The Agent Skills Directory

[SAFE]: The skill utilizes a bundled Python script (scripts/semantic_compressor.py) for all deterministic operations. This script has no external dependencies beyond the Python standard library, minimizing the attack surface.
[COMMAND_EXECUTION]: The skill executes the local script for file discovery, metric calculation, and atomic file replacement. The script includes robust path validation logic (reject_symlink_components and is_within) that restricts operations to the project root and rejects symbolic links, effectively preventing path traversal attacks.
[PROMPT_INJECTION]: The skill is designed to handle potential indirect prompt injection. It explicitly instructs the agent to treat candidate content as untrusted and never execute embedded shell snippets or instructions. This is reinforced by a workflow that delegates verification to a fresh, independent read-only reviewer subagent.
[SAFE]: The skill implements a secure backup and restoration system. It creates unique run directories within the project structure and uses atomic file replacement to prevent data corruption during the compression process.
[SAFE]: No hardcoded credentials or sensitive data exfiltration patterns were detected. The Python script uses os.urandom for generating unique session IDs and follows best practices for temporary file management and file permissions (using 0o600 for sensitive artifacts).

semantic-compressor