verify

Determines whether diagnosed code behavior is a bug or intended design.

Updated Sep 8, 2026
One-click install
npx skills add https://github.com/Thigh-Food/emdash --skill verify-thigh-food
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: verify
Source: https://github.com/Thigh-Food/emdash/tree/main/.flue/skills/verify
Command: npx skills add https://github.com/Thigh-Food/emdash --skill verify-thigh-food

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When triaging issue reports, a diagnosed code path that explains a symptom is not automatically a bug. This Skill prevents wasted fix work by gating the fix stage: it cross-references code, comments, documentation, and tests to decide whether the behavior is a genuine defect, intentional design, or genuinely ambiguous. ## Core Features & Use Cases - Three-verdict classification: Returns exactly one of bug, intended-behavior, or unclear, each with cited reasoning pointing to specific comments, docs, or tests. - Evidence-based cross-referencing: Reads surrounding code, docstrings, AGENTS.md, CONTRIBUTING.md, docs/, and existing tests to establish documented intent before judging. - Read-only safety: Enforces hard prohibitions on source edits, commits, pushes, and external network calls, so verification never mutates the repository. - Use Case: A reporter claims the API should return a bare array, but the code returns { items, nextCursor }. The Skill finds the pagination contract documented in AGENTS.md, returns intended-behavior, and drafts a comment pointing the reporter to the docs. ## Quick Start Ask the agent to verify whether the diagnosed behavior in the current issue is a bug or intended behavior, citing the relevant docs and tests.

Frequently Asked Questions about verify

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

FAQPage Schema
How do I decide if reported behavior is a bug or intended design?

Cross-reference the diagnosed code against comments, docstrings, AGENTS.md, docs, and existing tests. If the code matches documented intent, it is intended behavior; if it contradicts documented or clearly implied intent and the reporter's expectation is reasonable, it is a bug.

How to triage GitHub issues before writing a fix?

Run a diagnosis to locate the explaining code, then verify intent by reading surrounding code, documentation, and tests. Only proceed to the fix stage when the verdict is bug with high confidence; otherwise post a comment citing the relevant docs.

Does an existing test prove the behavior is intentional?

Not necessarily. A test asserting the current behavior is a strong signal of intent, but a test that asserts wrong behavior is itself part of the bug. Read what the test asserts and why before treating it as proof of intent.

What happens when documentation is silent about the behavior?

Return an unclear verdict and list what information would be needed to decide. The maintainer then makes the call rather than guessing, avoiding both false fixes and wrongly dismissed reports.

Can the verification step modify code or run tests?

No. Verification is strictly read-only: no source edits, no commits or pushes, no test runs, and no external network calls. It only reads code, docs, and tests to produce a verdict with cited reasoning.