discovery-drafting
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-drafted prose to perform entity extraction and update the story bible (markdown records).
- Ingestion points: User-provided story chapters and scene files are read by the agent to identify characters, locations, and plot promises (see SKILL.md and reconcile-loop.md).
- Boundary markers: The instructions do not specify explicit boundary markers for separating user prose from instructions during the extraction phase, though they emphasize using the user's spelling for names.
- Capability inventory: The skill uses the story CLI (or a fallback Node.js script) to rebuild indexes, update word counts, and create new markdown files for bible entries.
- Sanitization: The skill specifies that IDs for new entities must be converted to ASCII kebab-case, providing a level of sanitization for filenames. It also mandates user approval (human-in-the-loop) before any new bible files are created or records updated.
- [COMMAND_EXECUTION]: The skill relies on the story CLI tool or a local Node.js script (path: ../story-maintenance/scripts/story.js) to perform maintenance tasks.
- Evidence: Execution of commands like
story reindex .,story wordcount . --write, andstory check .as described in SKILL.md. - Context: The skill instructs the agent to prefer Node.js over Bun for execution to prevent the unintended loading of potentially malicious configuration files (such as bunfig.toml or .env) found in the project directory, demonstrating awareness of environment-based code execution risks.
Audit Metadata