FFFF Overview · adeelahmad/aloop · GitHub
[go: up one dir, main page]

Skip to content

Security: adeelahmad/aloop

SECURITY.md

Security policy

aloop's gates are its security boundary: a hook decides what an agent may write, run or merge. So a way around a gate is a security bug, even when no secret is involved.

Report a vulnerability

Please report privately through GitHub: Report a vulnerability. If that form does not open for you, open an issue titled "Security contact request" with no details of the problem, and the maintainer will set up a private channel. Never put the details of a vulnerability in a public issue.

Include what you ran, what the gate should have blocked, and what happened instead. A failing test (a scripts/test/*.test.py case or a conformance fixture) is the best report: it becomes the regression test of the fix.

In scope

  • A worker writing outside its worktree or its task's files without a BLOCK.
  • A stop or a tool call that passes although its gate failed, crashed or was skipped.
  • A BLOCK downgraded to a WARN, or a gate that fails open.
  • A forged or edited record in the mission ledger (docs/agents/mission/ledger.jsonl) that a check accepts, or an approval that no human gave.
  • Anything that makes the supervisor commit on main, dispatch on another model, or write production code during a sprint.

Supported versions

Fixes land on master and ship in the next release. Only the latest release is supported.

There aren't any published security advisories

0