The Agent Skills Directory

[DYNAMIC_EXECUTION]: The helper script scripts/run.py dynamically modifies the Python search path (sys.path) by resolving the repository root at runtime based on the presence of a marker file (AGENTS.md). It then performs imports from the tooling package located at this computed path.
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from external sources to generate its output, creating a vulnerability surface where instructions could be embedded in the input data.
Ingestion points: Processes data from outline/subsection_briefs.jsonl, outline/evidence_drafts.jsonl, and citations/ref.bib.
Boundary markers: Utilizes a manual approval mechanism in DECISIONS.md (Approve C2) and explicit instructions to identify and reject scaffold markers.
Capability inventory: The skill and its associated script have the capability to write files in the workspace, specifically output/DRAFT.md, output/SNAPSHOT.md, and updates to DECISIONS.md via the gate check process.
Sanitization: Instructions require the agent to perform a self-check for scaffold text, ellipsis, and placeholder leakage before finalizing output.

prose-writer