@namch/agent-assistant 1.3.0 → 1.3.2

package/documents/knowledge-domain/04-business-rules.md

@@ -0,0 +1,554 @@
+# Agent Assistant — Business Rules
+
+> **Purpose**: Complete business logic documentation — governance laws, execution rules, state machines, and constraint systems
+> **Parent**: [00-index.md](./00-index.md)
+> **Last Updated**: 2026-03-26
+> **Generated By**: docs-core skill
+
+---
+
+## Table of Contents
+
+1. [Orchestration Laws (L1-L10)](#orchestration-laws-l1-l10)
+2. [Tiered Execution Rules](#tiered-execution-rules)
+3. [Phase Sequencing](#phase-sequencing)
+4. [HSOL Fitness Scoring](#hsol-fitness-scoring)
+5. [Trust Progression](#trust-progression)
+6. [Error Classification (E1-E4)](#error-classification-e1-e4)
+7. [Golden Triangle Debate Rules](#golden-triangle-debate-rules)
+8. [Deliverable Integrity](#deliverable-integrity)
+9. [Language Compliance](#language-compliance)
+10. [Constraint Propagation](#constraint-propagation)
+11. [Anti-Pattern Registry (A1-A10)](#anti-pattern-registry-a1-a10)
+
+---
+
+## Orchestration Laws (L1-L10)
+
+These 10 laws are immutable governance rules defined in `rules/CORE.md` v4.1. They constrain all Orchestrator and agent behavior. No law can be overridden by any command, variant, or agent.
+
+| Law | Name | Rule | Enforcement |
+|-----|------|------|-------------|
+| **L1** | Single Point of Truth | The platform entry file loads `CORE.md` first; all other rules load on-demand | Prevents conflicting instructions from multiple sources |
+| **L2** | Requirement Integrity | 100% fidelity — parse EVERY requirement, zero loss, no assumptions | Requirements Registry must capture all items before Phase 1 |
+| **L3** | Explicit Loading | State what you loaded before using it | Prevents phantom references to unloaded rules/agents |
+| **L4** | Deep Embodiment | When embodying an agent (TIER 2), follow that agent's Directive + Protocol + Constraints exactly | Agent file overrides default AI patterns |
+| **L5** | Sequential Execution | Phase N must complete (exit criteria met) before Phase N+1 starts | No parallel phase execution, no phase skipping |
+| **L6** | Language Compliance | Respond in user's language; code, comments, and report files always in English | Dual-language protocol |
+| **L7** | Recursive Delegation | Meta agents (tech-lead, planner) coordinate only — they NEVER implement directly | Prevents single points of failure from meta agents coding |
+| **L8** | Stateful Handoff | Prior phase deliverables are IMMUTABLE constraints for subsequent phases | No modifying decisions made by earlier agents |
+| **L9** | Constraint Propagation | Scouter → Planner → Implementer chain — constraints tighten, never loosen | Downstream phases cannot relax upstream decisions |
+| **L10** | Deliverable Integrity | Files created by an agent define the standard — the deliverable IS the truth | Output artifacts are authoritative |
+
+### Law Dependencies
+
+```mermaid
+graph LR
+    L1[L1: Single Point of Truth] --> L3[L3: Explicit Loading]
+    L2[L2: Requirement Integrity] --> L5[L5: Sequential Execution]
+    L5 --> L8[L8: Stateful Handoff]
+    L8 --> L9[L9: Constraint Propagation]
+    L4[L4: Deep Embodiment] --> L7[L7: Recursive Delegation]
+    L9 --> L10[L10: Deliverable Integrity]
+    L6[L6: Language Compliance]
+```
+
+---
+
+## Tiered Execution Rules
+
+The Orchestrator delegates work to agents through a mandatory two-tier system. TIER 1 is always preferred; TIER 2 is fallback only.
