Skip to content

v0.9.31 — Safe Agent Command Execution (Phase A)

Latest
Latest

Choose a tag to compare

View all tags
@sourjya sourjya released this 17 Jul 12:36
· 17 commits to main since this release
v0.9.31
d56b0e2
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Safe Agent Command Execution — Phase A + STRIDE threat model

Security hardening at TracePulse's command-execution chokepoint. Agents route their shell through TracePulse and now get containment + sanitization that raw shell never had — with no loss of capability.

Added / changed

  • Redact run_and_watch raw_output before returning it to the agent, with [REDACTED:<n>] length hints so debugging value is preserved.
  • Scrub secret-shaped env vars from spawned commands (keeps ordinary config like NODE_ENV/CI); agent-declared env passes through, inherit_env opts out.
  • verify_mcp gets the same env-scrub + output-redaction parity.
  • Commands run outside the project root are surfaced (cwd_outside_project_root), not blocked.
  • env / inherit_env exposed in the run_and_watch + verify_mcp MCP schemas.

Security

  • AWS-Labs STRIDE threat model (docs/audits/security/THREAT_MODEL.md) — 14 threats, trust-boundary map.
  • Red-team / bypass test suite.

Full test suite: 1380 passing.

🤖 Generated with Claude Code

Assets 2
Loading