Audit Skill for Claude
Run a corpus-scale, STATS-ONLY PII audit over a folder of session transcripts LOCALLY and produce an aggregate report — counts by type and by layer, the per-session redaction-rate distribution, document lengths, and a coarse residual proxy. Use when the user says "audit my sessions", "scan folder for PII", "how much PII across these transcripts", "PII stats for my corpus", "is my redaction holding at scale", or points at a directory of transcripts and asks how much personal data it contains. Fully local — raw text never leaves the machine; the report carries ZERO PII values, transcript substrings, or filenames (only anonymized own-NN ids and counts), so the aggregates are safe to surface. Run it on a RED (raw) corpus to size the PII, or on a GREEN (already-redacted) corpus to check residual leakage.
Install
one command, takes just this skill from the repositorynpx skills add https://github.com/glebis/claude-skills --skill auditWhat comes with it
17 648 bytes besides the instructionThe instruction itself
7 sections, as written by the authorconfide:audit — corpus-scale, stats-only PII audit
Measure how much PII lives across a whole folder of sessions, without ever exposing any
of it. The audit runs the layered LOCAL detector stack from shared/confide_core.py
(regex → Natasha → local LLM) over each file and emits only aggregates. This mirrors
the real_session_eval privacy contract: read text only in-process, emit counts.
Privacy invariants (do not violate)
- Local-only. No cloud APIs. Raw transcript text never leaves the machine.
- Stats-only output. The report (markdown + json + optional HTML) contains ONLY
counts and rates — never a transcript substring, never a detected PII value.
- No filenames. Per-file rows are keyed by anonymized ids
own-00,own-01, …
The original path/name is never written or printed. On an unreadable file, only the
index + exception class name is recorded.
- Safe to surface. Because it is counts-only, the aggregate report can be shared
with a cloud agent or pasted into a chat. The PII stays on the machine.
What it reports
n_files, total / mean / min / max document charsspans_by_type(PERSON, EMAIL, PHONE, DATE, …) andspans_by_layer(regex / natasha / llm)overall_redaction_rateplus the per-session redaction-rate distribution
(min / median / mean / max)
- a coarse residual proxy: spans still detectable after redaction — ~0 on a clean RED
corpus, a leakage signal on a GREEN corpus.
Run it
Point it at a folder (recurses, processes every .md/.txt; skips confide's own
*.green.md / *.stats.json outputs):
python3 skills/audit/scripts/audit.py FOLDER
Options:
--list paths.txt— also/instead audit absolute paths listed one per line.--layers regex,natasha,llm— choose detection layers (default from config).
Use --layers regex for a fully offline, deterministic pass (no models/network).
--out report.md— report path; areport.jsonsibling is written alongside.--html— also write a Tufte-ish dashboard (report.html, counts only).
Writes the markdown + json report (and optional HTML) and prints the aggregate summary —
all counts only.
RED vs GREEN
- RED (raw) corpus: sizes the PII problem before any redaction.
- GREEN (redacted) corpus: the residual proxy and remaining
spans_by_typetell you
whether redaction is holding at scale.
After running
- Report the aggregate summary (file count, span totals by type/layer, redaction-rate
distribution, residual proxy) — never paste PII.
- If residual is non-trivial on a GREEN corpus, point the user at confide:anon to
re-redact and confide:red to probe re-identification risk.
Setup
Layer availability (Natasha, local LLM via Ollama) comes from config — run
confide:setup if they aren't installed. --layers regex always works offline.
Other skills for the same job
different authors, same section of the catalogue
> Document Quality Check skill for Datasite deal rooms. Use this skill whenever a deal team wants to audit document quality before going live to buyers. Triggers files", "check for blank documents", "PII check", "redaction review", "find corrupted files", "document audit", "quality check the data room", "are there any blank or broken files", "check for unredacted personal data", or any request to verify that documents in the data room are complete, accessible, and safe to share. Use this skill proactively before a data room goes live. Do not use for renaming files (use smart-file-renaming) or for identifying missing sections (use gap-analysis).
> Data Room Gap Analysis skill for Datasite deal rooms. Use this skill whenever a sell-side deal team wants to audit what is missing, sparse, or incomplete in their "what's missing from the data room", "check the data room coverage", "flag empty folders", "what haven't we uploaded yet", "data room readiness check", "find gaps before we go live", "are all the contracts in there", "check we have everything", or any request to assess completeness of the data room by section. Use this skill proactively whenever a deal team is preparing to launch a data room and wants to know what still needs to be uploaded or organised. Do not use for document quality issues such as PII or redaction (use document-quality-check), or for drafting Q&A responses (use bulk-qa-answers).
