Skip to content

Security: Cognitive-Delivery/contract

Security

SECURITY.md

Security policy

Reporting

Report a vulnerability privately to jason@cognitivedelivery.co.uk, or through GitHub's private advisory form.

Please do not open a public issue for a security problem. You will get an acknowledgement within five working days.

What is in scope

This repository holds JSON Schemas, a fixture corpus and a conformance runner. There is no service here and nothing that handles a credential, so the interesting failures are ones where the contract itself permits something it should not:

  • A privacy property the schemas fail to enforce. Audit and Index evidence is append-only and retained indefinitely, so anything that reaches it is effectively permanent. Prompts are never recorded; prompt_hash and request_hash are SHA-256; workspace_id is a hash of the git remote URL; the actor records a kind and a model, never a name, email or git identity. A shape that lets identifying data into a permanent record is a security issue, not a style one.
  • A sealing bypass. config-core.sealed lists fields an upper layer has fixed. A lower layer may match a sealed value or make it stricter, never looser. A schema that permits loosening would let a workspace quietly undo a control its organisation set.
  • A lease escape. agent-lease-manifest and agent-lease describe what an agent may reach. A shape that would let a manifest declare, or a lease grant, more than the harness intends is in scope.
  • A path in the corpus or runner that executes fixture content rather than reading it as data.

What is not in scope

  • A product that implements this contract badly. Report that to the product.
  • A missing feature, an unrecognised phase value, or a source form the reader reports as unsupported. Those are documented behaviour.
  • Anything requiring a fork of this repository to be published under this name; see the licence.

Versions

The 1.x schema set is supported. Within it, changes are additive only, and the additive-only guard runs on every change.

There aren't any published security advisories