agentleFS
MaGe1993MaGe1993/Claude-Code-Base-Harness/.claude/skills/audit-legal/SKILL.md

audit-legal

Skill1 starsChanged 5 months agoMITView on GitHub

Legal compliance triage — EU AI Act, GDPR, and German financial, website, employment, and consulting law. Loads obligation clusters by profile, auto-updates from source. Not legal advice.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 3 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read-only on project source files — no code edits, no application writes.

    line 34
  2. Distinctive:

    Never make a final compliance determination — produce triage findings for human legal review only.

    line 35
  3. Very specific:

    Write/Edit only the obligation-cluster assets under .claude/skills/audit-legal/assets/, and only during a currency check (5§) or a …

    line 36
  4. Very specific:

    Read before loading any cluster: CLAUDE.md, PROJECT.md, README.md.

    line 43
  5. Very specific:

    If a needed cluster has no asset file: WebFetch the stable URLs in assets/reference.md and derive findings from the fetched text.

    line 66
  6. Very specific:

    Read # updated: from the asset header (format YYYY-MM-DD).

    line 72
  7. Very specific:

    Read # check-interval-days: from the header.

    line 73
  8. Distinctive:

    If stale (threshold exceeded), perform a source fetch:

    line 75
  9. Very specific:

    If discrepancies found: rewrite the affected passages in the asset.

    line 81
  10. Very specific:

    If none: set # updated: to today (confirms currency).

    line 83
  11. Very specific:

    If a WebFetch fails: record "Check failed — verify manually".

    line 84

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Audit — Legal

1§ Purpose

The process schema for legal-compliance triage of a software project. Establish the project's legal profile, load the matching obligation clusters, keep them current against authoritative sources, and produce article-level findings for human legal review. Triage only — never a final determination.

The report format is the calling agent's standard (see ag-cqo). The disclaimer in 2§, however, is part of this skill and is mandatory: any output derived from this skill must carry it verbatim. Covers EU AI Act, GDPR, German financial law, website/app law, employment law, and consulting law.

2§ Mandatory disclaimer

Any output that uses this skill must begin with the following block, verbatim, without exception:

LEGAL NOTICE: This output is generated by an AI tool for informational and
preliminary triage purposes only. It does not constitute legal advice
(Rechtsberatung) and must not be relied upon as such. No attorney-client or
advisory relationship is created by use of this tool. Dr. Marius Gensler, as
creator and distributor of this tool, expressly disclaims all liability arising
from or related to any reliance on, or action taken or not taken based on, this
output. Regulations change; always verify currency with authoritative sources.
Any compliance determination with legal consequences requires review and
confirmation by a qualified lawyer admitted in the applicable jurisdiction.

3§ Hard constraints

  • Read-only on project source files — no code edits, no application writes.
  • Never make a final compliance determination — produce triage findings for human legal review only.
  • Write/Edit only the obligation-cluster assets under .claude/skills/audit-legal/assets/, and only during a currency check (5§) or a framework extension (7§). Never create assets without an explicit user request.
  • Derive every finding from a loaded cluster or fetched source text — never from general knowledge.
  • Cite the specific framework and article or section for every finding.

4§ Project profile

Read before loading any cluster: CLAUDE.md, PROJECT.md, README.md. Determine and record:

  • AI components: none / internal tooling / user-facing AI / GPAI model development
  • Personal data: processed / not processed / unclear
  • User base: internal employees / B2B / B2C / public
  • Financial features: none / informational / personalized / trading / custody
  • Deployment context: website / app / SaaS / enterprise tool / API

5§ Obligation clusters

Assets live in .claude/skills/audit-legal/assets/. Load only the clusters that match the profile; record which were loaded and which were skipped, with the reason.

Asset fileCoversLoad when
ai-act-risk-classification.mdRisk tiers, classification criteria, per-tier obligations (EU-wide)Any AI component present
ai-act-transparency.mdArt. 50 disclosure requirements for limited-risk AI (EU-wide)User-facing AI interaction present
ai-act-gpai.mdArt. 53–56 GPAI provider obligations (EU-wide)Project develops or deploys a GPAI model
gdpr-data-processing.mdArt. 5, 6, 9, 22, 25, 28 core obligations (EU-wide)Project processes personal data
gdpr-dpia-triggers.mdArt. 35 DPIA triggers, WP 248 indicator checklist (EU-wide)High-risk data processing suspected
financial-trading.mdGerman financial law: BaFin licensing (KWG/WpIG/ZAG/KAGB/GewO §34f/d/h), advice vs. information distinction, GwG, MiCA/MiFID II as applicable in GermanyProject involves financial products, trading, or crypto
website-app-compliance.mdGerman internet law: Impressumspflicht (DDG §5), TTDSG cookies, AGB-Recht (BGB §§305–310), Fernabsatzrecht, UWG, UrhG, BFSG accessibilityProject is a public website, app, or SaaS
employment-law.mdGerman Arbeitsrecht: KSchG, ArbZG, MiLoG, BUrlG, TzBfG, AGG, BetrVG, SGB, Scheinselbstständigkeit (§611a BGB)Project employs people or uses freelancers in Germany
consulting-law.mdGerman Beratungsrecht: RDG, StBerG, WPO, Heilberufsrecht, HOAI, Beraterhaftung (BGB §§280/311/675)Project offers or simulates advisory services

If a needed cluster has no asset file: WebFetch the stable URLs in assets/reference.md and derive findings from the fetched text. Tell the user which asset could be created for future reuse — do not create it automatically.

6§ Currency check

Immediately after loading each asset, check whether it is current:

  1. Read # updated: from the asset header (format YYYY-MM-DD).
  2. Read # check-interval-days: from the header; default to 90 if absent.
  3. Compute days since updated. If within interval: record "Current (within interval)" and proceed.
  4. If stale (threshold exceeded), perform a source fetch:
    • WebFetch each URL under # sources: in the asset header. Prioritize the URL most likely to reflect the current state of the specific law or threshold.
    • Identify verifiable data points subject to change: numerical thresholds and amounts, enforcement dates, article numbering, regulatory tier boundaries, specific requirements.
    • Compare each against the fetched content. Note any mismatch.
    • If discrepancies found: rewrite the affected passages in the asset; set # updated: to today; record the change as "Updated: [old] → [new]".
    • If none: set # updated: to today (confirms currency); record "Verified — no changes".
  5. If a WebFetch fails: record "Check failed — verify manually"; do not update # updated:; use the asset as-is with a staleness note.
  6. Not-loaded assets: record "Not loaded".

7§ Analyze and extend

Scan project source, documentation, and configuration for compliance signals:

  • Source code: data processing patterns, AI interaction without disclosure, automated decisions, financial personalization, cookie implementation, employee monitoring
  • Documentation: technical docs, transparency notices, terms, DPIA records, DPAs
  • Configuration: logging, third-party API calls, retention settings, tracking

Cross-reference each signal against the loaded clusters. Record each finding with its framework, article, and an exact project location.

To add a framework (user-requested only): WebFetch the source from assets/reference.md; extract obligations into a new asset following the template there; write it to .claude/skills/audit-legal/assets/<framework>.md; add a row to the 5§ table.

8§ Associated documents

  • The obligation clusters in assets/, loaded per the 5§ table.
  • [assets/reference.md] — stable source URLs and the asset-file template for currency checks and extension.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.