+
+### Tier Definitions
+
+| Tier | Name | When | Mechanism | Context |
+|------|------|------|-----------|---------|
+| **TIER 1** | Sub-agent | `runSubagent` tool exists on the platform | Isolated sub-agent invocation with fresh context | Isolated from parent |
+| **TIER 2** | Embody | Sub-agent tool unavailable or returned error | Orchestrator reads agent file and acts as that agent | Shared with parent |
+
+### Tier Decision Rules
+
+```mermaid
+stateDiagram-v2
+    [*] --> ToolDiscovery: Delegation needed
+    ToolDiscovery --> CheckSubagent: First delegation only
+    CheckSubagent --> TIER1: runSubagent available
+    CheckSubagent --> TIER2: runSubagent unavailable
+    TIER1 --> VerifyOutput: Sub-agent returns
+    TIER1 --> TIER2: Sub-agent error
+    TIER2 --> EmbodyAgent: Read agent file completely
+    EmbodyAgent --> ExecuteAsAgent: Follow Directive + Protocol + Constraints
+    ExecuteAsAgent --> VerifyOutput: Agent work complete
+    VerifyOutput --> [*]: Exit criteria met
+```
+
+### Prohibitions
+
+| Forbidden | Correct Behavior |
+|-----------|-----------------|
+| Using TIER 2 when `runSubagent` is available | Attempt TIER 1 first |
+| Skipping TIER 1 because task seems "simple" | Always try TIER 1 regardless of task complexity |
+| Using TIER 2 for "efficiency" or "token savings" | TIER 1 is mandatory when available |
+
+### TIER 2 Embody Protocol
+
+1. Log: `⚠️ TIER 2: {reason for fallback}`
+2. Read agent file completely
+3. Extract: Directive, Protocol, Constraints, Output Format
+4. Announce embodiment with prescribed format
+5. Execute as agent following their protocol
+6. Exit embodiment, resume as Orchestrator
+
+### Tool Discovery Caching
+
+Tool discovery (checking if `runSubagent` exists) runs once per session. The result is cached — subsequent delegations skip the check and use the cached tier.
+
+---
+
+## Phase Sequencing
+
+Workflow execution follows strict sequential phase ordering governed by Law L5.
+
+### Execution Loop
+
+```mermaid
+flowchart TD
+    PRE[Pre-Flight: Load CORE.md + PHASES.md + AGENTS.md] --> REQ[Requirements Intake: Parse ALL requirements into Registry]
+    REQ --> P1[Phase 1: Emit header, load agent, delegate, verify exit criteria]
+    P1 --> D1{Deliverable > 150 lines?}
+    D1 -->|Yes| CHUNK1[Write chunked folder]
+    D1 -->|No| SINGLE1[Write single file]
+    CHUNK1 --> P2[Phase 2: Read prior deliverable as constraint, delegate next agent]
+    SINGLE1 --> P2
+    P2 --> PN[Phase N: Same pattern...]
+    PN --> DELIVER[Deliver final result to user]
+```
+
+### Requirements Registry
+
+Before any phase executes, ALL user requirements must be parsed:
+
+```
+| ID | Requirement | Priority | Status |
+|----|-------------|----------|--------|
+| R1 | {extracted} | H/M/L    | ⏳     |
+```
+
+Rule: 100% fidelity — every requirement extracted, no assumptions, no omissions (Law L2).
+
+### Phase Execution Rules
+
+| Rule | Description |
+|------|-------------|
+| One phase at a time | Execute Phase 1 → Phase 2 → ... sequentially. No batching. |
+| Load only current phase's needs | Do not pre-load agents for Phase 2 while executing Phase 1 |
+| Prior deliverables are immutable | Read completely, lock as constraint (Law L8) — do not modify |
+| Missing required deliverable | Halt with notice, create via appropriate agent first, then resume |
+| Emit before execute | Announce phase start before doing work |
+| Verify after execute | Check exit criteria after agent completes |
+| Continue in same reply | Do not stop between phases — execute all in one response |
+
+### Deliverable Size Management
+
+| Condition | Strategy |
+|-----------|----------|
+| <= 150 lines AND < 4 sections | Single file |
+| > 150 lines OR >= 4 sections | Chunked folder with `00-index.md` |
+
+---
+
+## HSOL Fitness Scoring
+
+The Hybrid Skill Orchestration Layer calculates a fitness score to determine which skills to inject into an agent for a given task.
