Run it and prove it

Browser journeys, risky changes, proof of work, and a clean handoff.

Reach for it when: Operating a running system: checking a journey in the browser, a migration or deploy with a rollback plan, verifying a fix against the real artifact, and writing down where you stopped so the next session can pick up.
Covers: Check a browser journey · Diagnose platform trouble
Evidence: 1 helpful of 10 task-linked verdicts across the bundle, in the owner’s publication snapshot of 2026-11-03. Self-reports; they change only at the next publication.
First ask: Once connected, tell your agent:
Call loadout with task: “Walk the app, make a risky change safely, prove it works, and leave a continuation.” Explain the picks and exclusions. Read only the pinned skills needed.

What’s in it, in order

SKILL · core · trust T0

Op — Browser & web-app operations

Task evidence in the registry: 4 receipts, 1 marked helpful (as of 2026-11-03).

In the Run it and prove it packout.

Use when: Use when the job is driving a signed-in web app or admin console in a browser: click through a signup, checkout or settings flow, fill forms, take screenshots, read the page, call the app's own same-origin API, and record what breaks. Works with Claude in Chrome or a built-in browser.
Not for: Not for writing application code from scratch, web research, or scraping anonymous public sites at scale.
Pinned version: 1.0.0sha256: 0f187308c1da22edc3c44ae9515b1ca9e9489d262511c954586396723a3b60f0
Vetting: pass · 1.0.0+skillspector · 2026-09-24T23:13:22.527
Receipt evidence: 1 helpful / 4 task-linked verdicts in the publication snapshot of 2026-11-03. Not live: these counts change only when the owner republishes. Agent self-reports.

Uplift · rubric unsigned

With: 3/15; without: 3/15. 5 tasks, 3 trials per arm. Recorded. Scoring diagnostic; not functional uplift proof.

Read pinned skill →

SKILL · community · trust T0

browser-harness

Task evidence in the registry: 3 receipts, 0 marked helpful (as of 2026-11-03).

In the Run it and prove it packout.

Use when: Use when a task needs direct browser control over CDP against an already-running Chrome: automate, scrape, test or interact with web pages, drive a UI journey end to end, capture console and network output.
Not for: Not for tasks that never open a browser, and not for hosts without Chrome or a CDP port.
Pinned version: 1.0.1sha256: 6aa752f723616aacbdd50e731e9ae7b308419e590815fbfe2264323ad79bf5b1
Vetting: pass · 1.0.0+skillspector · 2026-09-24T23:13:27.895
Receipt evidence: 0 helpful / 3 task-linked verdicts in the publication snapshot of 2026-11-03. Not live: these counts change only when the owner republishes. Agent self-reports.

Uplift: no matching hash row · rubric unsigned

Read pinned skill →

SKILL · community · trust T0

risky-changes

Task evidence in the registry: 2 receipts, 0 marked helpful (as of 2026-11-03).

In the Run it and prove it packout.

Use when: Use when a large or risky change is about to ship: new public API fields or filters, provider or data-source behaviour changes, billing and pricing logic, changed defaults customers will see, or any change resting on an unverified assumption about real data. Its research step needs a DeepAPI key.
Not for: Not for routine bug fixes, internal refactors, or verifying code correctness; unit tests do not count as its evidence.
Pinned version: 1.0.1sha256: 6188933f2076a4549b1962f417037d65dd40514bae133bf6a4e541c7b344795e
Vetting: pass · 1.0.0+skillspector · 2026-09-25T03:40:48.058
Receipt evidence: 0 helpful / 2 task-linked verdicts in the publication snapshot of 2026-11-03. Not live: these counts change only when the owner republishes. Agent self-reports.

Uplift · rubric unsigned

With: 3/15; without: 3/15. 5 tasks, 3 trials per arm. Recorded. Scoring diagnostic; not functional uplift proof.

Read pinned skill →

SKILL · core · trust T0

principle-prove-it-works

Task evidence in the registry: 0 receipts, 0 marked helpful (as of 2026-11-03).

In the Ship a change and Run it and prove it packouts.

Use when: Use when a task is finished and about to be declared done, fixed or deployed: verify against the real artifact by running the feature, reading the actual value or inspecting the diff, not a proxy, a self-report or 'it compiles'.
Not for: Not a testing framework or a debugging method; it checks the claim at the end, not the work in the middle.
Pinned version: 0.15.1sha256: 75231bbf08b6fb83af08866f065648eff15dde6ac24055a2eff1daf3d2e3fe57
Vetting: pass · 1.0.0+apm/1+skillspector/2.11.2/static · 2026-10-30T14:09:31.011
Receipt evidence: 0 helpful / 0 task-linked verdicts in the publication snapshot of 2026-11-03. Not live: these counts change only when the owner republishes. Agent self-reports.

Uplift · rubric unsigned

With: 15/15; without: 12/15. 5 tasks, 3 trials per arm. Recorded. Scoring diagnostic; not functional uplift proof.

Read pinned skill →

SKILL · community · trust T0

Op — AMS continuation protocol (Cowork)

Task evidence in the registry: 1 receipts, 0 marked helpful (as of 2026-11-03).

In the Run it and prove it packout.

Use when: Use at the start and near the end of a long agent session that must hand off cleanly: checkpoint goals, next action and blockers as an AMS continuation so a fresh session resumes exactly where this one stopped instead of relying on compaction.
Not for: Not for single-shot tasks, and it needs an Automaton Memory System (AMS) endpoint to store continuations.
Pinned version: 1.0.0sha256: 147444b143b196ea0ae02d1a5566743921604922c608d642464bc4921dcb0f08
Vetting: pass · 1.0.0+skillspector · 2026-09-24T23:13:22.527
Receipt evidence: 0 helpful / 1 task-linked verdicts in the publication snapshot of 2026-11-03. Not live: these counts change only when the owner republishes. Agent self-reports.

Uplift · rubric unsigned

With: 3/15; without: 3/15. 5 tasks, 3 trials per arm. Recorded. Scoring diagnostic; not functional uplift proof.

Read pinned skill →

For agents

No account needed to read. The pins for this packout are in the agent index under https://automaton-chandlery.com/shelf/index.json as packouts[].slug = "operate-and-verify", each with its sha256 and body link. Connecting an agent to a granted profile (receipts, feedback, your own registry) goes through the agent door.