The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data to coordinate agent execution, creating a surface for indirect injection attacks.
Ingestion points: The agent reads project state from project-graph.json, canvas manifest.json, and source prompt Markdown files located in the workspace to determine routing and execution steps.
Boundary markers: Absent. The instructions do not define delimiters or specific "ignore embedded instructions" commands to separate plan metadata from control flow.
Capability inventory: The skill uses the <pw> (PlanWeave) CLI to execute state changes, claims work, and spawns subagents (plan-runner, plan-reviewer, plan-recovery) with specific instructions.
Sanitization: Absent. The skill interprets content from ingested configuration files directly to decide which executors to run and which subagents to spawn.
[COMMAND_EXECUTION]: The skill relies heavily on a command-line interface to manage plan execution and monitoring.
The coordinator is instructed to execute numerous CLI commands including <pw> paths, <pw> claim-next, <pw> run, <pw> run-status, and <pw> run-session to interact with the PlanWeave runtime.
It also performs subagent lifecycle management using platform-native tools to spawn, monitor, and close worker processes.

plan-coordinator