+
+### Formula
+
+```
+fitness = 0.35 × SEMANTIC_MATCH
+        + 0.25 × SPECIFICITY
+        + 0.20 × TRUST_LEVEL
+        + 0.10 × FRESHNESS_SCORE
+        + 0.10 × SUCCESS_RATE
+```
+
+### Weight Rationale
+
+| Factor | Weight | Why |
+|--------|--------|-----|
+| Semantic Match | 0.35 | Highest weight — relevance to the actual task is most important |
+| Specificity | 0.25 | Specialized skills outperform general ones |
+| Trust Level | 0.20 | Matrix skills (trusted) preferred over unproven community skills |
+| Freshness | 0.10 | Newer/recently verified skills may reflect current best practices |
+| Success Rate | 0.10 | Historical performance indicates reliability |
+
+### Decision Thresholds
+
+| Fitness Range | Decision | Discovery Action |
+|---------------|----------|-----------------|
+| >= 0.8 | Matrix Sufficient | Execute with matrix skills, skip dynamic discovery |
+| 0.75 — 0.8 | Matrix Adequate | Execute with matrix skills, run async discovery as background recommendation |
+| < 0.75 | Matrix Insufficient | Blocking discovery — must wait for `find-skills` result before proceeding |
+
+The superiority delta of 0.15 means a dynamic skill must exceed the best matrix skill's fitness by at least 0.15 to be preferred over it.
+
+### Variant-Discovery Matrix
+
+| Variant | Dynamic Discovery |
+|---------|-------------------|
+| `fast` | Always skipped — matrix only |
+| `hard` | Enabled — full resolution |
+| `team` | Enabled — full resolution |
+
+### Complexity Gate
+
+| Assessment | Action |
+|------------|--------|
+| Simple | Skip HSOL resolution — base knowledge sufficient |
+| Complex | HSOL resolution mandatory — skipping is a protocol violation |
+
+---
+
+## Trust Progression
+
+Dynamic skills (community-installed via `find-skills`) progress through a trust lifecycle.
+
+### State Machine
+
+```mermaid
+stateDiagram-v2
+    [*] --> NEW: Skill installed
+    NEW --> EVALUATING: 3 successful executions
+    EVALUATING --> VALIDATED: 10 successful executions
+    VALIDATED --> PROMOTED: Auto-promote to matrix
+    PROMOTED --> [*]: Added to matrix-skills YAML
+
+    NEW --> BLOCKED: Failure or policy violation
+    EVALUATING --> BLOCKED: Success rate drops below 0.85
+    VALIDATED --> BLOCKED: Inactive > 30 days
+
+    note right of NEW: Trust = 0.3
+    note right of EVALUATING: Trust = 0.5
+    note right of VALIDATED: Trust = 0.7
+    note right of PROMOTED: Trust = 1.0
+```
+
+### Trust Values
+
+| State | Trust Score | Gate to Next State |
+|-------|-------------|-------------------|
+| NEW | 0.3 | 3 successful executions |
+| EVALUATING | 0.5 | 10 successful executions |
+| VALIDATED | 0.7 | Auto-promote when all promotion criteria met |
+| PROMOTED | 1.0 | Added to matrix — lifecycle complete |
+| BLOCKED | N/A | Manual review required |
+
+### Promotion Criteria (All Required)
+
+| Criterion | Threshold |
+|-----------|-----------|
+| Minimum executions | >= 10 |
+| Minimum success rate | >= 0.85 (85%) |
+| Maximum inactive days | <= 30 |
+| Automatic | Yes — no human review queue |
+
+### Trust Decay
+
+Trust decays after 90 days of inactivity. This prevents stale skills from maintaining high trust scores.
