The Agent Skills Directory

[SAFE]: The skill consists exclusively of text-based instructions and formatting templates intended to guide the agent's reasoning process for task decomposition. It does not include executable code, script files, or automation commands.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided task descriptions (parent tasks) to generate structured sub-task contracts.
Ingestion points: Processes user requests within the chat context as defined in SKILL.md (Capture the parent in chat).
Boundary markers: Utilizes Markdown headers (e.g., # Parent, ### Slices) to delineate different sections of the structured output.
Capability inventory: The skill explicitly disables model invocation (disable-model-invocation: true) and does not perform file-system writes, network operations, or subprocess executions.
Sanitization: While no specific sanitization logic is present for user-provided task titles, the skill's primary function is formatting for organizational purposes rather than execution, significantly limiting the impact of any injected instructions.
[EXTERNAL_DOWNLOADS]: The file contains relative references to internal documentation (e.g., ../pack-shared/standards.md, ../pack-shared/asking.md). These are local file path references within the expected repository structure and do not represent external network downloads or unverifiable dependencies.

split-task