completion-evaluator_skill

Overview

This skill evaluates deliverables against defined acceptance criteria and returns objective pass/fail verdicts with evidence and actionable feedback. It is technology-agnostic and intended to produce binary decisions, clear remediation steps, and next-action assignments. The output is structured for automation and human review.

How this skill works

The skill loads acceptance criteria and builds an evaluation checklist across completeness, quality, and functionality indicators. For each indicator it gathers evidence (code presence, tests, static analysis, coverage, benchmarks, or manual review), compares measurements to thresholds, and records PASS/FAIL with supporting proof. A binary overall verdict is computed: any failed indicator yields FAIL; all indicators passing yields PASS. The result includes prioritized feedback, assigned agents for rework, and iteration guidance.

When to use it

• Final verification before marking a task complete
• Automated gate in CI pipelines or workflow orchestrators
• Peer-review step to produce evidence-backed feedback
• Decision point after an implementation or delivery
• Quality checks when acceptance criteria are supplied or default standards apply

Best practices

• Require explicit acceptance criteria where possible; fall back to default indicators only when none are provided
• Collect concrete, reproducible evidence for every indicator (test output, linter logs, coverage report, benchmarks, or review notes)
• Provide prioritized, specific, and actionable remediation items with assigned agents
• Keep the verdict binary: avoid conditional pass states to reduce ambiguity
• Track iteration counts and escalate to the user after the configured maximum iterations

Example use cases

• Automated evaluation of a pull request against project standards and acceptance tests
• Post-delivery check that all documentation, tests, and error handling are present
• Gatekeeper in a release pipeline deciding whether to proceed to deployment
• Quality reviewer issuing prioritized defects and assigning responsible agents for rework
• Iteration management when multiple fix cycles are allowed before escalation