+
+### Support States
+
+| State | Meaning |
+|-------|---------|
+| `supported` | Actively maintained and functioning |
+| `experimental` | New or untested — use with caution |
+| `blocked` | Known issues or policy violations — do not use |
+
+---
+
+## Error Classification (E1-E4)
+
+Defined in `rules/ERRORS.md`, the error classification system ensures every error leads to either successful completion or explicit user decision. Silent halts are forbidden.
+
+### Error Types
+
+```mermaid
+stateDiagram-v2
+    [*] --> Detected: Error occurs
+    Detected --> Classified: Classify E1/E1b/E2/E3/E4
+
+    Classified --> E1_Retry: E1 Transient
+    E1_Retry --> Success: Retry succeeds (max 3)
+    E1_Retry --> E2_Alt: Max retries exceeded
+
+    Classified --> E1b_Chunk: E1b Output Overflow
+    E1b_Chunk --> Success: Chunked strategy works
+
+    Classified --> E2_Alt: E2 Recoverable
+    E2_Alt --> Success: Alternative approach works
+    E2_Alt --> E3_Block: No alternative viable
+
+    Classified --> E3_Block: E3 Blocking
+    E3_Block --> AutoRecover: Safe point + best option
+    AutoRecover --> Success: Recovery works
+    AutoRecover --> UserEscalation: Recovery fails
+
+    Classified --> E4_Cascade: E4 Cascading
+    E4_Cascade --> Rollback: Stop propagation
+    Rollback --> UserEscalation: Report impact
+
+    UserEscalation --> [*]: User decides
+    Success --> [*]: Resume execution
+```
+
+### Error Classification Table
+
+| Code | Type | Description | Recovery Action | Max Attempts |
+|------|------|-------------|----------------|-------------|
+| E1 | Transient | Timeout, network failure, temporary unavailability | Retry with exponential backoff | 3 |
+| E1b | Output Overflow | File too large for single creation tool call | Switch to chunked deliverable strategy | 1 (strategy switch) |
+| E2 | Recoverable | Logic error, incorrect approach, wrong tool usage | Log error, attempt alternative approach | Until alternative found |
+| E3 | Blocking | Critical failure, safe state needed | Save to safe point, pick best recovery option, auto-recover | 1 (then escalate) |
+| E4 | Cascading | Error affects downstream phases/agents | Stop propagation immediately, rollback affected work, report | 0 (immediate stop) |
+
+### User Escalation Protocol
+
+When auto-recovery fails (E3) or cascade is detected (E4), the user is presented with options:
+
+```
+## ⚠️ BLOCKED — Decision Required
+**Error**: {description}
+**Impact**: {affected items}
+**Options**:
+A) {Alternative} — {tradeoff}
+B) {Skip with gap} — {limitation}
+C) {Provide input} — {what's needed}
+D) {Modify requirements} — {suggestion}
+```
+
+### Core Principle
+
+Every error must lead to either:
+1. Successful completion via automated recovery, OR
+2. Explicit user decision with clear options
+
+Silent halts, unexplained terminations, and swallowed errors are forbidden.
+
+---
+
+## Golden Triangle Debate Rules
+
+The Golden Triangle is the team collaboration protocol activated by `:team` variants. It spawns exactly 3 agents per phase.
