larkcowork/lark-cowork-plugins/permission-audit
Claude Code and agent skill by larkcowork/lark-cowork-plugins
- community
Checking that the SKILL.md file is still in the repository.
Read-only scan of Drive/Doc/Wiki/Base for risky permissions (public, external, PII). Triggers "permission audit", "quét quyền", "PII leak".
Install
npx vskill@latest install larkcowork/lark-cowork-plugins/permission-auditSKILL.md
permission-audit
Read-only scan → one severity-ranked report. No auto-fix, no auto-revoke, no notification.
Verbs (the installed lark-* skills): drive +search (--doc-types/--folder-tokens/--space-ids/--creator-ids — inventory files; NO permission flag), docs +fetch (content for PII scan), base +table-list. Raw CLI has no verb that returns per-file share/permission state — get permission metadata via the atomic lark-drive skill (or note the gap in the report). Atomic shapes + token flags: connectors/LARK-RECIPES.md.
Workflow
- Scope. Default: all Drive/Wiki/Base the user admins. Narrow via
--folder-tokens, --space-ids, --creator-ids. >1000 items → warn + suggest batching.
- Inventory. `drive +search --doc-types doc,sheet,bitable,file
--folder-tokens <t> --jq '.data' (peek shape; + --space-ids for wiki; needs search:docs:read scope). Permission state → atomic lark-drive` skill per item (batch). Cache for the session — no re-fetch.
- Classify by memory rule:
public-internet→ HIGH (CRIT if
strict policy) · anyone-with-link → MED · external_users >0 → MED (HIGH if folder is Confidential/HR/Finance) · inactive_owner → MED.
- PII/DLP scan on HIGH+ items. `docs +fetch --doc <url>
--limit 1 (cap content), regex memory/regulated-data-types.md: VN CCCD \b\d{9}\b|\b\d{12}\b, Luhn-valid 13-19 digit cards, email+phone batch (>10/file), API keys (sk-, ghp_, AKIA, xoxb-`). Each match: severity +1, mask 50% digits.
- Dedupe vs `incident-log.md` — already-raised → mark RECURRENT.
- Format report — SUMMARY (counts by severity) + CRIT/HIGH
findings (token/url/owner/share/PII/policy ref/action/status) + OWNER EXPOSURE top 5 + RECOMMENDED ACTIONS. Ends "read-only output, no changes made."
Hard rules
- READ-ONLY — never call
+update/+apply-permission/+delete.
"Fix luôn" → redirect to admin Lark UI or offer a task (confirm).
- Mask sensitive data —
4111-****-****-1234, CCCD 3+3 digits only. - Don't dump file contents — report token + URL only.
- Scan only where user has admin.
403→ skip silent, no brute force. - Severity from policy — every finding refs a
memory/policies.md§. - Report to admin privately — 1:1 chat, never wide group.
Memory
Required: policies.md (sharing rules + severity map), regulated-data-types.md (PII patterns). Recommended: incident-log.md (dedupe recurrent).
Empty memory → conservative defaults (anyone-with-link = HIGH) + warning to set up policies.md.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckinggovernance/skills/permission-audit/SKILL.md
Popularity
- 7-day trend score
- 5
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/larkcowork/lark-cowork-plugins/permission-audit)