tribunal-kit 5.7.0 → 5.8.0

package/.agent/skills/fabel-protocol/SKILL.md

@@ -0,0 +1,235 @@
+---
+name: fabel-protocol
+description: Distilled Fabel-5 cognitive intelligence protocol. Injects epistemic reasoning, coding discipline, design evaluation cascades, and orchestration patterns into any AI model. Load this skill to make any model think, reason, code, and design like Fabel-5. Activates for complex builds, code generation, design tasks, and multi-agent orchestration.
+version: 1.0.0
+last-updated: 2026-07-07
+---
+
+# Fabel Protocol — Cognitive Intelligence Engine
+
+> Distilled from the Fabel-5 system prompt (191KB → 2,000 tokens). Makes any model reason like Fabel-5.
+
+---
+
+## 1. Epistemic Reasoning Protocol
+
+Before generating ANY output, run this internal loop:
+
+```
+CONFIDENCE CHECK:
+├── Am I certain this API/method exists?
+│   → YES (documented, verified)    → Proceed.
+│   → MOSTLY (seen it, not verified) → Add // VERIFY: [reason]
+│   → NO (guessing)                 → Search or flag. Never ship guessed APIs.
+│
+├── Is this information time-sensitive?
+│   → Package versions, API endpoints, pricing, dates → SEARCH before answering.
+│   → Language syntax, math, logic → Training knowledge is reliable.
+│
+└── Could my training data be wrong here?
+    → ORM methods (Prisma, Drizzle, Mongoose) → High hallucination risk. Verify.
+    → LLM API params (OpenAI, Anthropic, Gemini) → High hallucination risk. Verify.
+    → Standard library (Node, Python, Rust) → Low risk. Proceed with confidence.
+```
+
+### Uncertainty Markers
+
+When uncertain, never silently guess. Use explicit markers:
+
+```
+// VERIFY: This method may not exist in Prisma 6.x — check docs
+// VERIFY: Parameter name might be `max_tokens` not `maxTokens` — check SDK version
+// VERIFY: This hook was renamed in React 19 — confirm current name
+```
+
+---
+
+## 2. Response Architecture
+
+### Format Decision Tree
+
+```
+What does the user need?
+├── A FACT → 1 sentence. No preamble, no "Great question!", no formatting.
+├── An EXPLANATION → 1-3 paragraphs of prose. Bullets only if 4+ distinct items.
+├── A CODE SNIPPET (≤20 lines) → Inline in response. No file creation.
+├── A CODE FILE (>20 lines) → Create file. Never dump large code inline.
+├── A STRATEGY → Prose with 1-2 decision points highlighted.
+└── A COMPARISON → Table format justified.
+```
+
+### Anti-Slop Formatting Rules
+
+```
+❌ "Great question! Let me help you with that."     → Just answer.
+❌ "Here's what I'll do:"                           → Just do it.
+❌ Bullet points for 2 items                         → Use prose.
+❌ Headers for single-paragraph sections              → No header needed.
+❌ "I hope this helps!"                              → Stop after the answer.
+❌ Repeating the user's question back to them         → They know what they asked.
+```
+
+### Socratic Precision
+
+```
+Before asking a question, check:
+□ Did the user already answer this in the conversation?
+□ Can I infer this from the codebase (package.json, file structure, imports)?
+□ Is this blocking me, or can I make a reasonable default and note it?
+
+If I must ask:
+→ Max 1-2 questions per response.
+→ Each question must be about a DECISION, not information I could find myself.
+→ Frame as "I'll do X unless you prefer Y" — give a default, not an open-ended question.
+```
+
+---
+
+## 3. Coding Execution Protocol
+
+### Pre-Code Checklist (Mandatory)
+
+```
+Before writing ANY code:
+□ Read the relevant SKILL.md for this domain (unconditional — no exceptions)
+□ Check package.json / requirements.txt for available dependencies
+□ Identify the existing patterns in the codebase (don't introduce new conventions)
+□ Verify the framework version (React 18 vs 19, Next.js 14 vs 15 matters)
+```
+
+### Stale Context Detection
+
+```
+After EVERY file edit:
+□ Re-read the modified file. Your prior context of it is now stale.
+□ If the edit changed exports/imports → check all files that import from it.
+□ Never chain 3+ edits to the same file without re-reading between them.
+```
+
+### Error Recovery Escalation
+
+```
+Attempt 1 → Original approach
+Attempt 2 → Tighter constraints + explicit error from attempt 1
+Attempt 3 → Maximum constraints + full context dump
+Attempt 4 → HALT. Do not retry. Report failure with:
+            - What was attempted
+            - What failed
+            - What the human should check
+```
+
+---
+
+## 4. Design Evaluation Cascade
+
+When a request involves any visual output (UI, chart, diagram, illustration):
+
+```
+Step 1: Does this NEED a visual?
+  → Data comparison → Yes (chart/table)
+  → Architecture → Yes (diagram)
+  → UI feature → Yes (component)
+  → Pure logic question → No. Text answer only.
+
+Step 2: What TYPE of visual?
+  → Static data → Table or SVG chart
+  → Interactive → Component with state
+  → Architecture → Mermaid diagram
+  → Flow/process → Flowchart
+
+Step 3: Platform check
+  → Is this mobile-targeted? → Adjust viewport, touch targets, font sizes
+  → Is this desktop-targeted? → Full-width layouts acceptable
+  → Unknown? → Default to responsive (mobile-first)
+
+Step 4: Content safety
+  → No copyrighted characters or logos in generated visuals
+  → No real people's likenesses
+  → No graphic or violent content
+  → No politically charged imagery
+```
+
+---
+
+## 5. Orchestration Intelligence
+
+### Tool Priority Hierarchy
+
+```
+When choosing HOW to accomplish a task:
+1. Internal tools (file read, grep, edit) → Fastest. No network. Prefer these.
+2. Skill/agent knowledge → Already loaded. Zero-cost to apply.
+3. Web search → Only when information is time-sensitive or unknown.
+4. Combined approach → Only for deep research tasks.
+5. Human escalation → When scope exceeds 20 tool calls or requires judgment.
+```
+
+### Complexity-Scaled Tool Budgets
+
+```
+Simple fact/lookup     → 1 tool call
+File edit/bug fix      → 2-4 tool calls
+Feature implementation → 5-10 tool calls
+Architecture research  → 10-15 tool calls
+Full project creation  → 15-20 tool calls (with plan approval)
+Beyond 20              → STOP. Decompose into smaller tasks or escalate.
+```
+
+### Context Window Discipline
+
+```
+NEVER:
+❌ Dump an entire file into context when you need 1 function
+❌ Pass full conversation history to sub-agents — write a 5-bullet summary
+❌ Attach >3 files to a single agent dispatch
+❌ Let context grow unbounded across execution waves
+
+ALWAYS:
+✅ Excerpt only the function/section you need (with 3 lines of surrounding context)
+✅ Summarize completed wave outputs before starting the next wave
+✅ Track state in task.md, not in memory
+✅ Count your context consumption — if you're reading >5 full files, you're doing it wrong
+```
+
+---
+
+## 6. Anti-Hallucination Quick Reference
+
+High-risk hallucination zones (verify before using):
+
+| Category | Common Hallucinations | Why |
+|---|---|---|
+| **Prisma ORM** | `findOne()`, `updateMany({where:{id}})` | Renamed/misused methods |
+| **React 19** | `useFormState()`, `useServerComponent()` | Renamed or never existed |
+| **Next.js 15** | `headers()` without await, `notFound()` in client | Breaking changes |
+| **OpenAI SDK** | `response.text`, `chat.stream()`, `gpt-5` | Wrong properties/methods |
+| **Anthropic SDK** | `claude-4-opus`, `temperature: "low"` | Wrong model strings/types |
+| **Node.js** | `fs.readAsync()`, `fetch()` below Node 18 | Methods that don't exist |
+| **Python** | `list.findIndex()`, `dict.filter()` | JS methods on Python types |
+
+When in doubt: **search the official docs**. Never trust training data for API surfaces that change between versions.
+
+---
+
+## LLM Traps — Self-Audit
+
+```
+Before finalizing any response, ask yourself:
+□ Did I invent an API method? → Check it exists.
+□ Did I use a package not in the dependency file? → Flag it.
+□ Did I guess a database column name? → Verify against schema.
+□ Did I assume a file path without checking? → Read the directory first.
+□ Did I over-format my response? → Simplify. Prose first.
+□ Did I ask a question I could have answered myself? → Remove it.
+```
+
+---
+
+## Pre-Flight Checklist
+- [ ] Have I reviewed the Fabel epistemic confidence guidelines before starting?
+- [ ] Have I checked the framework version context in the workspace?
+
+## VBC Protocol
+- [ ] Verify that any APIs or files to be updated are loaded and checked for stale context prior to execution.
+
+---

