The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes destructive rm -rf commands targeting contents within .claude/skills/, .claude/agents/, .claude/commands/, and .claude/hooks/. While intended to resolve duplication issues, these operations can lead to data loss if those directories are symlinked to critical sources or contain project-specific configurations.
[COMMAND_EXECUTION]: The skill runs a local Python script located at plugins/dev-utils/scripts/optimize_context.py. Because the script's content is not part of the analyzed package, its full functionality and safety cannot be verified.
[PROMPT_INJECTION]: The skill reads and summarizes local instruction files like CLAUDE.md and GEMINI.md. This creates an indirect prompt injection surface where malicious instructions embedded in these files could potentially influence the agent's behavior or outcomes during the summarization and rewrite phase.
Ingestion points: Reads CLAUDE.md, GEMINI.md, and .github/copilot-instructions.md.
Boundary markers: None explicitly defined for the ingestion of file content.
Capability inventory: Uses Bash for file deletion and Write for modifying instruction files.
Sanitization: Includes a human-in-the-loop step requiring confirmation before writing changes.

optimize-context