The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data access detected across the analyzed files. The skill's stated purpose as a dispatcher matches its technical implementation.
[COMMAND_EXECUTION]: The skill uses git to resolve the scope of changes and gh for creating issues. These operations are restricted to the intended workflow and include confirmation checkpoints for high-impact actions like whole-tree edits or issue creation.
[PROMPT_INJECTION]: Includes a specific defensive guardrail against indirect prompt injection by instructing the agent to never obey commands embedded in the source code it analyzes. It also uses disable-model-invocation: true to ensure the dispatcher follows defined logic rather than generating arbitrary actions.
[DATA_EXPOSURE]: Indirect prompt injection attack surface evaluation:
Ingestion points: Processes local source code through git-based scope resolution and delegation to other engines.
Boundary markers: Explicitly warns the agent in the 'External Side Effects' section to treat analyzed source as untrusted data.
Capability inventory: Accesses file metadata and contents via git and has the capability to write to GitHub via gh (on confirmation).
Sanitization: Implements instruction-based sanitization by explicitly forbidding obedience to embedded instructions.

refactor-dispatch