factory-release
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements strong security invariants for release management, including 'exact-head' SHA verification and multiple human approval gates, which prevents unauthorized or accidental code promotion.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from external sources such as pull request (MR) bodies and tracker issues.\n
- Ingestion points: Untrusted data enters the agent context through MR URLs and tracker records specified in
workflow.ymlfor steps likeverify,gate-merge, and canary authorizations.\n - Boundary markers: The skill proactively mitigates injection risks by including explicit instructions in
workflow.ymlandSKILL.md(Invariant 6) to treat all tracker and MR content as raw data and to ignore any embedded directives.\n - Capability inventory: The skill possesses capabilities to merge code, create git tags, modify the product roadmap file (
docs/adlc/product/roadmap.md), and interact with milestone APIs.\n - Sanitization: It employs a 'message-is-data' policy where external inputs are isolated from the control flow, ensuring that gate authorizations rely only on structured gate reads rather than free-form text.
Audit Metadata