data-loss-gate
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill functions as a security guardrail by interposing a mandatory confirmation step before destructive commands (such as rm -rf, bulk git rm, or database truncations) are executed. It requires the agent to halt, perform a recoverability analysis, and wait for an explicit human 'yes'.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests untrusted data from the filesystem to populate the confirmation card.
- Ingestion points: The agent is instructed to inspect target files, git logs, and brain backlinks to generate the card's content (SKILL.md, Step 1).
- Boundary markers: The skill utilizes a structured card template and enforces a strict human-in-the-loop approval process ('yes' or 'do it'), which serves as a safety boundary against accidental obedience to instructions embedded in target data.
- Capability inventory: The skill utilizes local git tools and vendor-specific gbrain commands for inspection, with logging directed to a local daily/ directory.
- Sanitization: While no specific data sanitization is mentioned, the manual confirmation requirement is the primary mitigation against the agent being misled by external content.
Audit Metadata