colleague-skill

This skill is coherent with its stated goal (automatically collecting workplace messages/docs and building a Persona + Work Skill), but there are important mismatches and privacy implications you should consider before installing: - Metadata vs reality: The registry claims 'no required env vars' and 'instruction-only', but the package contains scripts that require app_id/app_secret, OAuth/user tokens, and will save config to ~/.colleague-skill and write generated skills to disk. Treat those as required side effects. - Credentials & least privilege: If you try it, create dedicated, limited-scope app credentials (not your org admin keys). For Feishu/DingTalk/Slack, grant only the minimal scopes needed and prefer short-lived tokens. Do not reuse high-privilege or organization-wide secrets. - Privacy & consent: The tool harvests private chats and documents (it even instructs sending messages to obtain chat IDs). Only collect data you are authorized to access and ensure affected people consent where required by policy or law. - Storage & cleanup: The tool persists tokens and session data in ~/.colleague-skill and writes generated Skills to ./colleagues (or configured base dir). Inspect and delete these files if you stop using the skill; consider running in an isolated environment or disposable VM/container. - Code review: Because the skill executes network requests and can run Playwright/browser scraping, review the included Python scripts (tools/*.py) yourself or run them in a sandbox. There were no static malware signals, but the scripts perform sensitive operations by design. - Safer alternatives: If you only want to generate a persona from manual description or uploaded documents, avoid supplying OAuth tokens or enabling automatic collectors; use manual upload/paste mode instead. If you want help with specific mitigations (e.g., checklist to create limited-scope Feishu app, or commands to sandbox the skill), tell me which platform you plan to run it on and I can provide step-by-step guidance.