOT}/engines/compliance-compiler/recon.py"

• status: NOT_A_GIT_REPO → stop; tell the user to run inside a git repo.
• existing_dir set → ADOPT (refresh) install; reuse that name.
• Note catalog_built, existing_hooks, timezone, clean.

Phase B — Ask

Use AskUserQuestion to confirm:

1. Catalog dir name — default compliance-base (or the detected existing_dir).
2. Frameworks — default all three (gdpr, soc2, iso27001); allow a subset.
3. Extract now? — offer only when the tree is clean; extraction makes API calls and runs ~30 agents.

Phase C — Execute

python3 "${CLAUDE_PLUGIN_ROOT}/engines/compliance-compiler/install.py" \
  --catalog-dir <NAME> [--extract]
uv sync --directory <NAME>

Then tell the user to commit <NAME>/ and .claude/settings.json. After install:

• Every PRP plan write is checked automatically.
• Manual extract: /neurawork-cc-harness:co-extract.
• Manual validate: /neurawork-cc-harness:co-validate <plan>.