@pennyfarthing/core 7.9.3 → 8.0.0

package/pennyfarthing-dist/guides/agent-coordination.md

@@ -0,0 +1,480 @@
+# Pennyfarthing Agent Coordination Architecture
+
+## Overview
+
+This document describes how Pennyfarthing agents are coordinated. The framework supports both single-repo and multi-repo projects.
+
+**Key Principle:** Single entry point (`/new-work` or `/work`), state detection via session files, handoffs via Haiku subagents.
+
+## The TDD Flow
+
+```
+/new-work → SM → TEA → Dev → Reviewer → SM (finish)
+              │     │     │        │
+           setup  tests  impl   review
+              │     │     │        │
+              └─────┴─────┴────────┘
+                 subagent handoffs
+```
+
+**Entry points:** `/new-work` (new story) or `/work` (smart resume/start)
+**State detection:** Agents read session file on activation
+**Handoffs:** Agents spawn Haiku subagents to update session file
+**Finish:** SM handles when status = `approved`
+
+## Architecture Principles
+
+### 1. Single Source of Truth
+- **Agent definitions:** `.pennyfarthing/agents/`
+- **Subagent prompts:** `.pennyfarthing/agents/`
+- **Scope configuration:** `.pennyfarthing/guides/agent-scopes.yaml`
+- **Sprint tracking:** `sprint/current-sprint.yaml`
+- **Session state:** `.session/{STORY_ID}-session.md`
+
+### 2. Hierarchical Agent Structure
+```
+Strategic Agents (Full Scope)
+├── orchestrator  → Orchestrates everything
+├── pm           → Plans across all repos
+├── architect    → Designs across all repos
+└── devops       → Infrastructure and deployment
+
+Tactical Agents (Story-Scoped) - THE TDD FLOW
+├── sm           → Story setup and finish
+├── tea          → Write failing tests (RED)
+├── dev          → Implement to GREEN, create PR
+└── reviewer     → Adversarial code review
+
+Support Agents
+├── tech-writer  → Documentation
+└── ux-designer  → UI design
+```
+
+### 3. Context Loading Strategy
+- **Strategic agents** load full project context (both repos)
+- **Tactical agents** load only target repo context based on active story
+- **Context budget** kept under 500-800 lines per agent
+- **Lazy loading** - load docs only when specific task requires them
+
+## Directory Structure
+
+```
+/$CLAUDE_PROJECT_DIR/
+├── .claude/                              # Pennyfarthing coordination directory
+│   ├── agents/                         # Agent definitions (symlinks to pennyfarthing-dist)
+│   │   ├── orchestrator.md             # Master orchestrator
+│   │   ├── pm.md                       # Product Manager
+│   │   ├── sm.md                       # Scrum Master (+ sm-*.md subagents)
+│   │   ├── architect.md                # System Architect
+│   │   ├── dev.md                      # Developer (+ dev-handoff.md)
+│   │   ├── tea.md                      # Test Engineer (+ tea-handoff.md)
+│   │   ├── reviewer.md                 # Code Reviewer (+ reviewer-*.md subagents)
+│   │   ├── tech-writer.md              # Technical Writer
+│   │   ├── ux-designer.md              # UX Designer
+│   │   └── devops.md                   # DevOps Engineer
+│   │
+│   ├── commands/                       # 42 slash commands (symlinks)
+│   ├── guides/                         # Behavior guides (symlinks)
+│   ├── skills/                         # 18+ knowledge domain skills (symlinks)
+│   ├── scripts/                        # Utility scripts (symlinks)
+│   │
+│   ├── project/                        # Project-specific overrides
+│   │   ├── agents/                     # Agent sidecars (patterns, gotchas, decisions)
+│   │   └── commands/                   # Custom project commands
+│   │
+│   ├── persona-config.yaml             # Project default theme (shared with team)
+│   └── CLAUDE.md                       # Project instructions
+│
+├── .session/                           # Active work sessions
+│   └── {story-id}-session.md           # Session files (one per story)
+│
+├── sprint/                             # Sprint tracking
+│   ├── current-sprint.yaml             # Active sprint and stories
+│   ├── context/                        # Epic technical context
+│   ├── archive/                        # Completed sprints
+│   └── sidecars/                       # Agent learning files
+│
+└── pennyfarthing-dist/                 # Source of truth (if Pennyfarthing project itself)
+    ├── agents/                         # 10 main agents + 14 subagents
+    ├── commands/                       # 42 slash commands
