QA

You are performing functional validation of a feature, bugfix, or deployment. Your job is to prove it works (or doesn't) by executing test scenarios and capturing evidence.

Working Agreement

All user-facing questions go through AskUserQuestion (when not Autopilot) — see desplega:ask-user for conventions.

File-review is on by default — invoke it on the QA report when ready (skip only if Autopilot).

Ownership: QA Doc Lives Here, Not in Plans

QA test cases live in QA docs (thoughts/<username|shared>/qa/YYYY-MM-DD-<feature>.md), not inline in plans. When a plan has a ### QA Spec (optional): block, that block links to a QA doc this skill produces.

This skill is reserved for cross-cutting or evidence-heavy QA — not routine per-phase checks. Routine per-phase agent verification belongs in the plan's inline #### Automated QA: bucket, executed by phase-running. Use this skill when:

• The QA work spans multiple phases or the whole feature
• Evidence (screenshots, recordings, logs) needs to live somewhere durable
• A formal verdict (pass/fail/blocked) is required

Invocation paths:

• Planning Step 3 ("Generate QA docs") calls this skill before handoff, for any phase whose QA needs a separate doc.
• Implementing / phase-running call this skill at phase-completion if the phase reports QA Doc: <path>.
• Direct user invocation (/qa) creates a standalone doc not linked to a plan.

QA Approach

Unless Autopilot, ask once at the start:

Note on agent-browser availability: Run command -v agent-browser. It is the agent-browser npm CLI (vercel-labs), a headless Chromium driven over CDP. Also check whether the project's CLAUDE.md specifies a different testing tool. If agent-browser is not available and no alternative is specified, default to manual testing. Never use qa-use unless the user names it.

When to Use

This skill activates when:

• User invokes /qa command
• Another skill references desplega:qa
• User asks to test, validate, or QA a feature
• The implementing skill encounters a phase with a QA spec

Autonomy Mode

At the start of QA, adapt your interaction level based on the autonomy mode:

The autonomy mode is passed by the invoking command. If not specified, default to Critical.

Process Steps

Step 1: Context & Scope

Determine the source of QA specs:

Plan-driven (called from planning/implementing/phase-running) → A plan path + phase reference is supplied. Read the relevant phase, design test cases for the deliverable named in its Overview, and produce the QA doc the plan's QA Spec block will link to.

Separate QA spec document provided → Read the standalone spec, use it as the basis for the QA session.

Direct user invocation, no source → Build test cases from scratch with the user. Use AskUserQuestion to establish:

In all cases, create the QA document from cc-plugin/base/skills/qa/template.md at thoughts/<username|shared>/qa/YYYY-MM-DD-<topic>.md. Use the user's name when known; fall back to thoughts/shared/qa/.

For plan-driven invocation: after the doc exists, update the plan's ### QA Spec (optional): block in the relevant phase to point at the doc path. Do not inline scenarios in the plan.

Step 2: Test Case Design

Define test cases covering:

• Happy path — The main scenario works as intended
• Edge cases — Boundary conditions, empty inputs, large inputs
• Error scenarios — Invalid inputs, permission failures, network errors

If sourced from a plan: Aggregate phase QA specs into the test case list. Augment with additional exploratory cases.

If sourced from a separate spec: Use those scenarios as the starting point and augment with exploratory cases.

If browser automation was selected and agent-browser is available: Design agent-browser steps for each test case (open, snapshot, act on refs, screenshot).

Write all test cases into the QA document's ## Test Cases section.

Step 3: Execute Tests

For browser automation (agent-browser): if an agent-browser skill is installed (the swarm seeds one), follow it. Otherwise:

1. Load the version-matched guide once per session with agent-browser skills get core (do not guess flags from memory), then agent-browser open <url>
2. agent-browser snapshot -i to read the page state and its @eN refs
3. Act on refs (click @eN, fill @eN "text", type, press) and re-snapshot after every navigation or state change; refs are only valid for the snapshot that produced them
4. agent-browser screenshot /tmp/<name>.png for evidence, then agent-browser close
5. To share a screenshot outside the machine, upload it to agent-fs with agent-fs write <path> --file /tmp/<name>.png -m "<what it shows>" (binary-safe; --content mangles PNGs), keep it there as the original, and embed it in a PR or GitHub comment as a GitHub user attachment (github-attach skill): ![caption](https://github.com/user-attachments/assets/<uuid>). Use agent-fs share-create <path> for Slack or Linear. If agent-fs is not configured, report the local path and say the upload was skipped

For manual testing:

1. Present each test case's steps to the user
2. Guide them through execution
3. Use AskUserQuestion to collect results:

For CLI verification:

1. Execute the CLI commands directly
2. Compare output against expected results
3. Record pass/fail automatically

OPTIONAL SUB-SKILL: When a CLI test case needs the same multi-step probe re-run later (regression coverage, post-deploy smoke, multi-environment validation) and no existing script captures it, invoke desplega:script-builder inline. The generated script is committed to the project's scripts/ directory with the PASS/FAIL + /tmp log convention; this QA session uses it for the current run, and future QA/verifying sessions discover it via the auto-added <important if> block in CLAUDE.md.

For each test case, record the actual result and pass/fail status.

Step 4: Capture Evidence

Gather evidence for the QA report:

• Screenshots: Via agent-browser screenshot (kept on agent-fs; GitHub user attachments in a PR, share-create links elsewhere) or user-provided
• Videos: Session recording URLs (Loom, etc.)
• Logs: Console output, error messages, relevant log lines
• External links: Sentry issues, CI/CD runs, Grafana dashboards, PR URLs

Add all evidence to the QA document's ## Evidence section.

Step 5: Record Results

Update the QA document with:

• Actual results for each test case
• Pass/fail/blocked/skipped status per test case
• Evidence links inline with relevant test cases
• Any issues found in the ## Issues Found section with severity tags (critical/major/minor)

Step 6: Verdict

Aggregate results into an overall verdict:

Write a 1-2 sentence summary in the QA document's ## Verdict section. Update the frontmatter status field to match (pass/fail/blocked).

Step 7: Persist QA Knowledge

If the project's CLAUDE.md or agents.md doesn't document how QA is done for this project, offer to add a QA section describing:

• Testing approach used (manual, browser automation, CLI)
• Tools used (agent-browser, Playwright, etc.)
• Common test patterns discovered

Use AskUserQuestion with:

Also persist useful patterns to memory if they emerge across QA sessions.

Review Integration

File-review is on by default (unless Autopilot):

• After the QA report is complete, invoke /file-review:file-review <qa-report-path> for inline human comments
• Process feedback with the file-review:process-review skill

Learning Capture

OPTIONAL SUB-SKILL: If significant insights, patterns, gotchas, or decisions emerged during this workflow, consider using desplega:learning to capture them via /learning capture. Focus on learnings that would help someone else in a future session.

Workflow Handoff

After the QA report is complete, use AskUserQuestion with:

Based on the answer:

• Verify: Invoke the desplega:verifying skill
• Review: Invoke the desplega:reviewing skill on the QA document
• Done: Finalize the QA report