harness-termination-check

Validates completion claims through three layers of static, runtime, and system checks.

Updated Jul 29, 2026
One-click install
npx skills add https://github.com/MaiconGambini/opencode-harness-guide --skill harness-termination-check-maicongambini
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: harness-termination-check
Source: https://github.com/MaiconGambini/opencode-harness-guide/tree/main/skills/harness-termination-check
Command: npx skills add https://github.com/MaiconGambini/opencode-harness-guide --skill harness-termination-check-maicongambini

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Agents often declare work done based on assumptions rather than evidence, leading to broken handoffs and false passing states. This Skill enforces a three-layer verification gate before any task can be marked complete or set to passing. ## Core Features & Use Cases - Three-Layer Verification: Checks static analysis results, runtime behavior against acceptance criteria, and system-level startup confirmation before allowing a done claim. - Evidence-Based Fail Protocol: Records exact commands, exit codes, metric values, and thresholds when checks fail, then sets tasks to blocked instead of passing. - Quality Gate Integration: Verifies that harness-quality-gate ran at the required mode and treats missing or stale reports as unverified rather than passed. - Use Case: Before writing a completed handoff at the end of a sprint, run this check to confirm lint passes, acceptance criteria were observed with real command output, and the startup command exits 0. ## Quick Start Run the termination check before marking this task as done and record any failures with their exact command output.

Frequently Asked Questions about harness-termination-check

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify a task is really done before marking it complete?

Run a three-layer check: confirm static checks and lint pass, verify acceptance criteria with actual recorded command output, and confirm the startup command exits 0. If any layer fails, record the exact failure and set the task to blocked.

What evidence is required before setting a task status to passing?

Evidence must contain actual command output, not descriptions. For metrics, record the metric name, its value, and its threshold. A missing, stale, or unconfigured quality gate report counts as unverified, not passed.

What happens when a termination check layer fails?

The fail protocol records the exact command, exit code, and output, sets the task to blocked in the feature state, documents the blocker in the progress file, and prohibits claiming done or writing a completed handoff.

Can unavailable metrics be treated as passing in a quality gate?

No. Every unavailable metric must be listed as a gap rather than treated as green. Breaches are only acceptable in observe mode and must be named in the evidence.

When should regression checks be skipped in a completion check?

Regression checks for previously passing work must either be run or explicitly scoped out with a recorded reason. Skipping without a documented reason fails the runtime behavior layer.