add-api-endpoint
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill instructions and references emphasize defensive programming and security best practices. Key findings include:
- Enforcement of RFC 9457 (Problem Details) ensures that 5xx errors are sanitized to remove stack traces, internal paths, and credentials, preventing sensitive data exposure.
- Mandatory schema-first invariant contracts (OpenAPI 3.1) and runtime boundary validation (Zod/Pydantic) prevent the injection of malformed or malicious data into the service boundary (ASI01).
- Requirements for default-deny authentication and authorization on all non-public endpoints prevent unauthorized access (ASI03).
- The mandate for Level 0 air-gapped sandbox testing (
--network=none) ensures that the code and its tests cannot perform unauthorized network operations during development or CI/CD cycles. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface by reading external data from
feature-ticket.jsonand ADRs. This risk is mitigated by the skill's strict workflow requiring the definition and validation of a machine-readable schema contract before any implementation occurs. - Ingestion points:
feature-ticket.json, ADR documents (SKILL.md). - Boundary markers: Machine-readable OpenAPI/JSON Schema meta-validators (SKILL.md).
- Capability inventory: File manipulation and command execution (SKILL.md frontmatter).
- Sanitization: Rigid runtime boundary enforcement via Zod/Pydantic and global RFC 9457 exception filters (SKILL.md, references/api-contracts-and-error-specs.md).
Audit Metadata