The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core functionality involves reading and executing commands from a .coord/plan.yml file, which could be generated or modified by untrusted agents or external inputs.
Ingestion points: .coord/plan.yml (specifically the success_criteria field) and .ai/*.result.json files are used to drive the gating logic.
Boundary markers: The skill lacks explicit boundary markers or "ignore embedded instructions" warnings when processing the plan's success criteria.
Capability inventory: The skill can execute arbitrary shell commands (e.g., pytest, mypy, npm test, grep) and invoke other skills based on the contents of the plan file.
Sanitization: There is no evidence of command sanitization or validation to ensure that the success_criteria strings do not contain malicious payloads.
[COMMAND_EXECUTION]: The skill is designed to execute shell commands to verify project state and code quality.
Evidence: In SKILL.md and references/acceptance_checklist.md, the workflow describes translating assertions into shell commands like [ -f path ] && grep -q pattern and running test suites like pytest and ruff.
[DYNAMIC_EXECUTION]: The skill builds and runs commands at runtime by interpolating parameters into command templates.
Evidence: The multi-locale-mirror-sync.yml preset allows the execution of a user-defined script via the anchor_script parameter: python {anchor_script} --strict.

agent-acceptance-gate