Agent Wallclock is local-only software: CLI and MCP read your system clock and a JSON ledger under ~/.agent-wallclock/. The project itself makes no outbound network calls.

Security reports should focus on:

• Local privilege escalation or unsafe file permissions
• Store corruption, lock handling, or data loss bugs with security impact
• MCP trust boundary issues (unexpected writes, path traversal in store paths)
• Supply-chain concerns in dependencies (@modelcontextprotocol/sdk, zod, etc.)

Out of scope for this repo:

• Host-side paste/upload of briefings to Claude, ChatGPT, or Cursor (that is the host’s privacy model)
• Social engineering via model prompts (mitigated by adapters, not enforced in code)

Preferred: GitHub Security Advisories on TelivityAI/agent-wallclock.

Alternative: Open a private security issue if advisories are unavailable — title prefix [security], minimal reproduction, no secrets in the body.

Include:

1. Affected version or commit
2. Steps to reproduce locally
3. Impact assessment (confidentiality / integrity / availability on the user’s machine)
4. Suggested fix if you have one

Maintainers aim to acknowledge reports within 7 days and provide a triage update within 14 days. Critical local data-loss or RCE-class issues are prioritized.

Please do not open public issues with exploit details before a fix is available. We will coordinate disclosure and credit if you wish.

• MCP write tools (start_effort, log_session) are off unless AGENT_WALLCLOCK_WRITES=1
• Store directory 0700, store file 0600 where the OS supports it
• No telemetry or auto-update channels in the core product