project-skill-audit
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses sensitive local directories containing agent session histories, specifically
~/.claude/projects/,~/.codex/sessions/, and~/.gemini/tmp/. These files contain plaintext records of previous tool executions, file contents, and command outputs, posing a data exposure risk.\n- [COMMAND_EXECUTION]: To perform its audit, the skill uses shell commands likejq,grep,find, andgitto process local files and history data.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests large volumes of untrusted data from repository files and session transcripts.\n - Ingestion points: Processes
.jsonltranscripts,SKILL.md,CLAUDE.md,AGENTS.md, andMEMORY.mdfiles.\n - Boundary markers: No explicit delimiters or technical boundary markers are used when interpolating this data into the agent's context.\n
- Capability inventory: Utilizes shell tools (
jq,grep,git,ls) to filter and extract information.\n - Sanitization: Relies on natural language instructions for the agent to redact secrets rather than programmatic sanitization.
Audit Metadata