The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the claude CLI tool using subprocess.run in scripts/compress.py to perform text compression tasks when the SDK is not present.
[DATA_EXFILTRATION]: Natural language content is sent to the Anthropic API. The skill implements a security denylist in scripts/compress.py (is_sensitive_path) to skip files matching patterns like .env, id_rsa, or directories like .ssh to prevent accidental data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external memory files, creating a surface for injection. Mandatory evidence: (1) Ingestion points: scripts/compress.py reads file content using Path.read_text; (2) Boundary markers: Prompts in scripts/compress.py lack explicit delimiters or instructions to ignore embedded commands; (3) Capability inventory: Skill uses subprocess.run for commands and Path.write_text for file modification; (4) Sanitization: Heuristic path-based filtering (is_sensitive_path) is present but does not sanitize content.
[DYNAMIC_EXECUTION]: The script uses Python's subprocess module to dynamically invoke external system tools for model interaction.

compress