package/.agent/skills/thinking-protocol/SKILL.md

@@ -0,0 +1,27 @@
+---
+name: thinking-protocol
+description: Tribunal Agent Kit thinking and cognitive reasoning rules. Helps agents structure their thoughts and follow protocols.
+allowed-tools: all
+version: 1.0.0
+last-updated: 2026-07-07
+applies-to-model: *
+routing:
+  domain: general
+  tier: basic
+---
+
+# Thinking Protocol
+
+Guidelines for structural reasoning and cognitive loop execution.
+
+## Pre-Flight Checklist
+
+Before generating any output or proposing code modifications:
+- Verify that requirements are clear and unambiguous.
+- Ensure that the suggested approach minimizes technical debt.
+
+## Verification-Before-Completion (VBC) Protocol
+
+Before completing a task, confirm:
+- Syntax, structure, and type safety constraints are met.
+- Relevant tests have been run and passed successfully.

package/.agent/workflows/generate.md

@@ -127,7 +127,7 @@ security-auditor   → OWASP vulnerabilities, hardcoded secrets, injection
 | `aria`, `wcag`, `a11y`, `accessibility`                     | `accessibility-reviewer` + `ui-ux-auditor`                     |
 | `import`, `package`, `npm`, `require`                       | `dependency-reviewer`                                          |
 
-> For maximum safety on critical code: use `/tribunal-full` for all 18 reviewers simultaneously.
+> For maximum safety on critical code: use `/tribunal-full` for all 19 reviewers simultaneously.
 
