audit
Warn
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONDYNAMIC_CONTEXT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The skill reads .env files to identify connected tools and verify API key presence for domain coverage assessment. This practice exposes sensitive credentials directly to the model's context.
- [DYNAMIC_CONTEXT_INJECTION]: The skill uses the !
date +%Y-%m-%dsyntax to inject the current date into the prompt at load time. This mechanism allows for shell command execution automatically when the skill is accessed. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project data from manuals, memory files, and decision logs. This creates an attack surface where instructions in those files could influence the audit report. 1. Ingestion points: CLAUDE.md, MEMORY.md, and project skills folder. 2. Boundary markers: Absent. 3. Capability inventory: Local file reading, shell command execution via dynamic context, and writing the final audit report to disk. 4. Sanitization: None identified.
- [COMMAND_EXECUTION]: The skill executes the
datecommand during context gathering to determine the timestamp for the audit.
Audit Metadata