The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is configured with the run_terminal_cmd tool and explicitly describes a workflow for generating and executing shell commands and Python scripts. This is part of its intended 'LIFT' stage for automating proven procedures.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection through its EXTRACT and FOLLOW methods. These methods ingest untrusted data (user-defined Markdown procedures) and use them to generate executable Python code or perform agent actions.
Ingestion points: Data enters the agent context through PROCEDURE.md files and sister script manifests (SISTER.yml).
Boundary markers: The provided templates (PROCEDURE.md.tmpl) lack explicit delimiters or instructions to ignore embedded malicious prompts within the procedure steps.
Capability inventory: The skill utilizes write_file to create scripts and run_terminal_cmd to execute them, creating a path from untrusted text to system execution.
Sanitization: No explicit sanitization or validation of the input procedure content is defined before the code generation process.

sister-script