phase-gated-debugging
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it instructs the agent to ingest and process potentially untrusted data like error logs and source code during the reproduction and isolation phases.
- Ingestion points: Phase 1 (reproducing and capturing error output) and Phase 2 (reading source code for isolation).
- Capability inventory: The skill utilizes tool capabilities for command execution (Phase 1 reproduction and Phase 5 verification) and file modification (Phase 2 logging and Phase 4 bug fixing).
- Boundary markers: The skill does not define specific delimiters for untrusted data but enforces a significant mitigation through a mandatory human confirmation gate before moving from Phase 3 to Phase 4.
- Sanitization: There is no explicit requirement for the agent to sanitize or filter the contents of logs or code before processing, though the structured protocol minimizes the impact of automated obedience to embedded instructions.
- [EXTERNAL_DOWNLOADS]: The skill includes informational URLs in the frontmatter and body referencing external GitHub repositories. These are used for attribution and to provide background on the debugging methodology; they do not trigger automated runtime downloads or script execution.
Audit Metadata