The Agent Skills Directory

[NO_CODE]: The skill package contains no executable scripts or binaries. It consists of markdown instructions, configuration YAML, and license/attribution files.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied plans, designs, and ideas, which it then audits using available environment tools and sub-agents. This creates a surface where a malicious plan (especially if read from an external file or PR) could contain instructions to influence the agent's investigation.
Ingestion points: User-provided ideas and formed plans processed by the discovery and audit modes in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the prompt templates.
Capability inventory: The skill is instructed to investigate the environment, use sources/tests, and dispatch sub-agents to verify facts as described in SKILL.md and references/grilling-upstream.md.
Sanitization: No sanitization or escaping of the user-provided content is performed.
[SAFE]: The skill includes proper attribution and references to third-party interaction models (Matt Pocock's skills repository). These references are for documentation and licensing purposes and do not involve remote code execution or untrusted downloads.

deep-grill