Phase 1: Project Structure Analysis

Agent: Explore | Action: Scan project and gather intelligence

Context: BC_PLUGIN_ROOT is available in your context (injected by pre-task.mjs hook).

Scan Commands

EXECUTE using Bash tool — gather project info:

bash "scripts/setup.sh" scan && echo "✅ scan" || echo "❌ scan FAILED"

STOP if ❌ — check script exists and plugin is installed.

Phase 2: Intelligence Analysis

Agent: Plan | Action: Consolidate findings and create adaptation strategy

Create Adaptation Plan

Adaptation Plan

Tech Stack

• Language: [Java/Node.js/Python/Go/Rust]
• Framework: [Spring Boot/Express/Django/etc]
• Build: [Maven/Gradle/npm/pip/cargo]

Testing

• Framework: [JUnit 5/pytest/Jest/Go testing]
• Assertion: [AssertJ/Hamcrest/Chai/assert]
• Mocking: [Mockito/unittest.mock/Sinon]
• Data: [DBRider/Testcontainers/fixtures]

Database

• Type: [PostgreSQL/MySQL/MongoDB/Redis/ClickHouse]
• Access: [JOOQ/JPA/Hibernate/Sequelize/SQLAlchemy]

Project Agents

• agent-name: purpose (model)

Key Patterns (from CLAUDE.md)

• Pattern 1
• Pattern 2
• Pattern 3

Template Adaptations

• Update AGENTS section with project agents
• Add tech-specific constraints
• Customize verification checklists
• Add database-specific final review agent

Phase 3: Template Generation

Agent: developer | Action: Generate adapted template at .claude/tasks/templates/PLAN.md.template

Create Structure

EXECUTE using Bash tool:

bash "scripts/setup.sh" structure && echo "✅ structure" || echo "❌ structure FAILED"

STOP if ❌ — verify .claude/tasks directory is writable.

Copy/Update Templates

EXECUTE using Bash tool — sync templates from plugin (always overwrites if changed):

bash "scripts/setup.sh" sync && echo "✅ sync" || echo "❌ sync FAILED"

STOP if ❌ — verify plugin templates exist.

Note: Templates synced from plugin. Rules created once (never overwritten). Review skill adapted by AI.

Template Modifications

Section Adaptation

Agents Add project-specific agents from .claude/agents/

Reference Examples Fill with project's reference files (controllers, services, tests)

Phase V agents Customize reviewer focus for detected testing/code patterns

Final Review Add project agents (db_expert, etc.) if relevant tech detected

Required Sections

Preserve universal structure. Key sections to adapt:

Agents — Add project agents above Core Agents

Reference Examples — R1..RN with project's canonical files

Phases — Each phase has: Agent, Status, Context (C#), Refs (R#)

Phase NV: Verification — 2+ agents, one checks patterns compliance

Final Review — 3+ agents parallel

Context Index — C1..CN task-specific files

Phase 3.5: Copy and Adapt Review Skill

Agent: developer | Action: Copy review skill template and adapt for project

Create Review Skill Directory and Copy Template

EXECUTE using Bash tool — create directory and copy review skill template:

bash "scripts/setup.sh" review && echo "✅ review" || echo "❌ review FAILED"

STOP if ❌ — verify review template exists in plugin.

Adapt Review Skill

Replace placeholders based on Phase 2 analysis:

Placeholder Source Example

{PROJECT_AGENTS_TABLE}

.claude/agents/ scan | db_expert | PostgreSQL | DB layer |

{TECH_SPECIFIC_CHECKS}

Detected tech stack See Tech-Specific Checks below

{PROJECT_RULES}

CLAUDE.md patterns AssertJ rules, Lombok, logging

{MAIN_AGENT}

Project agent or reviewer

reviewer

{TEST_AGENT}

Project agent or tester

tester

{DB_AGENT}

Project agent or sql_expert

db_expert

{CUSTOM_GROUPS}

Additional review groups Security, API validation

{CODEBASE_BLOCKS}

Detected source patterns src/main/** , src/test/**

Tech-Specific Checks Templates

Java/Spring:

Node.js/TypeScript:

Python:

Go:

Validation

EXECUTE using Bash tool — verify review skill:

test -f .claude/skills/brewcode-review/SKILL.md && echo "✅ Review skill created" || echo "❌ Review skill MISSING" grep -q "Tech-Specific|tech-specific|Category.*Checks" .claude/skills/brewcode-review/SKILL.md && echo "✅ Tech checks" || echo "❌ Tech checks MISSING"

STOP if any ❌ — review skill must be created before continuing.

