Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Guide for using pagent - a PRD-to-code orchestration tool. Use when users ask how to use pagent, run agents, create PRDs, or transform requirements into code.
.claude/skills/majiayu000-pagent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -49% | 0% |
Pagent orchestrates specialist AI agents to transform Product Requirement Documents (PRDs) into working code.
bash# Interactive TUI (recommended) pagent ui # Run with a PRD file pagent run ./prd.md # Check agent status pagent status
Pagent runs 5 specialist agents in dependency order:
| Agent | Output | Purpose | |-------|--------|---------| | architect | architecture.md | Technical design, API specs, data models | | qa | test-plan.md | Test cases, acceptance criteria | | security | security-assessment.md | Threat model, security requirements | | implementer | code/* | Working code implementation | | verifier | *_test.go, verification-report.md | Tests and validation |
Level 0: architect
Level 1: qa, security (parallel)
Level 2: implementer
Level 3: verifierbash# Run all agents (parallel by default) pagent run ./prd.md # Run specific agents pagent run ./prd.md --agents architect,qa # Sequential mode pagent run ./prd.md --sequential # Resume (skip up-to-date outputs) pagent run ./prd.md --resume # Force regeneration pagent run ./prd.md --force # Custom output directory pagent run ./prd.md -o ./docs/
bashpagent ui # Start fresh pagent ui ./prd.md # Pre-fill with PRD pagent ui --accessible # Screen reader support
bashpagent status # Check running agents pagent logs <agent> # View agent output pagent message <agent> "text" # Send guidance pagent stop <agent> # Stop specific agent pagent stop --all # Stop all agents
bashpagent mcp # Stdio (Claude Desktop) pagent mcp --transport http --port 8080 # HTTP mode pagent mcp --transport http --oauth \ --issuer https://company.okta.com \ --audience api://pagent # With OAuth
Control implementation style:
| Persona | Use Case | |---------|----------| | minimal | MVP, prototype - ship fast | | balanced | Standard projects (default) | | production | Enterprise - comprehensive testing, security |
bashpagent run ./prd.md --persona production
Initialize config:
bashpagent init
Creates .pagent/config.yaml:
yamloutput_dir: ./outputs timeout: 300 persona: balanced preferences: api_style: rest # rest | graphql | grpc language: go # go | python | typescript testing_depth: unit # none | unit | integration | e2e containerized: true include_ci: true stack: cloud: aws compute: kubernetes database: postgres cache: redis
A good PRD includes:
markdown# Product: [Name] ## Problem Statement What problem are we solving? ## Features - Feature 1: description - Feature 2: description ## Requirements - Functional requirements - Non-functional requirements (performance, security) ## Constraints - Technology constraints - Timeline constraints
bashpagent run ./prd.md --agents architect # Review architecture.md, iterate on PRD
bashpagent ui ./prd.md # Select production persona # Run all agents cd outputs/code && go build ./...
bashpagent run ./prd.md --agents architect # Review architecture.md pagent run ./prd.md --resume # Run remaining agents
| Issue | Fix | |-------|-----| | Timeout | pagent run ./prd.md --timeout 600 | | Port in use | pagent stop --all | | Incomplete output | pagent message <agent> "Please complete..." | | Agent stuck | pagent stop <agent> then re-run |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,055 | 11,487 | +127% | 1 | 1 | 0% | 944 | 1,536 | +63% | 0 | 0 | — |
case-02 | fail→pass | 7,676 | 2,375 | -69% | 1 | 1 | 0% | 1,470 | 1,604 | +9% | 0 | 0 | — |
case-03 | fail→pass | 9,756 | 7,296 | -25% | 1 | 1 | 0% | 860 | 1,464 | +70% | 0 | 0 | — |
case-04 | pass→pass | 7,597 | 7,224 | -5% | 1 | 1 | 0% | 1,542 | 1,444 | -6% | 0 | 0 | — |
case-05 | pass→pass | 13,415 | 7,941 | -41% | 1 | 1 | 0% | 1,448 | 1,570 | +8% | 0 | 0 | — |
case-06 | pass→pass | 4,697 | 7,483 | +59% | 1 | 1 | 0% | 865 | 1,423 | +65% | 0 | 0 | — |
case-07 | fail→pass | 8,153 | 1,796 | -78% | 1 | 1 | 0% | 1,379 | 1,306 | -5% | 0 | 0 | — |
case-08 | fail→pass | 15,229 | 1,854 | -88% | 1 | 1 | 0% | 2,643 | 1,343 | -49% | 0 | 0 | — |
case-09 | fail→fail | 11,881 | 1,825 | -85% | 1 | 1 | 0% | 1,249 | 1,337 | +7% | 0 | 0 | — |
case-10 | fail→fail | 14,081 | 2,506 | -82% | 1 | 1 | 0% | 1,607 | 1,314 | -18% | 0 | 0 | — |
case-11 | fail→fail | 13,027 | 1,641 | -87% | 1 | 1 | 0% | 1,432 | 1,370 | -4% | 0 | 0 | — |
case-12 | pass→fail | 10,135 | 7,043 | -31% | 1 | 1 | 0% | 1,798 | 1,358 | -24% | 0 | 0 | — |
case-13 | fail→pass | 12,706 | 6,642 | -48% | 1 | 1 | 0% | 1,341 | 1,265 | -6% | 0 | 0 | — |
case-14 | pass→pass | 16,877 | 1,511 | -91% | 1 | 1 | 0% | 2,341 | 1,333 | -43% | 0 | 0 | — |
case-15 | fail→pass | 17,127 | 7,738 | -55% | 1 | 1 | 0% | 2,302 | 1,599 | -31% | 0 | 0 | — |
case-16 | fail→pass | 13,759 | 6,989 | -49% | 1 | 1 | 0% | 1,593 | 1,409 | -12% | 0 | 0 | — |
case-17 | fail→pass | 13,378 | 7,593 | -43% | 1 | 1 | 0% | 1,356 | 1,443 | +6% | 0 | 0 | — |
case-18 | fail→pass | 13,371 | 1,510 | -89% | 1 | 1 | 0% | 1,493 | 1,277 | -14% | 0 | 0 | — |
case-19 | pass→pass | 9,644 | 7,980 | -17% | 1 | 1 | 0% | 855 | 1,622 | +90% | 0 | 0 | — |
case-20 | fail→pass | 13,574 | 1,815 | -87% | 1 | 1 | 0% | 1,513 | 1,288 | -15% | 0 | 0 | — |
case-21 | pass→pass | 11,924 | 7,965 | -33% | 1 | 1 | 0% | 1,988 | 1,583 | -20% | 0 | 0 | — |
case-22 | fail→fail | 40,557 | 18,461 | -54% | 1 | 1 | 0% | 1,314 | 3,750 | +185% | 0 | 0 | — |
case-23 | fail→pass | 11,551 | 9,835 | -15% | 1 | 1 | 0% | 2,039 | 1,942 | -5% | 0 | 0 | — |
case-24 | pass→pass | 16,170 | 14,977 | -7% | 1 | 1 | 0% | 2,098 | 3,119 | +49% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted, and 22 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +46 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.