The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to list and remove redundant skill copies in the .claude/ directory (e.g., rm -rf .claude/skills/*) and to run a bundled Python script (optimize_context.py) for project analysis. These actions are aligned with the stated purpose of context cleanup.
[PROMPT_INJECTION]: The skill processes and rewrites project instruction files, which constitutes an indirect prompt injection surface.
Ingestion points: The agent reads CLAUDE.md, GEMINI.md, and .github/copilot-instructions.md in Phase 3.
Boundary markers: No explicit delimiters are used to isolate ingested data during analysis, although the skill requires explicit user confirmation before writing changes.
Capability inventory: The skill has file-writing capabilities (Write tool) and command-execution capabilities (Bash tool).
Sanitization: No explicit sanitization or filtering of instruction file content is performed.
[DATA_EXFILTRATION]: No evidence of network exfiltration or unauthorized data access was found. All file operations and analysis are performed locally within the project scope.
[EXTERNAL_DOWNLOADS]: No remote dependencies or external script downloads from untrusted sources were identified. The included Python script uses standard libraries and performs local operations only.

optimize-context