Phase 3.6: Copy Configuration

Agent: developer | Action: Copy configuration template for runtime settings

EXECUTE using Bash tool — copy/update config template:

bash "scripts/setup.sh" config && echo "✅ config" || echo "❌ config FAILED"

STOP if ❌ — verify .claude/tasks/cfg directory exists.

Configuration Options

Setting Default Description

knowledge.maxEntries

100 Max KNOWLEDGE.jsonl entries after compaction

knowledge.maxTokens

500 Max tokens in ## K block injected to agents

knowledge.priorities

["❌","✅","ℹ️"]

Priority order for knowledge entries

stop.maxAttempts

20 Stop attempts before escape mechanism triggers

agents.system

[...] System agents (don't receive ## K injection)

Hooks-only architecture: No external runtime. All context management via Claude Code hooks.

Phase 4: Validation

Agent: developer | Action: Verify template structure

EXECUTE using Bash tool — ALL must pass:

bash "scripts/setup.sh" validate && echo "✅ validate" || echo "❌ validate FAILED"

STOP if any ❌ — go back to "Copy Templates" step and fix.

Validation Report

Check Status

PLAN template .claude/tasks/templates/PLAN.md.template

SPEC template .claude/tasks/templates/SPEC.md.template

KNOWLEDGE template .claude/tasks/templates/KNOWLEDGE.jsonl.template

Config file .claude/tasks/cfg/brewcode.config.json

Project agents [N] from .claude/agents/

Reference Examples [N] canonical files populated

Tech-specific adaptations Testing framework, DB patterns

Review skill .claude/skills/brewcode-review/SKILL.md

Phase 5: Update Global CLAUDE.md Agents

Agent: developer | Action: Update agents section in global CLAUDE.md

Step 1: Collect Agents

EXECUTE using Bash tool:

bash "scripts/setup.sh" agents > /tmp/agents-section.md && cat /tmp/agents-section.md

Output = Ready-to-insert content. Script collects system + global + plugin agents. Internal agents (bc-coordinator, bc-grepai-configurator, bc-knowledge-manager) are automatically excluded.

Step 2: Analyze Existing CLAUDE.md

READ ~/.claude/CLAUDE.md using Read tool.

LLM Analysis — find ALL agent-related sections:

• Agents

, ## Agent Selection , ### Core Agents , ### Global Utility Agents

• Any tables with columns like Agent | Model | Purpose

• Any lists of agent names (developer, tester, reviewer, etc.)

Identify boundaries:

• Start line number of agent section(s)

• End line number (before next unrelated ## heading)

Step 3: Ask User

ASK USER with AskUserQuestion:

• Question: "Found agent sections in ~/.claude/CLAUDE.md. Replace with optimized LLM-friendly format?"

• Options:

• "Yes — replace all agent sections"

• "No — keep current format"

Step 4: If YES — Replace

CRITICAL: Use EXACTLY the content from /tmp/agents-section.md . DO NOT add agents manually — the script already filters internal agents. Internal agents (bc-coordinator, bc-grepai-configurator, bc-knowledge-manager) are excluded by design.

Using Edit tool:

• Read /tmp/agents-section.md to get the exact replacement content

• Find the ## Agents — DELEGATE! section in ~/.claude/CLAUDE.md

• Replace that section with the EXACT content from /tmp/agents-section.md

• Preserve ### Global Skills subsection if it exists (append after agents table)

• Preserve all non-agent content

Key: LLM determines section boundaries, not grep. Content comes from script output.

Output Format

Template Adaptation Complete

Detection

Tech Stack

Adaptations

Templates

Plan template: .claude/tasks/templates/PLAN.md.template Review skill: .claude/skills/brewcode-review/SKILL.md

Usage

/brewcode:spec "Implement feature X" /brewcode:review "Check null safety"