ce-compound

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script utilities/skill-builder/scripts/record_skill_feedback.py to persist decision feedback. The command interpolates multiple arguments (decision, outcome, confidence, notes) that may be influenced by user interaction, constituting a command execution surface.
  • [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests and processes untrusted data.
  • Ingestion points: The skill reads MEMORY.md, conversation history, and existing markdown artifacts from directories like docs/brainstorms/ and docs/specs/.
  • Boundary markers: Instructions in references/learning-capture.md require labeling auto-memory as 'supplementary context' and prioritize primary evidence over it.
  • Capability inventory: The skill has the ability to write files to docs/solutions/ and .harness/ and to execute local Python scripts.
  • Sanitization: The skill contains a specific policy for 'PII/Secrets redaction' to prevent the inclusion of credentials or personal data in generated artifacts.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 01:13 PM
Security Audit — agent-trust-hub — ce-compound