+
+### Roles
+
+| Role | Agent | Authority | Personality |
+|------|-------|-----------|-------------|
+| Tech Lead | Domain-specific (see roster) | Final authority on all decisions | Pragmatic, decisive, ships quality |
+| Executor | Domain-specific implementer | Owns implementation, must defend work with evidence | Builder mindset, evidence-driven, pushes back |
+| Reviewer | Domain-specific critic | Can FAIL submissions, demand fixes, escalate | Skeptical, thorough, assumes defects exist |
+
+### Debate Flow
+
+```mermaid
+flowchart TD
+    TL[Tech Lead decomposes task] --> EX[Executor implements]
+    EX --> RE[Reviewer critiques]
+    RE --> PASS{PASS?}
+    PASS -->|Yes| CONSENSUS[Consensus stamp]
+    PASS -->|No: FAIL| EXDEF{Executor response?}
+    EXDEF -->|FIX| EX
+    EXDEF -->|DEFEND with evidence| REEVAL[Reviewer evaluates evidence]
+    REEVAL --> ACCEPT{Evidence valid?}
+    ACCEPT -->|Yes| CONSENSUS
+    ACCEPT -->|No| ROUND{Round <= 3?}
+    ROUND -->|Yes| EX
+    ROUND -->|No| ARB[Tech Lead arbitrates - BINDING]
+    ARB --> CONSENSUS
+```
+
+### Debate Rules
+
+| Rule | Detail |
+|------|--------|
+| Maximum rounds | 3 per task — after round 3, Tech Lead makes binding decision |
+| Defense requirement | Executor must provide technical evidence (benchmarks, specs, code references) |
+| "I disagree" without proof | Automatic FAIL — Reviewer wins by default |
+| Reviewer rejects valid evidence | Escalation to Tech Lead for arbitration |
+| Tech Lead evaluation | Based on evidence quality, not seniority or role |
+
+### Communication Protocol
+
+| Artifact | Owner | Rules |
+|----------|-------|-------|
+| Shared Task List | Tech Lead | State management for assignments, status, priorities |
+| Mailbox | All agents (append-only) | `./reports/{topic}/MAILBOX-{date}.md` — all submissions, reviews, defenses, decisions |
+
+Mailbox rules:
+- Append-only — no agent may edit or delete prior exchanges
+- All agents read the full Mailbox for shared context
+- One Mailbox per phase execution, timestamped by date
+
+### Foundation Enforcement Checkpoints
+
+| Checkpoint | Scope | Rule |
+|------------|-------|------|
+| C8-TEAMS-01 | BLOCK | Mailbox is append-only and required for all inter-agent exchanges |
+| C8-TEAMS-02 | BLOCK | Debate capped at 3 rounds; unresolved disputes escalate to Tech Lead |
+| C8-TEAMS-03 | BLOCK | Phase output requires explicit consensus stamp before release |
+
+### Consensus Protocol
+
+Output is released only when one of these conditions is met:
+
+| Condition | Trigger |
+|-----------|---------|
+| Clean pass | Reviewer approved submission with no disputes |
+| Resolved pass | Reviewer approved after Executor fixed issues or defended successfully |
+| Arbitrated pass | Tech Lead overrode after max 3 rounds — reasoning documented |
+
+Consensus stamp format: `✅ CONSENSUS: TechLead ✓ | Executor ✓ | Reviewer ✓`
+
+Without this stamp, no phase output is released.
+
+---
+
+## Deliverable Integrity
+
+Governed by Law L10: files created by an agent define the standard.
+
+### Rules
+
+| Rule | Description |
+|------|-------------|
+| Deliverable IS the truth | The output file is authoritative — it defines what was decided |
+| Immutable once complete | Once a phase produces a deliverable, subsequent phases treat it as a locked constraint (Law L8) |
+| Format governs quality | Deliverable must match the agent's Output Format specification |
+| Size-appropriate strategy | <= 150 lines → single file; > 150 lines → chunked folder with index |
+
+### Deliverable Paths
+
+Research/planning agents write to `./reports/{topic}/` with prescribed naming:
+
+| Agent | Path Pattern |
+|-------|-------------|
+| brainstormer | `./reports/{topic}/brainstorms/BRAINSTORM-{feature}.md` |
+| researcher | `./reports/{topic}/researchers/RESEARCH-{feature}.md` |
+| scouter | `./reports/{topic}/scouts/SCOUT-{feature}.md` |
+| designer | `./reports/{topic}/designs/DESIGN-{feature}.md` |
+| planner | `./reports/{topic}/plans/PLAN-{feature}.md` |
+| reporter | `./reports/{topic}/general/REPORT-{type}-{date}.md` |
+| debugger | `./reports/{topic}/debugs/DEBUG-{issue}.md` |
+| tester | `./reports/{topic}/tests/TEST-{feature}.md` |
+| business-analyst | `./reports/{topic}/requirements/REQ-{feature}.md` |
+| performance-engineer | `./reports/{topic}/performance/PERF-{component}.md` |
+
+Implementation agents (backend-engineer, frontend-engineer, etc.) write directly to the project's source code.
