Thanks for taking the time to look at paige's security surface.

Please do not open a public issue for a vulnerability.

Email the maintainer privately at the address linked from the project's GitHub profile, or open a private security advisory on the repository (GitHub → Security → Report a vulnerability).

What to include:

• A description of the issue and the steps to reproduce.
• The paige version (./scripts/prod.sh status shows the wheel).
• Whether the issue is exploitable today, or only under specific configuration.

What to expect:

• Acknowledgement within a few business days.
• A fix or mitigation timeline once the issue is confirmed.
• Credit in the release notes if you'd like (let us know in your report).

paige's threat surface is small but non-trivial:

• IM credentials — PAIGE_FEISHU_APP_ID / PAIGE_FEISHU_APP_SECRET live in ~/.paige/.env and grant outbound message + card access. The repo's prod.sh strips outbound-proxy env vars before launch.
• tmux pane control — paige sends keystrokes into the bound pane. A vulnerability that lets an unauthorised user trigger send_keys is effectively remote code execution on the host (whatever claude will run).
• Access control — PAIGE_ALLOWED_USERS / PAIGE_ADMIN_USERS gate who can interact with the bot. Default is open to anyone the bot can hear from; setting a real allow-list is strongly recommended on any shared deploy.

• Findings against the IM backend itself (Lark / Feishu) — please report those to the backend vendor.
• Findings against claude (the agent paige bridges to) — please report those to Anthropic.
• Issues that require root or local-shell access on the host paige runs on (already game-over by other means).

paige is pre-1.0 and ships from main. Security fixes target the latest released wheel; no long-term support branches yet.