The Agent Skills Directory

[SAFE]: The skill is entirely instructional, focusing on providing templates and methodologies for writing project rules. It does not include executable code, external network calls, or access to sensitive system credentials.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of behavioral instructions that persist across agent sessions.
Ingestion points: Triggered by agent self-correction during development tasks.
Boundary markers: Explicitly requires a contrastive structure (Incorrect vs. Correct) to minimize ambiguity.
Capability inventory: Limited to writing markdown documentation to the project's .claude/rules/ directory.
Sanitization: Includes a multi-step validation and refinement cycle (Decompose, Filter, Reweight) to prevent contradictory or redundant instructions.

create-rule