capture

Pass

Audited by Gen Agent Trust Hub on Apr 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute the Bash command date +%Y-%m-%d to retrieve the current date. This is a benign system command used for metadata generation.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data that could influence future agent actions.
  • Ingestion points: Untrusted user input is captured via the $ARGUMENTS variable in SKILL.md.
  • Boundary markers: No delimiters or ignore-instruction warnings are placed around the captured text.
  • Capability inventory: The skill utilizes Read and Edit tools to modify local file system content (SKILL.md).
  • Sanitization: No escaping or validation is performed; the instructions explicitly state to "append as-is".
  • [DATA_EXFILTRATION]: The skill accesses a hardcoded file path within a user's home directory (/home/mat/dev/gtd-pcm/01-inbox.md). While this is consistent with the skill's stated purpose of managing a GTD inbox, it involves reading and writing to a personal file location.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 2, 2026, 06:17 AM
Security Audit — agent-trust-hub — capture