Ccc E2e

/ccc-e2e — End-to-End Pre-Release Assessment

CC Commander · /ccc-e2e · Full-surface confidence before you ship

Composes /ccc-fleet + /ccc-testing to run a full pre-release assessment in parallel across 3 isolated worktrees. One command. Three workers. One verdict.

Session markers

Call mcp__ccd_session__mark_chapter at these phase transitions:

| Trigger | title | summary | |---------|-------|---------| | After user confirms scope, before fan-out | "E2E assessment: " | "3-worker fan-out starting on " | | When all workers have reported back | "E2E workers complete" | "QA: / Unit-TDD: / E2E: " | | After verdict is written | "E2E verdict: " | " Critical, High — verdict: " |

Sidebar chips (spawn_task)

After the verdict is rendered, spawn ONE mcp__ccd_session__spawn_task chip per Critical finding:

• title: imperative fix phrase under 60 chars
• prompt: self-contained — include the worker that reported it, file:line if available, issue summary, and fix direction. The spawned session has no memory of this conversation.
• tldr: 1-2 sentences plain English.

Only Critical findings get chips. High/Medium/Low go into the verdict artifact as TodoWrite items.

Response shape (EVERY time)

1. Brand header

**CC Commander** · /ccc-e2e · Full-surface pre-release confidence

2. Context strip

Four parallel reads (silent on failure):

• git rev-parse --abbrev-ref HEAD → current branch
• git rev-list --count main..HEAD 2>/dev/null → commits ahead
• git diff --shortstat main..HEAD 2>/dev/null → diff size
• test -f package.json && node -p "require('./package.json').version" → version

Render one line: > 🧭 Branch: ` · commits ahead · files changed · version: `

3. Scope picker — AskUserQuestion

question: "What scope should I assess?"
header: "CC Commander E2E"
multiSelect: false
options:
  - label: "🔬 Full assessment (QA + Unit-TDD + E2E)"
    description: "3 parallel workers. ~15-30 min. Use before tagging a release."
    preview: "Most thorough. 3 worktrees, 3 agents, severity-ranked verdict."
  - label: "🧪 QA + Unit-TDD only (no E2E)"
    description: "2 workers. ~8-15 min. Use for non-UI changes."
    preview: "Skips Playwright. Good for API/library changes."
  - label: "🎭 E2E only (Playwright)"
    description: "1 worker. ~5-10 min. Use when fixing a specific UI regression."
    preview: "Runs /ccc-testing e2e-testing + visual-regression only."
  - label: "🏃 Quick QA pass only"
    description: "1 worker. ~3-5 min. Use for sanity check, not release gate."
    preview: "Runs /ccc-testing qa-only. Finds but does not fix."

Recommendation logic (prepend to ONE label):

• Branch ahead of main + version bumped → "Full assessment"
• No UI/CSS files in diff → "QA + Unit-TDD only"
• Only UI/CSS files in diff → "E2E only"
• Small patch, low risk → "Quick QA pass only"

Step 2 — Fan out workers

On scope pick, dispatch the appropriate workers via the Agent tool with run_in_background: true. Each worker runs in its own isolated worktree to prevent merge conflicts and shared-state contamination.

Full assessment (3 workers)

Worker 1 — QA audit (../qa-audit/ worktree):

• Agent, subagent_type: qa-engineer, model: sonnet, run_in_background: true
• Task: "Run /ccc-testing qa-only in this worktree. Report: bug count by severity (Critical/High/Medium/Low), affected files, reproduction steps for any Critical. Return JSON: {worker: 'qa', status: 'pass|fail', findings: [{severity, file, description}], duration_s}"

Worker 2 — Unit-TDD (../unit-tdd/ worktree):

• Agent, subagent_type: qa-engineer, model: sonnet, run_in_background: true
• Task: "Run /ccc-testing tdd-workflow + test-strategy in this worktree. Execute the full test suite. Report: pass/fail counts, coverage delta vs last run, any failing test names. Return JSON: {worker: 'unit-tdd', status: 'pass|fail', tests_pass: N, tests_fail: N, coverage_delta: '+N%|-N%', duration_s}"

Worker 3 — E2E (../e2e/ worktree):

• Agent, subagent_type: qa-engineer, model: sonnet, run_in_background: true
• Task: "Run /ccc-testing e2e-testing + visual-regression in this worktree. Run the Playwright suite and screenshot baseline comparison. Report: pass/fail counts, any visual diffs found. Return JSON: {worker: 'e2e', status: 'pass|fail', tests_pass: N, tests_fail: N, visual_diffs: N, duration_s}"

QA + Unit-TDD (2 workers)

Dispatch Worker 1 + Worker 2 only. Skip Worker 3.

