The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on local CLI tools (ak, gh) to perform its functions. It executes commands to manage boards, tasks, agents, and Pull Requests. It also uses system utilities like ps and grep for health monitoring. These actions are within the expected scope of a task management tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content such as repository code, documentation, and PR diffs, creating a potential surface for indirect prompt injection.
Ingestion points: Files are read from the repository in Phase 1 (Step 2) and Phase 2 (Step 6). PR data is fetched from GitHub during review gates.
Boundary markers: The instructions guide the agent to evaluate content against technical and functional criteria but do not provide explicit delimiters or "ignore instructions" markers for the untrusted data.
Capability inventory: The skill has significant capabilities including merging code, creating/deleting tasks, and modifying agent configurations.
Sanitization: No specific sanitization or escaping of external content is defined before processing or interpolation into commands.
[DYNAMIC_EXECUTION]: The skill dynamically generates and applies Agent YAML configurations to define worker agent behaviors (soul) and capabilities (skills) at runtime. While this involves modifying the execution environment of sub-agents, it is a core feature of the delegation model and follows the architecture's design.

ak-task