skills/seanmars/mreq/mreq-define/Gen Agent Trust Hub

mreq-define

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local script at a path relative to the skill directory: ${CLAUDE_SKILL_DIR}/../mreq-shared/scripts/mreq.py. It uses uv run as the primary execution method, with fallbacks to python or python3. This script is utilized for creating and listing task documents within the .mreq/ directory.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user requests and the local codebase to generate plans and implement changes.
  • Ingestion points: The user-provided <request> argument and various project files (README, source code, and agent instruction files) read during the initial exploration phase (Step 1).
  • Boundary markers: The skill instructions mandate a "Restatement" phase where the agent must summarize the goal in its own words and receive explicit user confirmation before proceeding to write files or implement code.
  • Capability inventory: The agent has the capability to run a local Python script, read/write files in the repository (specifically in .mreq/), and execute shell commands for git operations and testing.
  • Sanitization: The mandatory confirmation step and the requirement to restate goals in the agent's own words serve to sanitize and validate input, reducing the risk that malicious instructions in the data are obeyed without user oversight.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 04:21 PM