common-learning-log
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied feedback and 'correction language' from the session to generate structured log entries in
AGENTS_LEARNING.md. This represents a data ingestion surface where an adversary could provide 'corrections' that contain malicious instructions. - Ingestion points: User input during mid-session corrections and automated feedback reports (SKILL.md, Protocol Step 1).
- Boundary markers: The skill uses a structured Markdown template (references/log-format.md) which provides some implicit separation, but lacks explicit instructions to ignore or escape instructions embedded within the user's 'mistake' data.
- Capability inventory: The skill is authorized to read and write to the filesystem specifically for
AGENTS_LEARNING.mdin the project root. - Sanitization: No evidence of sanitization or escaping of the user-provided correction text before it is interpolated into the Markdown log entry.
Audit Metadata