skills.rest
Navdeepgambhir9023
Navdeepgambhir9023COMMUNITY
Explore all skills by this author →
Website𝕏@Navdeep9023GitHub

verify

Verifies goal contracts by running checks and adjudicating criteria against declared confidence thresholds.

1|1|Updated Jun 21, 2026
Search Tags:#verification#acceptance criteria#evidence#quality gate#subagent#contract#confidence threshold
One-click install
npx skills add https://github.com/Navdeepgambhir9023/nav --skill verify-navdeepgambhir9023
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: verify
Source: https://github.com/Navdeepgambhir9023/nav/tree/main/skills/verify
Command: npx skills add https://github.com/Navdeepgambhir9023/nav --skill verify-navdeepgambhir9023

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Agents often claim work is done when it is not. This Skill independently re-checks whether a goal's contract is actually satisfied before allowing closure, blocking premature completion claims. ## Core Features & Use Cases - Mechanical Check Execution: Runs every command declared in the contract's verification_commands list and captures raw output as evidence. - Isolated Verifier Subagent: Dispatches a separate context to judge each acceptance criterion for satisfaction, evidence, and confidence without executor contamination. - Three-State Verdicts: Emits COMPLETE, INCOMPLETE, or UNVERIFIED based on mandatory criteria, evidence freshness, and the declared confidence threshold. - Use Case: Before closing a performance goal, run verification to confirm the p95 latency criterion is met with fresh benchmark evidence at confidence above the strict-mode 0.8 threshold. ## Quick Start Ask the agent to verify the active goal against its contract before closing it.

Frequently Asked Questions about verify

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

FAQPage Schema
How do I verify a goal is complete before closing it?▼

Run the verify skill against the active goal. It loads contract.json, executes all declared verification commands, dispatches an isolated verifier subagent to judge each criterion, and emits a COMPLETE, INCOMPLETE, or UNVERIFIED verdict with a score.

What is the difference between INCOMPLETE and UNVERIFIED verdicts?▼

INCOMPLETE means a mandatory criterion is not satisfied at all. UNVERIFIED means a criterion appears satisfied but confidence is below the declared threshold or no evidence reference exists, requiring fresh evidence or human sign-off.

Does the verifier subagent see the executor's work context?▼

No. The verifier runs in an isolated context containing only the contract, the evidence bundle, and a pointer to the goal directory. It cannot see the executor's diff or conversation, which keeps the judgment independent.

Why does verification fail when tests pass?▼

Passing tests only prove the test suite passes, not that acceptance criteria are met. Verification also checks evidence freshness, per-criterion confidence against the threshold, and whether stale evidence predates recent source edits.

Can the confidence threshold be lowered to make a goal pass?▼

No. The threshold is resolved once at contract intake based on completion_mode (strict 0.8, shippable 0.6, prototype best-effort) and cannot be lowered at verify time. This is the load-bearing rule of the gate.

Related Skills

Explore More Software Engineering Skills→
ogiboyogiboy

performance-goal

Runs an evaluator-gated performance optimization workflow with durable OMX goal artifacts.

Community
Intermediate
LinkDryLinkDry
1

archive

Archive an experiment line while preserving lessons, artifacts, and memory.

Community
Advanced
jongcheol-pakjongcheol-pak
3

plan

Creates an approved plan.md by interviewing requirements, measuring impact scope, and decomposing work into verifiable tasks.

Community
Advanced
faustellar1995faustellar1995
1

review-readiness

Evaluates GitHub pull-request branches for missing checks and recommends whether they are ready to merge.

Community
Basic