release-runbook-author
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data sources, specifically 'tribal knowledge' or 'chat threads', to generate structured release documentation. This constitutes an ingestion point for potentially malicious instructions embedded in the source material.
- Ingestion points:
SKILL.md(instructions to use chat threads or tribal knowledge as the primary source for runbook content). - Boundary markers: The skill mitigates risks by requiring the agent to use rigid, predefined markdown table templates and structured output formats as defined in
references/output-template.mdandreferences/cutover-output-template.md. - Capability inventory: The skill does not contain internal executable scripts, shell commands, or network operations. It relies on the agent's pre-existing, externally-managed CLI tools (e.g.,
gh,npm) for evidence collection. - Sanitization: The skill emphasizes recording evidence as literal strings, specific IDs, or captured command outputs, which enforces a layer of human-verifiable structure over the generated document.
- [NO_CODE]: The skill consists entirely of instructional Markdown files, templates, and metadata. It does not include any scripts (.py, .js, .sh), binaries, or configuration files that trigger automated code execution. All logic is provided via instructional prompts and structural guidance.
Audit Metadata