+
+---
+
+## Language Compliance
+
+Governed by Law L6 — a dual-language protocol.
+
+| Context | Language Rule |
+|---------|-------------|
+| Response to user | Same language the user used in their request |
+| Code and code comments | Always English |
+| Files in `./reports/{topic}/` | Always English |
+| Files in `./documents/` | Always English |
+| Agent file contents | Always English |
+| Rule file contents | Always English |
+
+---
+
+## Constraint Propagation
+
+Governed by Law L9 — constraints tighten through the workflow pipeline, they never loosen.
+
+### Pipeline Direction
+
+```mermaid
+flowchart LR
+    SC[Scouter - Discovers constraints] --> PL[Planner - Adds planning constraints]
+    PL --> IM[Implementer - Executes within ALL constraints]
+    IM --> VA[Validator - Verifies constraints met]
+    
+    SC -->|"Constraints accumulate"| PL
+    PL -->|"Constraints accumulate"| IM
+    IM -->|"Constraints accumulate"| VA
+```
+
+### Rules
+
+| Rule | Description |
+|------|-------------|
+| Tighten only | Each phase can ADD constraints but never REMOVE or RELAX constraints from prior phases |
+| Scouter findings locked | If a scouter identifies technology X is used, the planner must plan around X |
+| Planner decisions locked | If a planner specifies approach Y, the implementer must execute approach Y |
+| No renegotiation | An implementer cannot decide to use approach Z instead of planner's approach Y |
+| Violation handling | If a downstream agent violates an upstream constraint, self-healing triggers backtrack |
+
+### Ambiguity Handling
+
+When a requirement is ambiguous (before it becomes a constraint):
+
+1. PAUSE execution
+2. ASK user for clarification
+3. DOCUMENT the decision
+4. THEN proceed
+
+Guessing, assuming intent, or skipping unclear items is forbidden.
+
+---
+
+## Anti-Pattern Registry (A1-A10)
+
+Defined in `rules/ERRORS.md`, these are cataloged violations that the self-healing protocol watches for.
+
+| Code | Anti-Pattern | Correct Behavior |
+|------|-------------|-----------------|
+| A1 | Direct implementation by Orchestrator | Delegate to specialist agent |
+| A2 | Workflow bypass (skipping phases) | Follow exact phase order |
+| A3 | Silent halt (unexplained stop) | Notify user + provide options |
+| A4 | Skipped step within a phase | Backtrack + complete the step |
+| A5 | Lazy fallback (using TIER 2 when TIER 1 available) | Attempt TIER 1 first |
+| A6 | Assumed requirements | Ask for clarification |
+| A7 | Modified prior deliverable | Treat prior deliverables as immutable |
+| A8 | Batched phase loading (pre-loading agents) | Load one phase at a time |
+| A9 | Meta agent implementing (tech-lead coding) | Meta agents delegate only |
+| A10 | Unexplained termination | Always explain + provide options |
+
+### Self-Healing Protocol
+
+When a violation is detected:
+
+1. DETECT violation or error
+2. PAUSE at nearest safe point
+3. LOG: `⚠️ {violation_code}: {description}`
+4. BACKTRACK to correct state if needed
+5. RE-EXECUTE correctly
+6. UPDATE downstream if affected
+7. RESUME normal flow
+
+Maximum corrections per phase: 3. If limit is reached, escalate to user.
