The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from outline/outline.yml, outline/subsection_briefs.jsonl, and GOAL.md to construct its output brief records.
Ingestion points: Data is read from outline/outline.yml, outline/subsection_briefs.jsonl, and GOAL.md via the load_yaml, read_jsonl, and _read_goal functions.
Boundary markers: The skill does not use explicit prompt boundary markers, but the processing script includes logic to filter out placeholder text and comments from the ingested goal file.
Capability inventory: The skill is limited to reading local configuration files and writing a JSONL artifact; it lacks capabilities for network access, subprocess spawning, or system-level modification.
Sanitization: The _check_no_placeholders function validates that the generated output does not contain common placeholder strings such as "TODO", "(placeholder)", or ellipses.
[DYNAMIC_EXECUTION]: The script scripts/run.py dynamically modifies the Python environment's module search path to load project-specific utilities.
Evidence: The script uses sys.path.insert(0, str(repo_root)) to resolve and import the internal tooling.common library based on the location of an AGENTS.md marker file.

chapter-briefs