Skip to content

AGENTS.md

Latest commit

 

History

History
36 lines (30 loc) · 1.65 KB

AGENTS.md

File metadata and controls

36 lines (30 loc) · 1.65 KB
Raw

Agent Guidance

Charon is a credential-handling security boundary. Prefer a small, explicit, fail-closed design over convenience.

  • Agents MUST NOT commit, log, snapshot, or fixture real credentials.
  • Destination access MUST remain deny-by-default and use exact hostnames. Agents MUST NOT add wildcard or caller-selected credential references.
  • Credential values MUST use secret-holding types and MUST NOT implement Debug, serialization, or response conversion.
  • Redirects MUST remain disabled unless every hop is independently re-authorized.
  • The data plane MUST remain independent of any orchestrator. Integrators issue signed workload identity and policy; Charon does not query an application's database or control-plane API.
  • Agents MUST update docs/threat-model.md when a trust boundary or credential flow changes.
  • Agents MUST run mise run check before committing.

Read the documentation index, repository conventions, and the contribution policy before changing public contracts or workflows.

Use tmp/<task-name>/ for disposable task artifacts. Agents MUST NOT use tmp/ for credentials. Promote retained work to its owned tracked path and remove only scratch files created by the current task.

Branch workflow:

  • Agents MUST create feature branches from dev and target ordinary pull requests to dev.
  • Only release pull requests from dev CAN target main.
  • Maintainers MUST merge dev into main with a merge commit to preserve ancestry.
  • Agents MUST treat main as production and release immutable container versions from it.
  • Agents MUST NOT rewrite shared branch history.