The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to detect installed AI models and execute verification scripts.
Evidence: scripts/looper.py executes version probes for local CLIs (e.g., kiro-cli, ollama); templates/run-loop.py executes arbitrary commands defined as 'programmatic' criteria in the loop specification.
[DATA_EXFILTRATION]: The skill manages potential data exposure through automated redaction and explicit consent prompts.
Evidence: The redact_prompt_for_member function in templates/run-loop.py scans local files matching patterns like .env and secrets/** to remove their contents from prompts. The ensure_consent function blocks communication with non-local models until the user provides affirmative consent.
[COMMAND_EXECUTION]: The skill possesses an indirect prompt injection surface (Category 8).
Ingestion points: The skill reads external context files (e.g., process-notes.md) which are interpolated into prompts for the architect and reviewers.
Boundary markers: Relies on standard Markdown headers within generated prompts.
Capability inventory: Includes the ability to execute shell commands via subprocess.run for verification checks and model CLI calls.
Sanitization: Employs shlex.split for command construction and a string-replacement redaction system for sensitive data.

loop-architect