The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes a local Python script (scripts/run.py) for automated text processing. The script uses standard libraries and performs safe file operations within the designated workspace. It does not invoke shells, download external payloads, or use dynamic execution functions like eval or exec.
[DATA_EXFILTRATION]: The skill's operations are confined to the provided workspace directory. No network-enabled libraries are used, and no attempts to access sensitive system paths (e.g., SSH keys, cloud credentials) or exfiltrate data to remote servers were detected.
[INDIRECT_PROMPT_INJECTION]: The skill processes external text data, presenting a potential surface for indirect injection.
Ingestion points: Reads content from sections/*.md, outline/writer_context_packs.jsonl, and citations/ref.bib.
Boundary markers: None identified in the script or prompt instructions.
Capability inventory: File read/write access to the local workspace directory. No network or subprocess capabilities.
Sanitization: Content is filtered through deterministic regex patterns and keyword matching defined in assets/numeric_hygiene.json.
[SAFE]: The skill follows best practices for its intended purpose. It does not exhibit signs of obfuscation, privilege escalation, or persistence mechanisms. All logic is transparently defined in the provided markdown and script files.

evaluation-anchor-checker