oz-verify-claim

Independently verifies claims and PRs by re-running tests and confirming receipts exist.

Updated Aug 5, 2026
One-click install
npx skills add https://github.com/AmitabhainArunachala/dharma_swarmv.01 --skill oz-verify-claim-amitabhainarunachala
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: oz-verify-claim
Source: https://github.com/AmitabhainArunachala/dharma_swarmv.01/tree/main/.warp/skills/oz-verify-claim
Command: npx skills add https://github.com/AmitabhainArunachala/dharma_swarmv.01 --skill oz-verify-claim-amitabhainarunachala

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Teams often accept "done" claims, PR descriptions, or completion assertions without independent proof, letting phantom artifacts and narrated-as-done work slip through. This Skill provides a decorrelated verifier that re-runs the claimed tests and commands on a separate harness, confirms cited artifacts and spine receipts actually exist, and posts a falsifiable verdict. ## Core Features & Use Cases - Independent claim verification: Decomposes a PR, commit, or "done" assertion into a checklist of runnable commands and required files, then re-executes each one and captures exit codes. - Phantom detection: Confirms every cited artifact file and Runtime Truth Spine receipt exists on disk, flagging narrated-as-done work with missing evidence. - Structured verdicts: Posts a PASS, FAIL, UNPROVEN, or BLOCKED verdict comment via gh pr comment and writes a JSON receipt under reports/loop_closure/oz_verify/. - Use Case: A PR claims "Loop 5b closure shipped with receipt written." The verifier re-runs the closure command, finds it green, but discovers the receipt file is missing — and reports FAIL with the exact evidence row. ## Quick Start Ask the agent to independently verify PR 712 by re-running its tests, confirming its cited artifacts and spine receipts exist, and posting an OZ-VERIFY verdict.

Frequently Asked Questions about oz-verify-claim

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

FAQPage Schema
How do I independently verify a pull request claim?▼

Decompose the PR's claims into a checklist of runnable commands and required files, then re-run each test yourself and confirm every cited artifact exists on disk. Post a PASS, FAIL, or UNPROVEN verdict with the exact commands, exit codes, and evidence rows.

What is a phantom artifact in code review?▼

A phantom artifact is a file or receipt a claim references that does not actually exist on disk — work narrated as done but never produced. The verifier detects phantoms by checking that every asserted file path and spine receipt is present before accepting a claim.

Can the verifier merge or approve a pull request?▼

No. The verifier operates under a hard authority boundary: it may read code, run tests and read-only make checks, and post verdict comments, but it must never merge, approve, push, edit source, or expose secrets. Its verdict is evidence, never promotion.

When should a claim be marked UNPROVEN instead of FAIL?▼

Mark UNPROVEN only when a claim is genuinely unfalsifiable — no runnable check and no artifact to confirm. If a check exists but is inconvenient, run it; UNPROVEN is not a soft pass and must state the reason the claim cannot be tested.

Why must verification run on a separate harness?▼

Running on a different session, agent, or clean checkout decorrelates the verdict from the builder's context, preventing reused shell state or conversation history from contaminating evidence. The verdict must derive only from commands the verifier ran and files it confirmed itself.