The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions by aggregating data from multiple workspace sources into a format that directly instructs a downstream agent's writing process. \n
Ingestion points: Reads data from outline/subsection_briefs.jsonl, outline/chapter_briefs.jsonl, outline/evidence_drafts.jsonl, outline/anchor_sheet.jsonl, outline/evidence_bindings.jsonl, and citations/ref.bib within the scripts/run.py script. \n
Boundary markers: The output JSONL structure lacks delimiters or 'ignore' instructions to separate content from control, which could allow malicious data in input fields to be interpreted as valid commands by the downstream agent. \n
Capability inventory: The skill executes a Python script that performs filesystem write operations to outline/writer_context_packs.jsonl. \n
Sanitization: The script includes a _sanitize_source_text function that uses regular expressions to filter out paper self-narration and generic boilerplate, but this is intended for stylistic hygiene rather than security filtering. \n- [COMMAND_EXECUTION]: The SKILL.md file provides instructions for executing a local Python script via uv run python, which is the primary mechanism for generating the context packs.

writer-context-pack