story-import

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted novel text provided by the user to generate structured project data, which involves interpreting external content within the agent's context.
  • Ingestion Points: Novel text input via file path or direct pasting in SKILL.md (Phase 1, Step 3).
  • Boundary Markers: The skill defines an explicit directory structure (e.g., 拆文库/, 正文/, 追踪/) to separate analysis from project data.
  • Capability Inventory: File system writes and local Python script execution are used to create the project structure.
  • Sanitization: The tracking_commit.py script validates identifiers and filenames against invalid characters and length limits to mitigate command injection or file path traversal.
  • [DYNAMIC_EXECUTION]: The skill executes a bundled Python script, scripts/tracking_commit.py, to manage and validate project state files at runtime.
  • Evidence: Execution calls are documented in SKILL.md (Phase 3-L, Step 7) and references/tracking-transaction.md.
  • Logic: The script uses Python standard libraries (argparse, json, os, pathlib, etc.) to perform local file operations within the user-defined project directory, representing standard skill functionality.
  • [METADATA_POISONING]: The skill's metadata contains a source URL for the project's repository. This URL belongs to the author's organization and is used for reference purposes, posing no safety risk.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 03:38 PM
Security Audit — agent-trust-hub — story-import