quality-gate
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external files (e.g., .agents/quality/findings.jsonl, metrics.jsonl, flaky.jsonl, and debt.md) to determine release verdicts (GO/NO-GO).
- Ingestion points: The agent is instructed to read various JSONL and Markdown files under the .agents/quality/ directory, as well as registry/capabilities.yaml.
- Capability inventory: The skill has permissions to use Bash, Write, and Edit tools, which are used to modify CI and gate configurations based on the ingested evidence.
- Boundary markers: The instructions do not specify sanitization or delimiters for the data ingested from these evidence logs, potentially allowing data-driven instructions to influence the verdict.
- Sanitization: There is no evidence of validation or filtering of the external content before it is used to influence the agent's decision-making process or CI configuration writes.
- [DYNAMIC_EXECUTION]: The skill references and encourages the use of an external script refute.py to verify claims, which is not provided in the skill bundle.
- Evidence: SKILL.md contains the instruction 'refute — put it to the engines that did not make it'.
- Risk: The source code for refute.py is not provided for security auditing, while the skill possesses the Bash tool, allowing for the execution of unverified logic if the script is present in the environment.
Audit Metadata