The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Bash commands used for health-checking the availability of well-known developer CLI tools such as claude, gemini, codex, and opencode. These checks are performed within isolated temporary directories created via mktemp.
[DATA_EXFILTRATION]: The workflow involves sharing project artifacts and contracts with external CLI agents for review. This is the intended primary purpose of the skill and targets well-known technology services.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data returned from secondary AI models, which constitutes a potential injection surface.
Ingestion points: Step 4 (RECONCILE) processes finding artifacts from external reviewers in SKILL.md.
Boundary markers: The skill suggests using structured ARTIFACT and CONTRACT blocks to separate data from instructions during the review process.
Capability inventory: The agent can execute shell commands to interact with other CLI tools.
Sanitization: The instructions explicitly warn the agent not to interpolate artifacts into shell arguments, recommending the use of stdin or temporary files to prevent command injection.

doubt-early