improve-session
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes session context, task evidence, and external artifacts which introduces a potential attack surface for instructions embedded in data.
- Ingestion points: Workflow Step 1 and 2 in SKILL.md involve retrieving candidate context and external evidence.
- Boundary markers: The skill explicitly mandates that retrieved content is evidence and not authority, and cautions that prior assistant reports are not proof.
- Capability inventory: The skill is authorized to execute tools, commands, and perform targeted file modifications (reverts and code reductions) within existing authority.
- Sanitization: The protocol requires grounding all claims in inspected sources and applying the quality-validation protocol to all substantive corrections.
Audit Metadata