Skip to content

Security: HomericIntelligence/Athena

SECURITY.md

Security policy

Apply the ASD-STE100 technical-English policy to all English technical prose in this document.

Athena distributes plugins that contain instructions for artificial intelligence (AI) harnesses. Athena supplies security fixes for the latest tagged release and main.

Reporting

Send security reports privately to research@villmow.us. Include this information:

  • the affected revision;
  • the affected skill or manifest;
  • the reproduction steps;
  • the expected behavior;
  • the observed behavior; and
  • the disclosure status.

We aim to acknowledge reports no later than five business days after receipt.

Threat model

  • Skill instructions: Malicious or overly broad instructions can cause unsafe tool use. Frontmatter declares capabilities. Skill bodies define human gates and fail-closed behavior.

  • Dependency substitution: Owner overrides for Mnemosyne and Hephaestus can direct Athena to custom content. Thus, each override is an explicit trust decision. An automatically discovered fork can contain organization-specific changes. Athena accepts that fork only when all these conditions are true:

    • The current repository has an organization owner.
    • The authenticated viewer has WRITE, MAINTAIN, or ADMIN permission on that repository.
    • GitHub verifies that the candidate has the canonical repository as its parent.

    Athena reports the repository identity, commit SHA, and trust basis. Athena repeats the complete gate immediately before use. The origin of an existing checkout must identify the resolved repository.

  • Instruction and execution trust: Text from Mnemosyne enters the agent context. Automation from Hephaestus can execute commands. Before use, Athena reports the exact repository, commit, and trust basis. Athena stops when it cannot prove the identity, authority, ancestry, or checkout revision.

  • Marketplace redirection: Host manifests use the repository root. Athena validates the manifests before merge and release.

  • Supply chain: GitHub Actions use commit pins. The repository allowlist permits only the reviewed action revisions in the required and release workflows. Dependency checkouts verify their identity. A release contains repository resources and not executable package artifacts.

  • Untrusted fork code in continuous integration (CI): On pull requests, the required package job deliberately installs the plugin from the pull request head fork with pi install git:github.com/<fork>@<head> --no-approve and checks its source skill inventory through a Pi remote procedure call (RPC). It then installs the built archive and checks the archive skill inventory through a separate Pi RPC probe. Athena accepts this supply-chain risk because the job must inspect both inventories to verify the distributed artifact. The job has only contents: read permission. Its checkout does not persist credentials, Git prompts are disabled, the Pi runtime uses locked dependencies, and no token is exported to the step. Pi startup traffic is disabled for each probe, but the hosted runner retains public Internet egress. Athena accepts this residual risk. This step must never gain secrets, write-scoped tokens, or elevated workflow permissions. Any capability change requires an explicit review of this risk acceptance.

  • Secrets: Required CI scans the complete Git history. Repository policies prohibit credentials and private data.

Report a security issue in dependency code or its corpus to the resolved dependency repository. Athena security issues include unsafe resolution, invocation, permissions, packaging, or policy in this repository.

There aren't any published security advisories