skills.rest
douglasjarquin
douglasjarquinCOMMUNITY
Explore all skills by this author →
Website𝕏@douglasjarquinGitHub

evidence

Capture and compare before/after screenshots, screencasts, and CLI transcripts across Git revisions.

Updated Sep 5, 2026
Search Tags:#browser automation#git worktree#screenshot comparison#visual verification#evidence capture#screencast#github pull request
One-click install
npx skills add https://github.com/douglasjarquin/sum --skill evidence-douglasjarquin
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: evidence
Source: https://github.com/douglasjarquin/sum/tree/main/.agents/skills/evidence
Command: npx skills add https://github.com/douglasjarquin/sum --skill evidence-douglasjarquin

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve? Proving that a bug fix or feature actually works requires showing the user path failing before and passing after, but screenshots alone are easy to fake or misattribute. This Skill captures truthful, hash-verified before/after evidence from two separate Git checkouts and compares them into one auditable manifest. ## Core Features & Use Cases - Real Browser Capture: Drives a headless Chromium browser over the DevTools Protocol with real mouse and keyboard events, recording screenshots, screencasts, and visible-text assertions bound to the exact commit SHA. - CLI and HTTP Recipes: Records command transcripts with exit codes and HTTP responses with status codes for nonvisual changes, with automatic redaction of tokens, keys, and emails. - Honest Comparison Manifests: Produces red-green, mismatch, capture-failed, or before-unavailable verdicts; blocked captures are never simulated or converted into passes. - PR Publication: Publishes validated media into a pull request's marked before-and-after block via the GitHub CLI, with content-hash receipts, foreign-block detection, and read-back verification. - Use Case: After fixing a failing counter button, capture the broken behavior from the base commit and the working behavior from the candidate branch, then attach the verified screenshot pair to the PR for reviewers. ## Quick Start Ask the agent to capture before and after evidence for a scenario from VERIFY.md by running the base and candidate builds in separate worktrees and comparing the captures into one manifest.

Frequently Asked Questions about evidence

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

FAQPage Schema
How do I capture before and after screenshots for a bug fix?▼

Check out the base and candidate revisions into separate Git worktrees, run each build on its own port, then run evidence_capture.py capture twice with the same run id and roles before and after. Finish with the compare subcommand to produce a red-green verdict manifest.

How do I publish before/after evidence to a GitHub pull request?▼

Use evidence_publish.py to plan a publication from a comparison run, then publish it with gh pr edit --attach, which requires GitHub CLI 2.99 or newer. The script validates hashes, checks for secrets, and edits only the marked before-and-after block in the PR body.

What are the requirements for browser-based evidence capture?▼

Browser capture needs Node 22 or newer for the built-in WebSocket and a Chromium-family browser binary, discoverable automatically or set via EVIDENCE_BROWSER. The capabilities command reports what the current machine supports, and unsupported recipes are recorded as blocked rather than simulated.

Can I capture evidence for CLI or API changes without a browser?▼

Yes, the cli recipe records a command transcript with its exit code and the http recipe records status and body, both marked visual_proof not-applicable for nonvisual kinds. Assertions on exit codes, output text, and response status are recorded in the same manifest format.

What happens when the base build cannot be run for comparison?▼

Record it honestly with the unavailable subcommand and a reason instead of fabricating a failing screenshot. The comparison is then labeled after-only with before-unavailable, and the evidence remains publishable with a Preview label.

Does evidence capture handle secrets in transcripts and screenshots?▼

Recorded text is redacted for bearer tokens, API keys, passwords, and emails, with custom patterns added via --redact. Screenshots cannot be redacted, so captures containing redacted text are refused during PR publication and synthetic data should be used instead.

Related Skills

Explore More Software Engineering Skills→
withmartian-sandbox-darksidewithmartian-sandbox-darkside

using-git-worktrees

Create isolated git worktrees for parallel feature development without branch switching.

Official
Intermediate
real-casereal-case

pr-merge

Merge a GitHub pull request via gh and return to the updated base branch.

Community
Intermediate
element22llcelement22llc
2

preflight

Run automated pre-commit and pre-push verification checks on e22-plugins using mise and Bash scripts.

Official
Basic
telum-aitelum-ai
2

visual-testing-desktop-tauri

Automate visual regression testing for Tauri apps with WebdriverIO and tauri-driver.

Official
Advanced