The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to automate the configuration of the development environment.
It uses rsync and cp to copy template files from the environment-defined directory ${CLAUDE_PLUGIN_ROOT}.
It performs a shell check for an overlay directory using ls and environment variable interpolation ($POWERSUIT_OVERLAY).
It runs a local Python script located at .github/actions/gate/run.py to validate the generated agent settings.
It utilizes git commands to manage the tracking of configuration files, including git check-ignore and git add -f to ensure shared settings are tracked by version control.
[INDIRECT_PROMPT_INJECTION]: The skill processes configuration and instruction files from the repository and external paths which could contain instructions that influence agent behavior.
Ingestion points: The skill reads .claude/settings.json, CLAUDE.md, AGENTS.md, and any files found in the path specified by the POWERSUIT_OVERLAY environment variable (specifically README.md and files in ai-ops/agent-harness/).
Boundary markers: The skill does not define or use specific delimiters to separate the content of these ingested files from its own core instructions.
Capability inventory: The skill has the capability to modify the file system (rsync, cp, mkdir), execute shell commands, and run local Python scripts.
Sanitization: The skill relies on instructions to "Show every change" to the user for manual review and explicitly mandates the masking of any detected secrets, but it does not implement programmatic sanitization of the content read from external files.

agent-harness