proof-of-work

Proof-of-Work Documentation

When to Use

• "Capture a screenshot after deploying the change."
• "Document what you found in the logs."
• "Show me proof this worked."
• "Save the output of that command."
• After any investigation, audit, or multi-step task where a stakeholder needs to verify the outcome.

When Not to Use

• Purely generative tasks (writing code, drafting text) where no external state was observed.
• Tasks where the user has explicitly said output or artifacts are not needed.

Artifact Types

Type	Tool	Use For
Screenshot	agent-browser or playwright-mcp	UI state, browser-rendered output, visual regressions
Log capture	shell / ctx_execute	Terminal output, CI logs, service logs
Script output	shell / ctx_execute	Command results, API responses, test runs

Workflow

1. Identify evidence scope — determine which artifact type(s) are appropriate for the task. Output: evidence plan (screenshot / log / script output, destination path).

2. Capture the evidence — use the appropriate tool (see below). Output: raw artifact saved to disk or embedded in response.

3. Store artifacts — write to the evidence directory using the naming convention. Output: file path(s) confirmed.

4. Reference in response — include the file path and a one-line description of what it shows. Output: summary block appended to the task response.

Capturing Screenshots

With playwright-mcp

// Navigate and capture
await page.goto('https://example.com/dashboard');
await page.screenshot({ path: '.context/evidence/2026-04-07-dashboard.png', fullPage: true });

With agent-browser (computer use)

Use the screenshot action after performing browser interactions:

action: screenshot
save_to: .context/evidence/2026-04-07-<slug>.png

Capturing Logs

# Tail and save service logs
kubectl logs deploy/my-service --tail=100 | tee .context/evidence/2026-04-07-service-logs.txt

# Save CI step output
cat build.log > .context/evidence/2026-04-07-build-output.txt