+
+---
+
+## Evidence Sources
+
+| Source | Path |
+|--------|------|
+| Orchestration Laws (L1-L10) | `rules/CORE.md` |
+| Tiered Execution protocol | `rules/AGENTS.md` |
+| Phase Sequencing rules | `rules/PHASES.md` |
+| HSOL configuration (weights, thresholds) | `matrix-skills/_index.yaml` |
+| HSOL resolution algorithm | `rules/SKILLS.md` |
+| Trust progression and promotion | `rules/SKILLS.md` + `matrix-skills/_index.yaml` |
+| Error Classification (E1-E4) | `rules/ERRORS.md` |
+| Anti-Patterns (A1-A10) | `rules/ERRORS.md` |
+| Golden Triangle protocol | `rules/TEAMS.md` |
+| Foundation Enforcement Checkpoints | `rules/TEAMS.md` |
+| Deliverable paths | `rules/CORE.md` + `rules/REFERENCE.md` |
+| Language compliance | `rules/CORE.md` (Law L6) |
+| Constraint propagation | `rules/CORE.md` (Law L9) |
+| Ambiguity handling | `rules/CORE.md` |

package/documents/knowledge-overview/00-index.md

@@ -0,0 +1,107 @@
+# Knowledge Overview
+
+> **Purpose**: High-level summary, table of contents, and cross-references for the Agent Assistant project knowledge base
+> **Last Updated**: 2026-03-26
+> **Generated By**: docs-core skill
+
+---
+
+## Quick Summary
+
+Agent Assistant (`@namch/agent-assistant` v1.3.0) is a multi-agent orchestration framework that transforms a single AI coding assistant into a coordinated team of 21 specialist agents. It operates across five platforms — Cursor, GitHub Copilot, Claude Code, Codex, and Antigravity/Gemini — by installing Markdown and YAML configuration files into each tool's global directory. The framework uses a Hybrid Skill Orchestration Layer (HSOL) to dynamically resolve and inject skills from a matrix of 1,430+ skill definitions across 19 domains, requiring zero manual configuration per project.
+
+The core value proposition is efficiency at scale: one-time global installation provides structured workflows for every project a developer works on. The Orchestrator pattern enforces 10 governance laws, delegates to specialist agents via 14 commands (with 4 variant strategies each), and supports sub-agent spawning for parallel task execution on compatible platforms. The project claims 70% faster time-to-production, 70% bug reduction, and 85% token cost savings.
+
+Built with Node.js >=18.0.0 using only built-in modules (fs, path, os, readline), Agent Assistant has zero production dependencies. The project also includes a React 19 + Vite + Tailwind CSS v4 marketing website deployed on Vercel, and uses semantic-release with conventional commits for automated npm publishing.
+
+---
+
+## Table of Contents
+
+1. [Quick Summary](#quick-summary)
+2. [Sub-Files](#sub-files)
+3. [Quick Facts](#quick-facts)
+4. [Cross-References](#cross-references)
+5. [Known Gaps and Open Questions](#known-gaps-and-open-questions)
+
+---
+
+## Sub-Files
+
+| # | File | Description |
+|---|------|-------------|
+| 1 | [01-project-identity.md](./01-project-identity.md) | Project name, version, author, license, purpose, problem/solution statement, key benefits, and a First 60 Minutes onboarding checklist |
+| 2 | [02-tech-stack.md](./02-tech-stack.md) | Categorized technology stack table — runtime, CLI, frontend, build, styling, animation, CI/CD, release — each with version and evidence file |
+| 3 | [03-features.md](./03-features.md) | Key features list with descriptions, performance metrics, capability matrix, and agent/command inventories |
+| 4 | [04-getting-started.md](./04-getting-started.md) | Prerequisites, step-by-step installation, environment setup, first run commands, running tests, and common troubleshooting |
+
+---
+
+## Quick Facts
+
+| Key | Value |
+|-----|-------|
+| Package Name | `@namch/agent-assistant` |
+| Version | 1.3.0 |
