The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a variety of command-line tools for codebase analysis, type checking, and maintenance. These include git (for history tracing and commits), grep (for pattern searching), tsc (for TypeScript validation), and various dead-code analysis tools such as knip, vulture, and depcheck. It also involves direct file system operations, such as deleting files (rm).
[SAFE]: The skill implements robust safety protocols to mitigate the risks associated with its destructive core functionality. It mandates a 'plan mode' during the discovery and manifest creation phases to ensure no changes are made until a full strategy is approved. Infrastructure changes and database migrations are explicitly marked as 'HUMAN-GATED,' requiring the agent to propose actions for human execution rather than performing them autonomously.
[SAFE]: The skill has a surface for indirect prompt injection because it processes code from the repository which could contain malicious instructions. However, this is managed by the skill's structure and human-in-the-loop requirements.
Ingestion points: The skill reads repository files, git logs, and tool outputs (e.g., from knip or grep) during the tracing phase.
Boundary markers: The skill does not provide specific instructions to the agent to ignore instructions embedded within the codebase being processed.
Capability inventory: The agent has capabilities to delete files, modify code, execute local analysis tools, and propose database migrations.
Sanitization: No specific sanitization or filtering of the ingested code content is performed before it enters the agent's context.

remove-feature