+    ├── guides/                         # Behavior guides and patterns
+    ├── skills/                         # Knowledge domain skills
+    ├── personas/themes/                # 102 themed personas
+    └── scripts/                        # Utility scripts
+```
+
+## Agent Types
+
+### Strategic Agents (Full Scope)
+
+**Characteristics:**
+- See entire project (both repos)
+- Load all context on activation
+- Make cross-repo decisions
+- Coordinate work across repos
+
+**Agents:**
+- **Pennyfarthing Master:** Orchestrates all agents and workflows
+- **PM:** Plans sprints, prioritizes epics, manages backlog
+- **SM:** Creates stories, adds technical context, validates readiness
+- **Architect:** Makes design decisions, defines patterns, ensures consistency
+
+**Context Loading:**
+```yaml
+On Activation:
+  1. sprint/sprint-status.yaml     # Full sprint status
+  2. API/.claude/context.md      # API context
+  3. UI/.claude/context.md       # UI context
+  4. API/docs/epics.md         # Epic definitions (PM only)
+  5. .session/{STORY_ID}-session.md        # Active work
+```
+
+### Tactical Agents (Story-Scoped)
+
+**Characteristics:**
+- Focus on specific repo(s)
+- Load context based on active story
+- Implement/test/document specific features
+- Narrow, deep expertise
+
+**Agents:**
+- **Dev:** Implements features in target repo
+- **TEA:** Tests features in target repo
+- **Tech Writer:** Documents features in target repo
+- **UX Designer:** Designs UI interfaces (UI only)
+
+**Context Loading:**
+```yaml
+On Activation:
+  1. sprint/sprint-status.yaml     # Current sprint (story section)
+  2. .session/{STORY_ID}-session.md        # Active story
+  3. Determine target repo from story "Repos:" field
+  4. Load target repo context:
+     - If API:  API/.claude/context.md
+     - If UI:   UI/.claude/context.md
+     - If Both: Load both contexts
+```
+
+## Context Files
+
+### Base Directory Context
+
+#### `sprint/sprint-status.yaml`
+**Purpose:** Unified sprint tracking for all stories
+**Size:** ~100-200 lines
+**Loaded By:** All agents
+**Content:**
+- Sprint goals
+- All stories (API, UI, Both)
+- Story status (backlog, in-progress, review, done)
+- Story metadata (repos, priority, points, files)
+
+#### `.session/{STORY_ID}-session.md`
+**Purpose:** Current work session context
+**Size:** ~50-100 lines
+**Loaded By:** All agents
+**Content:**
+- Active story ID
+- Target repo(s)
+- Story context
+- Progress notes
+- Done checklist
+
+### Repository Context
+
+#### `API/.claude/context.md`
+**Purpose:** API-specific patterns and structure
+**Size:** ~30-50 lines
+**Loaded By:** Strategic agents + Dev/TEA when working on API
+**Content:**
+- Tech stack (Go, PostgreSQL, MongoDB, Redis)
+- Key patterns (Repository, DI, REST, Middleware)
+- Project structure
+- Quick links to key directories
+
+#### `UI/.claude/context.md`
+**Purpose:** UI-specific patterns and structure
+**Size:** ~30-50 lines
+**Loaded By:** Strategic agents + Dev/TEA/UX when working on UI
+**Content:**
+- Tech stack (React, TypeScript, TailwindCSS)
+- Key patterns (Components, Hooks, Context API)
+- Project structure
+- Quick links to key directories
+
+### Documentation
+
+#### `API/docs/epics.md`
+**Purpose:** All epic definitions for API
+**Size:** ~800+ lines (load summary only)
+**Loaded By:** PM, Architect
+**Content:**
+- Epic descriptions
+- Story breakdowns
+- Technical context
+- Dependencies
+
+## Agent Activation Protocol
+
+### Step 1: Load Agent Definition
+```
+Read: .pennyfarthing/agents/[agent].md
+Identify: Strategic or Tactical agent type
+```
+
+### Step 2: Load Base Context
+```
+Load files from agent-scopes.yaml:
+  - Sprint status
+  - Active work (if exists)
+  - Repo contexts (based on agent type)
+```
+
+### Step 3: Determine Scope (Tactical Agents Only)
+```
+Read: .session/{STORY_ID}-session.md
+Extract: "Repos:" field (API|UI|Both)
+Load: Appropriate repo context
+```
+
+### Step 4: Present Ready State
+```
+Display: Agent persona and menu
+Show: Loaded context summary
+Ready: For user input
+```
+
+## Agent Workflows
+
+### Strategic Agent Workflow (PM Example)
+
+```
+User: @/pm
+
+1. Load PM agent definition (.pennyfarthing/agents/pm.md)
+2. Load all strategic context:
+   - Sprint status (full)
+   - API context
+   - UI context
+   - Epic definitions
+   - Active work
+3. Present PM menu with options
+4. Execute PM tasks (sprint planning, backlog grooming, etc.)
