vault-guide
Pass
Audited by Gen Agent Trust Hub on Mar 3, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted input from candidates and uses it to shape agent behavior and generate persistent files.
- Ingestion points: Candidate responses in Phase 0, 1, 3, 3.5, and 5 (
SKILL.md). - Boundary markers: Absent; the agent is instructed to weave candidate values and interests directly into the workflow.
- Capability inventory: File reading (
ls,grep,glob,cat), file writing, and repository operations (git add,git commit,git push). - Sanitization: No evidence of escaping or validation of candidate-provided strings before they are incorporated into session chronicles or contract proposals.
- [COMMAND_EXECUTION]: The skill utilizes command-line tools for file system navigation and version control management.
- Evidence: Explicit instructions to execute
git status,git add,git commit, andgit pushin Phase 7 and the Hidden Assessment Protocol. - Evidence: Use of
ls,glob, andgrepto orient candidates to the workspace in Phase 1 and 4.
Audit Metadata