learning-aggregator

Pass

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes untrusted data from .learnings/ files (such as error summaries and user feedback) and includes it in a structured 'Gap Report' intended for other agents like harness-updater.
  • Ingestion points: Reads files LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md from the local .learnings/ directory.
  • Boundary markers: The skill does not define clear delimiters or instruction-ignore markers when outputting log evidence into the generated report.
  • Capability inventory: Analysis of the skill body confirms it is read-only; it contains no subprocess calls, network operations, or file-writing logic.
  • Sanitization: There is no evidence of sanitization or escaping of metadata fields (Summary, Details) before they are interpolated into the output Markdown report.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 16, 2026, 05:24 AM