+5. Update sprint status
+6. Hand off to SM or other agents as needed
+```
+
+### Tactical Agent Workflow (Dev Example)
+
+```
+User: @/dev
+
+1. Load Dev agent definition (.pennyfarthing/agents/dev.md)
+2. Load base context:
+   - Sprint status (story section)
+   - Active work
+3. Determine target repo from active work "Repos:" field
+4. Load target repo context:
+   - If API: API/.claude/context.md
+   - If UI: UI/.claude/context.md
+   - If Both: Load both
+5. Present Dev ready state
+6. Implement story in target repo(s)
+7. Update sprint status to 'review'
+8. Hand off to TEA or SM
+```
+
+## Agent Handoffs (TDD Flow)
+
+Handoffs are automated via Haiku subagents in `.pennyfarthing/agents/`.
+
+### The Flow
+
+```
+SM → TEA → Dev → Reviewer → SM
+│      │     │        │       │
+│      │     │        │       └── sm-finish PHASE=execute
+│      │     │        └── handoff VERDICT=approved/rejected
+│      │     └── handoff CURRENT_PHASE=green
+│      └── handoff CURRENT_PHASE=red
+└── sm-setup MODE=setup + sm-handoff
+```
+
+### Complete Subagent Inventory
+
+| Subagent File | Purpose | Model |
+|--------------|---------|-------|
+| **SM Subagents** | | |
+| `sm-setup.md` | Research backlog (MODE=research) or setup story (MODE=setup) | haiku |
+| `sm-finish.md` | Preflight checks (PHASE=preflight) or execute finish (PHASE=execute) | haiku |
+| `sm-handoff.md` | SM → TEA/Dev handoff with Jira/branch verification | haiku |
+| `sm-file-summary.md` | Summarize file changes for commits | haiku |
+| **Shared Subagents** | | |
+| `handoff.md` | Workflow-driven phase transitions (TEA, Dev, Reviewer) | haiku |
+| `testing-runner.md` | Run tests and report results | haiku |
+| **Reviewer Subagents** | | |
+| `reviewer-preflight.md` | Pre-flight checks before review | haiku |
+| **Utility Subagents** | | |
+| `workflow-status-check.md` | Check workflow status across repos | haiku |
+
+### SM → TEA (Story Setup)
+**Trigger:** User selects story via `/new-work`
+**Subagent:** `sm-setup MODE=setup` then `sm-handoff`
+**Action:** Claim Jira, write session file, create branches
+**Handoff phrase:** "TEA, Story X-Y needs tests. Write failing tests for these ACs."
+
+### TEA → Dev (Tests Written)
+**Trigger:** TEA completes failing tests
+**Subagent:** `handoff CURRENT_PHASE=red`
+**Action:** Update session file, transition to green phase
+**Handoff phrase:** "Dev, tests are RED and ready. Make them GREEN."
+
+### Dev → Reviewer (Implementation Complete)
+**Trigger:** Dev creates PR with passing tests
+**Subagent:** `handoff CURRENT_PHASE=green`
+**Action:** Update session file, transition to review phase
+**Handoff phrase:** "Reviewer, PR #N is ready. All tests GREEN."
+
+### Reviewer → SM (Approved)
+**Trigger:** Reviewer approves PR
+**Subagent:** `handoff VERDICT=approved`
+**Action:** Update status to `approved`, transition to finish phase
+**Handoff phrase:** "SM, Story X-Y approved. Run finish-story."
