The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates spawning subagents using data interpolated from external project files (e.g., roadmaps, research docs, task registries). This creates an attack surface where untrusted content in those files could influence subagent behavior.\n
Ingestion points: SKILL.md instructs the agent to read .planning/execution/tasks.json, roadmaps, and research docs.\n
Boundary markers: task-template.md uses Markdown headers and list structures to separate instructions from task data.\n
Capability inventory: Subagents are expected to write files to the .planning/execution/handoffs/ directory and execute shell commands for verification (e.g., npm test, pytest).\n
Sanitization: The instructions do not specify sanitization or validation logic for the interpolated task data.\n- [COMMAND_EXECUTION]: The skill recommends using shell commands (e.g., npm test, pytest, cargo test, npm run build) to verify task completion and gate merges. These commands are standard for development workflows and are used within the skill's primary purpose of managing software project execution.

wave-execution