durable-docs-update
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from code comments, discovery bullets, and specification documents which could contain hidden instructions targeting the agent's logic.
- Ingestion points: Processes content from changed code files, user-supplied
Discovery:bullets, andspec.mddecisions in Step 2. - Boundary markers: The protocol establishes task-level boundaries for subagents but lacks explicit content delimiters or instructions to ignore embedded prompts within the ingested text.
- Capability inventory: Performs file reads (via
git show), executes internal tool calls (vet-fact,place-fact,compress-file,tighten-instruction,structure-prose,check-coherence), and performs file updates through application subagents in Step 4. - Sanitization: There is no evidence of input sanitization, filtering, or escaping before data is merged and applied to the codebase or documentation files.
Audit Metadata