The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow involves ingesting data from untrusted external sources to guide the generation and execution of code.
Ingestion points: The research step uses web_search, web_extract, and read_file to collect information from the public internet (SKILL.md).
Boundary markers: The instructions lack explicit directives for the agent to treat external content as data only or to ignore embedded instructions (SKILL.md).
Capability inventory: The agent has access to powerful tools including terminal (shell execution), write_file (filesystem modification), and delegate_task (subagent orchestration) (SKILL.md).
Sanitization: The skill does not define methods for sanitizing or validating external documentation before it influences the code written and executed by the agent (SKILL.md).
[COMMAND_EXECUTION]: The skill facilitates the creation and execution of arbitrary code as part of its prototyping workflow.
The build step involves writing scripts to the filesystem and executing them via the terminal tool (e.g., python3 main.py) (SKILL.md). This is the intended primary purpose of the skill for feasibility prototyping.
[DYNAMIC_EXECUTION]: The skill generates and executes code at runtime based on research results.
Instructions guide the agent to write files and then run them using system interpreters like python3 or node via the terminal (SKILL.md).

spike