The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and documentation describe using python3 to execute the local script tiling_tree.py located within the skill's own directory. This is standard operational behavior for executing skill logic.
[EXTERNAL_DOWNLOADS]: The skill depends on the orchestrating-chat-agents skill. It modifies the Python path to import modules from /mnt/skills/user/orchestrating-chat-agents/scripts. This is a internal dependency between trusted local skills and does not involve untrusted remote sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes a user-provided 'problem' string which is interpolated into prompts for subagents.
Ingestion points: The problem argument in tiling_tree.py enters the agent context through the _splitter_prompt and _evaluator_prompt functions.
Boundary markers: Prompts use clear instruction/data separation with quotes (e.g., Problem space to partition: "{label}").
Capability inventory: The skill has capabilities to write files to /mnt/user-data/ and invoke parallel LLM calls via the claude_client dependency.
Sanitization: Standard prompt interpolation is used; while there is no explicit character-level sanitization, the use of system prompts and structural delimiters reduces the risk of accidental instruction obedience.

tiling-tree