Loop Engineer is distributed as source. Security fixes land on the latest release and on main. Pin a tag if you need a stable point.

Do not open a public issue for a security problem.

Use GitHub's private vulnerability reporting: Security → Report a vulnerability on the repository (https://github.com/SollanSystems/loop-engineer/security/advisories/new).

Please include reproduction steps and the affected file(s)/skill(s). You'll get an acknowledgement, and a fix or mitigation plan once the report is triaged.

Loop Engineer designs and operates autonomous agent loops, so a few of its own guarantees are security-relevant — issues in these areas are in scope:

• Verifier-gaming defenses. The anti-cheat scanner (scripts/anticheat_scan.py) and held-out gate (scripts/holdout_gate.py) exist to catch a loop editing its own tests/fixtures/spec to manufacture a pass. A bypass of these is a security issue.
• Approval boundary. The contract pauses at side-effect boundaries (destructive commands, secrets, production mutation, money movement). A path that lets a loop cross one without the approval gate is in scope.
• Plan-then-execute / prompt injection. The skills instruct loops to treat untrusted content (web, scraped docs, tool output) as data, never as instructions. A documented pattern that violates this is in scope.

Out of scope: vulnerabilities in your loop's task code, in optional third-party integrations, or in Claude Code itself (report those upstream).

This repository must contain no secrets. The no-secrets structural check (scripts/self_eval.py) scans for secret-shaped literals on every run; if you find one that slipped through, report it via the private channel above.