The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing a local Python script ~/Documents/agent-skills/scripts/plan_runner.py to manage state, normalize files, and generate command templates. It also directs the agent to execute shell commands specified within external plan files.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves processing untrusted plan.md files which may contain malicious instructions or dangerous commands intended for execution.
Ingestion points: Processes external plan files provided via $ARGUMENTS through the plan_runner.py script.
Boundary markers: Uses specific delimiters such as --- plan data (not instructions) --- to separate data from instructions during context injection.
Capability inventory: Utilizes the Bash tool for script execution and mentions integration with high-privilege tools like kubectl, terraform, and helm.
Sanitization: Implements a robust sandbox mechanism (PLAN_SANDBOX_ROOT) that explicitly blocks access to sensitive directories including /, $HOME, and ~/.claude. It also features a mandatory Requires-Approval gate for sensitive steps, which must be manually approved by a human user.

plan-run