The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/harness.py executes the git binary to resolve project paths and identities.
Evidence: The script uses subprocess.run with a list of arguments to call git. It specifically clears GIT_ environment variables to prevent environment-based hijacking and uses a timeout to prevent hanging.
[INDIRECT_PROMPT_INJECTION]: The skill manages and processes external knowledge files and handoff records which could contain instructions.
Ingestion points: The read and write operations in scripts/harness.py access Markdown files in the project's knowledge directory. The handoff operation processes text provided via stdin.
Boundary markers: Content is read and returned as raw text; the skill does not wrap findings in specific security delimiters.
Capability inventory: The skill performs file system reads/writes and executes local git commands.
Sanitization: The script validates that files are UTF-8 Markdown and uses path validation to prevent traversal outside the designated knowledge and project directories.

harness-init