Skip to content

ValentinNikolaev/agent-skills / agent-plugins/skills/verify-impl

verify-impl

Verified

Trace authoritative spec requirements to existing code evidence; not general diff or design review.

FileKindSize
LICENSE.sergeyklay-agentsother11.1 KB
NOTICE.sergeyklay-agentsother965 B
SKILL.mdskill md4.1 KB
agents/openai.yamlconfig254 B
---
name: verify-impl
description: Trace authoritative spec requirements to existing code evidence; not general diff or design review.
---

# Verify Implementation Against Specification

Compare a named implementation scope with an authoritative specification. Produce a traceable conformance verdict without editing code, tests, or the specification. Respond in chat unless a report file is explicitly requested.

## Route precisely

- Use this skill only when a specification and an implementation are both in scope and the user asks whether the code fulfills the spec.
- Use `review-spec` to assess a spec before implementation.
- Use `review-arch` to assess system structure without a requirement-by-requirement contract.
- Use `code-review` to find defects introduced by a diff; that review may cite a spec but does not imply exhaustive conformance.
- Use `plan-review` for a proposed implementation plan.

## Fix the authority and scope

1. Identify the exact spec file or revision and the implementation paths, module, commit, or PR being checked. If either side is unavailable, report what is missing before assigning a verdict.
2. Read applicable repository instructions, binding architecture decisions, and the spec in full. Note conflicts between sources rather than silently choosing one. Treat an explicitly superseded spec as historical, not authoritative.
3. Extract each independently verifiable obligation into a stable ID. Preserve a short exact quote and source location; distinguish `must`, `should`, explanatory notes, and examples. Do not turn every sentence into a requirement or infer a target count from document length.
4. Find the implementation through named paths and symbols, then search relevant alternate paths before claiming absence. Inspect callers, configuration, migrations, and tests when a requirement depends on them.

## Compare each obligation

Record one status per requirement:

| Status | Meaning |
| --- | --- |
| `PASS` | Evidence shows the required behavior or contract. |
| `PARTIAL` | Some required aspects are present; others are not. |
| `DRIFT` | The behavior differs from what the spec describes, in a way the quoted wording does not forbid: a different default, scope, ordering, or mechanism with observable effect. |
| `MISSING` | A bounded search found no implementation of a required behavior. |
| `CONFLICT` | The behavior violates the quoted wording itself; a test written from the quote would fail. |
| `UNVERIFIED` | Available code, runtime access, or evidence cannot settle the question. |

For each non-`PASS` result, cite the spec quote, the current code or test evidence, the scenario affected, and the smallest correction or verification step. Use `MISSING` only after a documented search of plausible implementation paths. A passing test supports `PASS` only when its assertions cover the actual obligation. Do not invent a confidence percentage.

After the first pass, re-read every non-`PASS` requirement and its candidate code path. Check whether another mechanism satisfies it and whether the spec is outdated or internally inconsistent. Record a spec defect separately; do not silently excuse code divergence or treat an invalid spec as immutable truth.

Check cross-cutting contracts only when the spec imposes them: dependency direction, authorization, data safety, lifecycle, concurrency, recovery, and public interfaces. Mark inapplicable dimensions rather than filling them with generic concerns.

## Report

Return the spec identity, implementation scope, evidence limits, and a compact requirement matrix: ID, short requirement, spec location, implementation evidence, status, and impact. Group actionable discrepancies by priority: `P0` immediate safety, security, or irreversible-loss gap; `P1` gap in a `must` that makes the feature incorrect or unsafe; `P2` gap in a `should` or a limited-impact deviation; `P3` minor mismatch. Give one verdict: `CONFORMS` (all applicable obligations supported), `DEVIATES` (confirmed material gap), or `INDETERMINATE` (essential evidence or authority is unresolved). A limited audit must say which requirements were not checked; never imply exhaustive coverage from a sampled matrix.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall