pi-sdlc

SDLC workflow extension for Pi. Proven pipeline inspired by pi-superagents with rule-based guidance.

Features

• Phased Workflow: spec → plan → execute → verify
• Auto Model Switching: Each phase uses appropriate model tier
• Rule-Based Guidance: AI loads relevant rules from docs/rules/ per task type
• Drift Detection: Pre-check and post-check on every task
• Verification Gates: tests > checklist > build with hard stop on failure
• Memory Integration: Syncs specs and completions to pi-memctx

Installation

# Install dependencies first
pi install pi-memctx
pi install npm:pi-subagents

# Then install pi-sdlc
pi install git:github.com/igun997/pi-sdlc

Required packages:

• pi-memctx - durable memory
• pi-subagents - task delegation

Commands

Command	Phase	Purpose
/sdlc-spec <idea>	Spec	Brainstorm and create feature specification
/sdlc-plan	Plan	Break spec into ordered tasks
/sdlc-execute	Execute	Implement tasks with verification gates
/sdlc-verify	Verify	Final verification with evidence report
/sdlc-settings	Config	Configure model tiers interactively

Model Tiers

Agents use abstract tiers mapped to concrete models:

Tier	Phase	Default Model
reasoning	Spec, Plan	anthropic/claude-opus-4-7
coding	Execute	anthropic/claude-sonnet-4-6
fast	Verify	openai/gpt-4o-mini

Configure in .sdlc/config.json (auto-gitignored):

{
  "sdlc": {
    "modelTiers": {
      "reasoning": { "model": "local-llm/alibaba/qwen-max", "thinking": "high" },
      "coding": { "model": "local-llm/alibaba/qwen3-coder-plus" },
      "fast": { "model": "local-llm/alibaba/qwen-flash" }
    }
  }
}

Rules

Rules are loaded by AI based on task type:

Task Type	Rules
Backend	backend/tdd.md, backend/api-design.md
Frontend	frontend/anti-slop.md, frontend/components.md
Go	golang/patterns.md
Rust	rust/patterns.md
Performance	performance/low-latency.md

Rules live in docs/rules/. AI reads only relevant rules, minimizing token usage.

Workflow

1. Spec Phase

/sdlc-spec Add user authentication with OAuth

• One question at a time
• Creates docs/specs/YYYY-MM-DD-{feature}/spec.md
• Syncs to memctx

2. Plan Phase

/sdlc-plan

• Breaks spec into ordered tasks
• Assigns task types: backend | frontend | mixed
• Initializes plan tracker

3. Execute Phase

/sdlc-execute

• Pre-check: confirm understanding
• Implement: follow rules by type
• Post-check: detect drift
• Gates: tests > checklist > build
• Hard stop on failure

4. Verify Phase

/sdlc-verify

• Full test suite
• Walk every acceptance criterion
• Generate verification report
• Mark spec complete

Configuration

Project Config

.sdlc/config.json in project root (gitignored):

{
  "sdlc": {
    "commands": {
      "sdlc-execute": {
        "autoAdvance": false,
        "useTDD": true,
        "gates": ["tests", "checklist", "build"],
        "onFail": "stop"
      }
    },
    "modelTiers": {
      "reasoning": { "model": "anthropic/claude-opus-4-7" },
      "coding": { "model": "openai/gpt-4o" },
      "fast": { "model": "openai/gpt-4o-mini" }
    }
  }
}

Design References

For frontend tasks, create _references/ folder with:

• README.md - Brand voice, visual foundations
• colors_and_type.css - Design tokens
• ui_kits/ - Component patterns

Anti-slop rules require AI to reference these before creating UI.

Comparison with pi-superagents

Feature	pi-superagents	pi-sdlc
Workflow	Superpowers	SDLC (spec→plan→execute→verify)
Agents	sp-recon, sp-implementer, etc.	Phase-based entrypoints
Rules	Skills injection	Rule files loaded by task type
Subagents	Bounded role delegation	Single agent per phase
Review	Plannotator integration	Verification gates

Credits

• pi-superagents for the proven architecture pattern
• Pi for the foundation