+| Type | Multi-agent orchestration framework |
+| Language | JavaScript (CommonJS for CLI, ESM for web) |
+| Runtime | Node.js >=18.0.0 |
+| License | MIT |
+| Author | NamCH |
+| Repository | https://github.com/hainamchung/agent-assistant |
+| npm | https://www.npmjs.com/package/@namch/agent-assistant |
+| Website | https://agent-assistant-ten.vercel.app |
+| Production Dependencies | 0 (zero) |
+| Dev Dependencies | 9 (semantic-release ecosystem + husky) |
+| Total Files | ~5,784 |
+| Total Directories | ~2,372 |
+| Specialist Agents | 21 |
+| Team Configurations | 17 teams (51 team agents) |
+| Commands | 14 |
+| Command Variants | 4 (fast, hard, team) |
+| Skill Modules | 1,430+ |
+| Skill Domains | 19 |
+| Rules | 7 |
+| Supported Platforms | 5 (Cursor, GitHub Copilot, Claude Code, Codex, Antigravity/Gemini) |
+
+---
+
+## Cross-References
+
+| Resource | Path | Relationship |
+|----------|------|-------------|
+| Knowledge Source Base | [../knowledge-source-base/00-index.md](../knowledge-source-base/00-index.md) | Deep-dive into directory structure, entry points, key modules, and configuration |
+| HSOL Assessment | [../HSOL-ASSESSMENT.md](../HSOL-ASSESSMENT.md) | Detailed evaluation of the Hybrid Skill Orchestration Layer |
+| SMART Skill Orchestration | [../SMART-SKILL-ORCHESTRATION-BLUEPRINT.md](../SMART-SKILL-ORCHESTRATION-BLUEPRINT.md) | Blueprint for skill orchestration design |
+| README | [../../README.md](../../README.md) | Public-facing project overview and quick start |
+| CHANGELOG | [../../CHANGELOG.md](../../CHANGELOG.md) | Version history and release notes |
+| Core Rules | [../../rules/CORE.md](../../rules/CORE.md) | Orchestrator protocol and 10 governance laws |
+| Skills Rules | [../../rules/SKILLS.md](../../rules/SKILLS.md) | HSOL resolution algorithm and skill decision flow |
+| Package Manifest | [../../package.json](../../package.json) | npm metadata, scripts, dependencies, and engine requirements |
+
+---
+
+## Known Gaps and Open Questions
+
+1. **No automated test suite shipped** — The `cli/install.test.js.example` file is an example, not an active test. The `npm test` script (`node --test cli/*.test.js`) expects test files matching that glob but none are committed.
+2. **Web application not documented in knowledge-source-base** — The `web/` directory contains a full React 19 + Vite application but is not covered in the source base analysis.
+3. **Dynamic skill discovery undocumented** — The `_dynamic.yaml` file and `find-skills` mechanism are referenced in rules but lack dedicated architecture documentation.
+4. **No performance benchmarks** — The claimed metrics (70% faster, 70% fewer bugs, 85% token savings) lack published methodology or benchmark data.
+5. **Team variant coverage** — Not all 14 commands have a `:team` variant; the exact availability per command is defined by the presence of `commands/{cmd}/team.md`.
+6. **Platform parity details** — While all 5 platforms are listed as "Full" support, the specific features or limitations per platform are not documented.
+
+---
+
+## Evidence Sources
+
+| Source | Path |
+|--------|------|
+| Package manifest | [../../package.json](../../package.json) |
+| README | [../../README.md](../../README.md) |
+| Core rules | [../../rules/CORE.md](../../rules/CORE.md) |
+| Skills rules | [../../rules/SKILLS.md](../../rules/SKILLS.md) |
+| CLI installer | [../../cli/install.js](../../cli/install.js) |
+| Agent definitions | [../../agents/](../../agents/) |
+| Commands | [../../commands/](../../commands/) |
+| Matrix skills registry | [../../matrix-skills/](../../matrix-skills/) |
+| Knowledge source base | [../knowledge-source-base/00-index.md](../knowledge-source-base/00-index.md) |