E2E only (1 worker)

Dispatch Worker 3 only.

Quick QA pass (1 worker)

Dispatch Worker 1 only, with task scoped to qa-only (no fixes).

Step 3 — Emit progress card

After dispatching, emit one card:

> 🔄 worker(s) running — ETA ~ min. I'll synthesize when all complete. > 📂 Verdict will land at tasks/reviews/ccc-e2e-.md. > 💡 Use /ccc-e2e status to check in. Or wait — I'll surface the verdict automatically.

Step 4 — Synthesize and write verdict

When all workers report back (or timeout at 30 min per worker), synthesize:

1. Collect all findings. Merge by severity.
2. Compute overall status: FAIL if any worker status: fail OR any Critical finding exists.
3. Compute coverage delta across all workers.
4. Write tasks/reviews/ccc-e2e-.md (create tasks/reviews/ if missing).

Verdict artifact format

# CCC E2E Assessment — 

**Branch:**  · **Version:**  · **Scope:** 
**Workers:**  · **Duration:** 

## Verdict: ✅ PASS / ❌ FAIL

> FAIL if any Critical finding or any worker status is `fail`.

## Worker Summary

| Worker | Status | Tests | Coverage | Duration |
|--------|--------|-------|----------|----------|
| QA audit | ✅/❌ | — | — | Xs |
| Unit-TDD | ✅/❌ | / |  | Xs |
| E2E | ✅/❌ | / | — | Xs |

## Findings

### Critical (N)
- `file:line` — `` — **Worker:** qa

### High (N)
...

### Medium (N)
...

### Low (N)
...

## Next steps
- [ ] 

## Recommendation
Ship / Patch required / Triage needed

When to use

• Before git tag v*.*.0 for a major/minor release
• After merging a feature that touches 5+ files
• When refactoring across module boundaries
• When "feels like we should test more" but unclear where

When NOT to use

• Quick bug-fix patches — use /ccc-review instead
• Single-file changes — use /ccc-testing directly
• Docs-only PRs — skip this, docs don't need E2E
• Hotfixes under time pressure — use /ccc-ship preflight instead

Flow (pseudocode)

1. AUQ → user picks scope

2. /ccc-fleet fan-out:
     Worker 1 (qa): /ccc-testing qa-only in ../qa-audit/
     Worker 2 (unit-tdd): /ccc-testing tdd-workflow in ../unit-tdd/
     Worker 3 (e2e): /ccc-testing e2e-testing+visual-regression in ../e2e/

3. Wait for all (timeout: 30 min per worker)

4. Merge findings → compute verdict:
     FAIL = any Critical OR any worker.status == fail
     PASS = all workers pass, zero Criticals

5. Write tasks/reviews/ccc-e2e-{YYYY-MM-DD}.md

6. Spawn chips for each Critical finding

7. Exit summary: PASS/FAIL + blocker count + artifact path

Depends on

• /ccc-fleet (parallel orchestration + worktree isolation)
• /ccc-testing (testing sub-skill router)
• git worktree (isolation — worktrees must already exist or be created)
• jq / node (for parsing worker JSON reports)

Anti-patterns — DO NOT do these

• Do not run all workers sequentially — always fan-out in parallel
• Do not merge results without checking worker status field
• Do not mark PASS if any Critical finding exists — verdicts must be honest
• Do not auto-fix findings during assessment — this is audit-only
• Do not skip the verdict artifact — /ccc-ship reads from tasks/reviews/
• Do not timeout workers before 30 min — E2E suites are slow

Brand rules

• 🔬 for full assessment · 🧪 for partial · 🎭 for E2E-only · 🏃 for quick pass
• ✅ PASS / ❌ FAIL — bold, on its own line in the artifact header
• Severity badges: 🔴 Critical · 🟠 High · 🟡 Medium · 🟢 Low
• Artifact always written — never output-only
• Never mention the CLI — Desktop-plugin flow

---

Bottom line: one command fans out into up to 3 parallel QA workers, each isolated in its own worktree. Severity-ranked verdict lands in tasks/reviews/. Criticals become spawn_task chips for immediate fix sessions.

---

> ⚙️ Fable contract: plan before build · verifier ≠ worker · prove before alarm · loops need gates · leave durable state — rules/fable-method.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: KevinZai
• Source: KevinZai/commander
• License: MIT
• Homepage: https://commanderplugin.com

Install and usage instructions live in the source repository linked above.