The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands provided through the --verify and --revise parameters. This is a core functionality intended for running tests (e.g., pytest) or linting tools, but it presents an inherent risk if the agent is directed to execute malicious strings.
[DYNAMIC_EXECUTION]: The skill invokes a local Python script (harness/scripts/loop-runner.py) to manage the control loop. This script handles the deterministic logic, state hashing, and guardrail enforcement.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it processes untrusted data from the workspace and external command outputs.
Ingestion points: Reads workspace files defined by the --state glob patterns and captures the output/exit codes of the shell commands executed via --verify and --revise.
Boundary markers: The SKILL.md does not specify the use of delimiters or specific instructions to the agent to ignore embedded content within the ingested workspace data.
Capability inventory: The skill can execute shell commands, perform file system writes (for logs and wiki updates), and calculate cryptographic hashes of files.
Sanitization: There is no mention of sanitizing or validating the output of the shell commands or the content of the workspace files before they are processed or logged.

loop-runner