verify

Runs a repository's VERIFY.md verification contract and records candidate-bound evidence.

Updated Sep 5, 2026
One-click install
npx skills add https://github.com/douglasjarquin/sum --skill verify-douglasjarquin
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: verify
Source: https://github.com/douglasjarquin/sum/tree/main/.agents/skills/verify
Command: npx skills add https://github.com/douglasjarquin/sum --skill verify-douglasjarquin

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Verifying that a code change actually works is often ad-hoc: checks get skipped, manual scenarios go unreported, and passing claims lack evidence. This Skill enforces a repository-owned VERIFY.md contract so every verification run produces a trustworthy, reviewable record bound to an exact commit. ## Core Features & Use Cases - Contract-driven verification: Reads the project-root VERIFY.md, validates its structure, confirms the canonical mise run verify task belongs to this repository, and runs it with a timeout. - Evidence capture: Records per-feature command and HTTP evidence (exact command, output, expectations met) into a run directory that teardown never deletes. - Truthful outcome reporting: Distinguishes pass, fail, blocked, not-run, and not-applicable; flags dirty trees as provisional and policy-file changes as requiring root review, so a candidate cannot certify itself. - Use Case: After fixing a bug on a branch, run the verify skill to execute the project's checks, capture evidence for the affected feature, and produce a run record a reviewer can inspect before merging. ## Quick Start Ask the agent to verify this repository by running the verify skill, which reads VERIFY.md at the project root and executes the canonical mise run verify task while recording the results.

Frequently Asked Questions about verify

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

FAQPage Schema
How do I verify a repository with a VERIFY.md contract?

Run verify_run.py from the skill's scripts directory at the project root. It reads VERIFY.md, validates the contract and feature maps, confirms the mise verify task belongs to this repository, runs `mise run verify`, and writes a run record under the artifacts directory.

How do I capture evidence for a manual test scenario?

Use verify_capture.py with the feature ID and either a run recipe (a command with expected exit code or output text) or an http recipe (a request with expected status or body text). Each capture writes a JSON evidence file into the run directory's evidence folder.

Does the verify skill work without sum or Herdr installed?

Yes, the skill is fully portable and works in any ordinary clone. It only requires Git, mise, and Python 3.11 or newer; everything it needs lives inside the repository being verified.

Why is my verification run marked blocked instead of failed?

Blocked means the run could not be trusted: VERIFY.md is missing or malformed, the verify task is missing or inherited from a parent project, a required command is absent, or the run timed out. Blocked is never reported as a pass.

What does it mean when a run requires root review?

A run requires root review when policy files such as VERIFY.md, mise tasks, or feature maps changed relative to the --base ref, or when no base was provided. Such a run cannot certify a commit SHA, since a candidate must not modify the rules that judge it.

Can a dirty working tree produce a certified verification result?

No. A dirty tree produces a provisional record that is not bound to an exact commit SHA. Commit your changes first when you need a result that certifies a specific SHA.