The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes source code to perform refactoring, creating a surface for indirect prompt injection if the code contains malicious instructions intended to mislead the agent.
Ingestion points: Reads local source code files for analysis and restructuring.
Boundary markers: No explicit instruction delimiters are defined for the code content, though the workflow includes manual approval gates.
Capability inventory: Performs file system modifications, executes shell commands for testing and analysis, and performs git commits.
Sanitization: Relies on manual human review (Gate 1 for planning and Gate 2 for pre-commit) to verify the neutrality and safety of changes.
[COMMAND_EXECUTION]: The workflow involves running external development tools and project-specific test suites to ensure behavior preservation.
Evidence: Mentions running analysis tools such as knip, depcheck, and ts-prune via a helper agent, and executing the project's own test suite throughout the refactoring process.

orch-refine-code