> Risk Analysis Audit skill for Datasite deal rooms. Use this skill whenever a sell-side deal team wants to audit, review, or flag risks across a data room before going live. "what are the risks in this deal", "audit the data room", "risk analysis", "flag issues before we go live", "what should we fix before launch", or any request to analyse deal risk by workstream (Tax, Finance, Legal, HR, IP, Commercial, Regulatory, ESG). Use this skill proactively whenever the user is preparing a data room for launch and wants a structured view of what might concern a buyer. Do not use for document quality issues like PII or redaction (use document-quality-check), or for identifying missing sections (use gap-analysis).
> Smart File Renaming skill for Datasite deal rooms. Use this skill whenever a deal team wants to standardise document names, clean up scanned file names, normalise naming across similar document types, or improve the professionalism of the data names", "standardise naming", "the file names are a mess", "fix the document names", "rename scanned documents", "make the naming consistent", "tidy up the data room", or any request to improve, clean, or normalise document naming across a Datasite project. Never apply any rename without explicit user confirmation. Do not use for document quality or PII checks — use document-quality-check for that. Never rename files without explicit user confirmation.
Penetration test and red team report writing methodology. Covers executive summary structuring (risk-led narrative for non-technical readers), technical finding format (title, severity, affected scope, narrative, reproduction steps, impact, remediation, references), CVSS v3.1 / v4.0 scoring with vector justification, OWASP risk rating, evidence hygiene (redacting credentials, hashing client data, time-stamping every action), screenshot and PoC artifact management, finding chain narratives, scope/limitations/assumptions documentation, retest evidence and remediation tracking, deliverable formats (PDF, DOCX, HTML, JSON for SIEM ingestion), client-customer-deliverable separation, and common report mistakes (over-CVSSing, undermining the triager, missing the 'so what'). Use at the end of an engagement when authoring a deliverable, when restructuring a draft for executive readability, or when establishing a reusable report template for a consulting practice.
> for GDPR violations, (2) drafting GDPR-compliant documents such as privacy policies, Data Processing Agreements (DPAs), and consent notices, (3) answering GDPR compliance questions with authoritative article citations, and (4) reviewing data flows and PII handling practices. Use this skill whenever the user mentions GDPR, data protection, privacy compliance, lawful basis, data subject rights, DPA, privacy notices, consent management, data breaches, DPIAs, controller/ processor relationships, cross-border data transfers, or any EU/UK data privacy topic. Also trigger for questions like "is this GDPR compliant?", "how do I handle personal data?", "what does a privacy policy need?", or any request involving PII, personal data, or data retention in a regulatory context.
> Expert HIPAA compliance assistant for healthcare and software contexts. Use this skill whenever the user mentions HIPAA, PHI (Protected Health Information), ePHI, covered entities, business associates, healthcare data privacy, medical records, health information security, BAA (Business Associate Agreements), or any compliance review involving patient data. Also trigger for requests to draft privacy notices, HIPAA policies, consent forms, security risk assessments, or breach notification letters. Use for developers building healthcare software who need technical safeguard guidance (encryption, access controls, audit logs), compliance officers reviewing documents or procedures, and anyone asking "is this HIPAA compliant?" or "what does HIPAA require for X?". When in doubt about whether a healthcare or data privacy question falls under this skill — use it.
Expert GDPR compliance assistant covering all four core workflows: (1) auditing code and systems for GDPR violations, (2) drafting GDPR-compliant documents such as privacy policies, Data Processing Agreements (DPAs), and consent notices, (3) answering GDPR compliance questions with authoritative article citations, and (4) reviewing data flows and PII handling practices. Use this skill whenever the user mentions GDPR, data protection, privacy compliance, lawful basis, data subject rights, DPA, privacy notices, consent management, data breaches, DPIAs, controller/ processor relationships, cross-border data transfers, or any EU/UK data privacy topic. Also trigger for questions like \"is this GDPR compliant?\", \"how do I handle personal data?\", \"what does a privacy policy need?\", or any request involving PII, personal data, or data retention in a regulatory context.
How to use it
Take glebis/audit from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.