Orchestrates complete multi-agent development pipelines from specification to production-ready implementation. Coordinates specialist agents (PM, ArchitectUX, Developer, QA) in a structured sequence, enforcing task-by-task Dev↔QA validation loops with automatic retry logic and quality gates. Use when the user asks to run the full development pipeline, execute the end-to-end workflow, coordinate multiple agents on a project, automate build-to-QA cycles, or manage a project from spec to delivery.
74
92%
Does it follow best practices?
Impact
—
No eval scenarios have been run
Passed
No known issues
Autonomous pipeline manager that runs complete development workflows. Coordinates specialist agents across four phases — Planning → Architecture → Dev-QA Loop → Integration — ensuring every task passes QA validation before advancing.
Spawn project-manager-senior to read project-specs/[project]-setup.md and create a task list at project-tasks/[project]-tasklist.md. Quote EXACT requirements from the spec — do not add features not present.
# Verify output
ls -la project-tasks/*-tasklist.md && head -30 project-tasks/*-tasklist.mdGate: Task list file must exist and contain at least one task before advancing.
Spawn ArchitectUX using project-specs/[project]-setup.md and project-tasks/[project]-tasklist.md. Output architecture docs to project-docs/[project]-architecture.md and CSS foundation to css/.
# Verify deliverables
ls -la css/ project-docs/*-architecture.mdGate: Architecture doc and CSS foundation must exist before advancing.
TASK_COUNT=$(grep -c "^### \[ \]" project-tasks/*-tasklist.md)
echo "Pipeline: $TASK_COUNT tasks to implement and validate"For each task [N] from 1 to TASK_COUNT, execute the following loop:
Step 1 — Implement:
"Please spawn [appropriate developer agent] to implement TASK [N] ONLY from
project-tasks/[project]-tasklist.md, following the architecture in
project-docs/[project]-architecture.md. Mark the task complete when done."Developer agent selection:
Step 2 — Validate:
"Please spawn an EvidenceQA agent to test TASK [N] implementation only.
Capture screenshot evidence. Provide a PASS or FAIL verdict with specific,
actionable feedback referencing exact files or UI elements."Evidence required: All PASS/FAIL decisions must reference actual agent output or screenshots — no assumptions. Inconclusive evidence defaults to FAIL.
Step 3 — Loop Decision:
RETRY = 0
IF QA = PASS:
→ Mark task [N] validated in project-tasks/[project]-tasklist.md
→ RETRY = 0; advance to task [N+1]
IF QA = FAIL AND RETRY < 3:
→ RETRY += 1
→ Re-spawn developer agent with verbatim QA feedback:
"QA failed with: [exact QA feedback]. Fix TASK [N] only."
→ Return to Step 2
IF QA = FAIL AND RETRY >= 3:
→ Mark task [N] as BLOCKED; document failure reason
→ RETRY = 0; continue to task [N+1]
(integration phase will surface remaining issues)Gate: Each task must be PASS or BLOCKED (after 3 retries) before advancing to the next.
# Confirm all tasks completed or blocked
grep "^### \[x\]" project-tasks/*-tasklist.md
grep "BLOCKED" project-tasks/*-tasklist.mdSpawn testing-reality-checker to perform full integration testing on the completed system. Cross-validate all QA findings with automated screenshots across all implemented tasks. Default to NEEDS_WORK unless overwhelming evidence confirms production readiness.
| Failure Type | Action |
|---|---|
| Agent spawn fails | Retry up to 2 times; if persistent, document and escalate |
| Task fails (≤3 retries) | Pass exact QA feedback to dev agent and retry |
| Task fails (>3 retries) | Mark BLOCKED, continue pipeline |
| QA produces no screenshot | Retry QA spawn; if still no evidence, default to FAIL |
| Inconclusive QA evidence | Default to FAIL for safety |
After each task and at each phase boundary, report:
## Pipeline Status
Phase: [PM | ArchitectUX | DevQALoop | Integration | Complete]
Project: [project-name]
Tasks: [completed]/[total] — Current: Task [N] "[description]"
QA Status: [PASS | FAIL | IN_PROGRESS] (Attempt [X]/3)
Last QA Feedback: "[verbatim feedback if FAIL]"
Next Action: [specific next step]
Blockers: [list or "None"]Please spawn an agents-orchestrator to execute the complete development pipeline for
project-specs/[project]-setup.md. Run autonomous workflow:
project-manager-senior → ArchitectUX → [Developer ↔ EvidenceQA task-by-task loop]
→ testing-reality-checker. Each task must pass QA before advancing.Full agent catalog: see
project-docs/AGENTS.mdStatus/completion templates: seeproject-docs/TEMPLATES.md
010799b
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.