The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script to manage state and transitions. It passes user-supplied file paths to this script via shell commands.
Evidence: Shell commands such as python3 ~/Documents/agent-skills/scripts/plan_runner.py normalize "$ARGUMENTS" --write are used throughout the workflow.
[INDIRECT_PROMPT_INJECTION]: The skill operates by reading and executing tasks from a plan.md file, which creates a vulnerability surface if the file content is untrusted.
Ingestion points: The skill parses plan.md files specified by the user in the $ARGUMENTS variable.
Boundary markers: There are no specific delimiters or instructions provided to the agent to ignore potentially malicious content embedded within the plan's task descriptions.
Capability inventory: The skill is permitted to use powerful tools including Bash, Agent, and Read, which could be abused if a plan contains malicious instructions.
Sanitization: Although the transition logic is implemented in Python, the content of the tasks is interpreted and executed by the LLM.

plan-run