The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes arbitrary shell commands through the buildCommand and checkCommand settings. These are invoked in scripts/brigade_lib/_edits/replay.py using subprocess.run to verify code changes within a temporary scratch directory.
[DYNAMIC_EXECUTION]: Several modules, including scripts/brigade_lib/_reviews/rounds.py, scripts/brigade_lib/_watch/runs.py, and scripts/brigade_lib/_spend/tokens.py, use runpy.run_path to dynamically execute local Python scripts like roles.py and activity.py at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill parses markdown files in the reports/ directory authored by worker agents. A compromised agent could output malicious content to influence the coordinator's logic.
Ingestion points: Report files (reports/<dish>.md) are read and parsed in scripts/brigade_lib/_tickets/follow_ups.py and scripts/brigade_lib/_reviews/threads.py.
Boundary markers: The skill relies on markdown headings (e.g., ## Follow-ups, ## Threads) to identify and delimit content for processing.
Capability inventory: Based on report content, the skill can file new work tickets and record execution thread metadata.
Sanitization: The skill removes control characters and attempts to filter private identifiers (UUIDs, email addresses, absolute paths), but it does not fully sanitize natural language instructions.

brigade