The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/verify-criteria utility uses the eval command to execute shell strings passed as arguments, enabling arbitrary shell execution including pipes and redirects.
[COMMAND_EXECUTION]: The references/plan-lifecycle.md documentation specifies a cleanup step using rm -rf on directory paths, which is a high-impact file system operation.
[DYNAMIC_EXECUTION]: The use of eval in the verify-criteria script is a dynamic execution pattern that runs string-based input as shell code.
[INDIRECT_PROMPT_INJECTION]: This skill functions as a plan generator that ingests untrusted specifications and produces executable briefings for other agents. Ingestion points: External specs or requirements are ingested as the source for plan generation (File: SKILL.md). Boundary markers: The generated briefings lack explicit delimiters or 'ignore' instructions for content sourced from requirements. Capability inventory: Generated briefings contain instructions for file writes, git commits, and shell command execution (File: references/briefing-template.md). Sanitization: The skill does not perform sanitization or escaping of requirement text before incorporating it into briefings.

writing-plans