Skip to content

Security: theRadicalSoftware/AIDesktopCompanion

Security

docs/SECURITY.md

Security

AI Desktop Companion is a local desktop tool. Treat pet packs, dropped files, and provider config as untrusted unless you created them.

Do Not Commit Secrets

Never commit:

  • API keys
  • Slack bot/user tokens
  • SSH private keys
  • personal Slack IDs or private channel IDs
  • local transcripts
  • runtime work outputs
  • .env files

Use environment variables or:

~/.config/ai-desktop-companion/secrets.json

Codex Work Drop

Dropped files are sent to Codex as prompt data, not shell-interpolated command strings. Review/explain/summarize defaults to read-only. Custom work can use workspace-write only when the user explicitly asks for edits.

Codex Approvals

The approval bridge answers the visible Codex terminal approval UI through X11/xdotool. It does not send approval waits as ordinary text replies.

GitHub Actions

GitHub actions run through local Git/SSH and block common unsafe states:

  • no Git repository
  • detached HEAD
  • dirty worktree
  • non-GitHub SSH remote
  • missing SSH key access

The bridge uses GIT_TERMINAL_PROMPT=0 and noninteractive SSH options so it does not hang on hidden credential prompts.

Public Pet Packs

Public pet packs should include only assets you are allowed to redistribute. Keep personal character assets in a private repo or ignored local folder.

There aren't any published security advisories