+
+### Reviewer → Dev (Rejected)
+**Trigger:** Reviewer finds issues
+**Subagent:** `handoff VERDICT=rejected`
+**Action:** Update session with issues, route back to implement phase
+**Handoff phrase:** "Dev, {N} issues found. See assessment for details."
+
+### SM → Done (Finish)
+**Trigger:** Status = `approved`
+**Subagent:** `sm-finish PHASE=execute`
+**Action:** Archive session, create summary, update sprint YAML, sync Jira
+
+## Support Agent Handoffs
+
+### Any → Architect
+**Trigger:** Need design decisions or architectural guidance
+**Context:** Technical problem or design question
+**Action:** Architect provides design and implementation guidance
+
+### Any → Tech Writer
+**Trigger:** Feature needs documentation
+**Context:** Implemented feature
+**Action:** Tech Writer creates documentation
+
+### Any → UX Designer
+**Trigger:** Feature needs UI design
+**Context:** User story and requirements
+**Action:** UX Designer creates UI design and specs
+
+## Context Budget Management
+
+### Target Budgets
+- **Agent file:** 200-400 lines
+- **Sprint status:** 100-200 lines (full) or 50 lines (story section)
+- **Repo contexts:** 30-50 lines each
+- **Active work:** 50-100 lines
+- **Total per agent:** 500-800 lines
+
+### Strategic Agent Budget
+```
+PM Agent Example:
+  - pm.md:                    200 lines
+  - sprint-status.yaml:       150 lines
+  - API/context:     30 lines
+  - UI/context:      30 lines
+  - epics.md (summary):       100 lines
+  - active work:               50 lines
+  Total:                      560 lines ✓
+```
+
+### Tactical Agent Budget
+```
+Dev Agent Example (API story):
+  - dev.md:                   300 lines
+  - sprint-status (story):     50 lines
+  - API/context:     30 lines
+  - active work:               50 lines
+  Total:                      430 lines ✓
+```
+
+## Benefits
+
+### ✅ Single Source of Truth
+- All agents defined in one place
+- No duplicate or conflicting agent files
+- Easy to update and maintain
+
+### ✅ Coordinated Planning
+- Strategic agents see full project scope
+- Unified sprint tracking
+- Clear cross-repo dependencies
+
+### ✅ Focused Implementation
+- Tactical agents load only what they need
+- Reduced context overhead
+- Faster agent activation
+
+### ✅ Clear Hierarchy
+- Strategic agents coordinate
+- Tactical agents execute
+- Clean handoffs between agents
+
+### ✅ Scalable
+- Easy to add new agents
+- Context budgets stay manageable
+- Works for solo dev or team
+
+## Architecture History
+
+### Previous Architecture (Pre-December 2025)
+- Separate commands: `/new-work`, `/pickup-work`, `/handoff-work`, `/finish-work`
+- Manual handoff documentation
+- No subagent extraction
+
+### Current Architecture (January 2026)
+- Smart entry point: `/work` (resumes or starts new)
+- Alternative: `/new-work` (explicitly start new story)
+- State detection via session file in `.session/`
+- Handoffs via Haiku subagents in `.pennyfarthing/agents/`
+- SM handles finish-story when status = `approved`
+- 102 themed personas for agent personality
+
+### Directory Structure
+```
+.pennyfarthing/agents/             # Agent definitions (symlinked)
+.claude/commands/           # 42 slash commands
+.claude/skills/             # 18+ knowledge domain skills
+.session/                   # Session files
+sprint/                     # Sprint tracking
+```
+
+### Commands Reference
+```bash
+# Entry points
+/work          # Smart entry - resume or start new
+/new-work      # Explicitly start new story
+
+# TDD Flow agents
+/sm            # Scrum Master (setup + finish)
+/tea           # Test Engineer (RED phase)
+/dev           # Developer (GREEN phase)
+/reviewer      # Code Reviewer
+
+# Support agents
+/architect     # Architecture design
+/tech-writer   # Documentation
+/ux-designer   # UI/UX design
+/devops        # Infrastructure
+
+# Utility
+/check         # Run quality gates before handoff
+/chore         # Quick commit for small changes
+/release       # Merge develop to main
+```
+
+---
+
+**Smart entry. State detection. Subagent handoffs. Themed personas.**