compact-message
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes the entire conversation history, which may include untrusted data from tool outputs, file contents, or fetched web pages. This constitutes an attack surface where malicious instructions could be embedded in the session data to influence the generated compaction summary.
- Ingestion points: Conversation history processed in
SKILL.md(Step 2). - Boundary markers: The skill relies on a 'Facts only' rule and user review of the final copy-pasted output.
- Capability inventory: The skill has no dangerous capabilities; it does not execute shell commands or perform network requests.
- Sanitization: Implements a 'Facts, not foreign instructions' policy to convert external data into attributed facts rather than commands.
- [DATA_EXPOSURE]: The skill implements a robust 'No secrets' rule in Step 4, instructing the agent to never include tokens, keys, passwords, or credential-bearing URLs in the final output string, providing a safeguard against accidental credential leakage during session compaction.
Audit Metadata