The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user arguments and brainstorm documents to generate implementation plans.
Ingestion points: User-supplied $ARGUMENTS and local markdown files in the docs/brainstorm/ directory (SKILL.md).
Boundary markers: Uses XML-style <feature_description> tags to delimit user input (SKILL.md).
Capability inventory: The skill performs file reads, codebase searches using grep, git operations, and writes new plan files to the docs/plan/ directory (SKILL.md).
Sanitization: No explicit sanitization or escaping of the external content is described before interpolation into the generated plan.
[COMMAND_EXECUTION]: The skill uses local system tools to gather context and manage the workspace.
Evidence: Executes ls to check for brainstorm documents, grep and glob for codebase searches, and git for branch management and verification.
[DYNAMIC_EXECUTION]: The skill instructs the agent to create machine-executable verification commands for inclusion in the generated plan.
Evidence: The 'Success Criteria Gate' section requires the generation of shell commands (e.g., verify: <command>) to provide automated proof of implementation success.

plan