reader-panel

Pass

Audited by Gen Agent Trust Hub on Oct 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Git commands (status, add, commit, tag) to manage project snapshots. It also runs a Story CLI tool (via 'story' or a local Node.js script) to generate context and build review copies. All operations are confined to the local project environment.\n- [CREDENTIALS_UNSAFE]: The skill proactively instructs the agent to scan for sensitive files such as .env, keys, or credentials using git status and git ls-files. It directs the user to add these to .gitignore to prevent accidental exposure in the repository history.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-authored chapters to generate feedback, creating a surface for indirect instructions. \n
  • Ingestion points: Reads chapter markdown files from the local filesystem. \n
  • Boundary markers: The agent is instructed to strip non-reader context (like future plot spoilers) and use isolated subagents for each persona to limit the scope of the LLM's attention. \n
  • Capability inventory: The skill can perform local file writes and execute Git/CLI commands. \n
  • Sanitization: The skill relies on subagent isolation and specific persona reference files to maintain behavior boundaries while processing narrative content.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 15, 2026, 03:07 PM