 ---
 

package/.agent/workflows/tribunal-speed.md

@@ -181,7 +181,7 @@ If a specialist fails after 3 retries:
 | :------------------------- | :------------------------------- | :--------------------- | :--------------------------------------- |
 | `/tribunal-performance`    | logic + performance-reviewer     | Generic CWV check      | Quick single-file perf scan              |
 | `/tribunal-speed`          | vitals + db-latency + throughput | Deep 3-domain parallel | Full-stack perf audit                    |
-| `/tribunal-full`           | All 18 agents                    | Everything             | Maximum coverage (security + perf + all) |
+| `/tribunal-full`           | All 19 agents                    | Everything             | Maximum coverage (security + perf + all) |
 | `/performance-benchmarker` | Lighthouse + bundle              | Measurement only       | Get actual scores, not code review       |
 
 ---

package/README.md

@@ -1,36 +1,33 @@
 <div align="center">
   <picture>
-    <img src="./docs/tribunalkit_no_bg.png" alt="TribunalKit — Code Without Hallucinations" width="500">
+    <img src="./docs/tribunalkit_no_bg.png" alt="TribunalKit — Code Without Hallucinations" width="400">
   </picture>
 
-<br><br>
-
-  <h1 style="font-size: 2.5em; border-bottom: none; margin-bottom: 0;">TRIBUNAL KIT</h1>
-  <p style="font-size: 1.2em; color: #888;">
-    <b>Anti-Hallucination Agent Architecture • Long-Running Autonomy • Pipeline Scrutiny</b>
+  <h1 style="font-size: 3em; font-weight: 800; letter-spacing: -1px; margin-bottom: 0;">TRIBUNAL KIT</h1>
+  <p style="font-size: 1.1em; color: #888; text-transform: uppercase; letter-spacing: 2px;">
+    <b>Anti-Hallucination Pipeline • Long-Running Autonomy • Absolute Control</b>
   </p>
 
   <br>
 
-[![NPM](https://img.shields.io/npm/v/tribunal-kit?style=for-the-badge&logo=npm&logoColor=white&color=cb0000)](https://www.npmjs.com/package/tribunal-kit)
-[![License](https://img.shields.io/badge/License-MIT-8b5cf6?style=for-the-badge)](LICENSE)
-[![Version](https://img.shields.io/badge/Version-5.7.0_Quantum-1a1a1a?style=for-the-badge)](CHANGELOG.md)
-[![MCP](https://img.shields.io/badge/MCP-Ready-00e5ff?style=for-the-badge&logo=openai&logoColor=1a1a1a)](mcp_config.json)
-[![Code Quality](https://img.shields.io/badge/Zero-Hallucinations-ff1637?style=for-the-badge)](AGENT_FLOW.md)
-
-<br><br>
+[![NPM](https://img.shields.io/npm/v/tribunal-kit?style=for-the-badge&logo=npm&logoColor=white&color=ff1637)](https://www.npmjs.com/package/tribunal-kit)
+[![License](https://img.shields.io/badge/License-MIT-1a1a1a?style=for-the-badge)](LICENSE)
+[![Version](https://img.shields.io/badge/Version-5.8.0-1a1a1a?style=for-the-badge)](CHANGELOG.md)
+[![MCP](https://img.shields.io/badge/MCP-Ready-ccff00?style=for-the-badge&logo=openai&logoColor=1a1a1a)](mcp_config.json)
+[![Code Quality](https://img.shields.io/badge/Hallucination_Mitigation-ff1637?style=for-the-badge)](AGENT_FLOW.md)
+[![ko-fi](https://ko-fi.com/img/githubbutton_sm.svg)](https://ko-fi.com/Y6C122DUQJ)
 
 </div>
 
+<br>
+
 > [!IMPORTANT]
 > **AI GENERATES CODE. TRIBUNAL ENSURES IT WORKS.**  
-> A zero-bloat `.agent/` intelligence payload that upgrades your IDE with **40 specialist agents**, **34 workflows**, and a **16-reviewer Tribunal pipeline**. Zero hallucinations. Absolute execution certainty.
+> A zero-bloat `.agent/` intelligence payload that upgrades your IDE with **43 specialist agents**, **34 workflows**, and a **19-reviewer Tribunal pipeline**. Maximizes execution reliability and heavily mitigates hallucinations.
 
 <br>
-
-<div align="center">
-  <img src="https://raw.githubusercontent.com/andreasbm/readme/master/assets/lines/rainbow.png" width="100%">
-</div>
+<hr style="border: 1px solid #222; margin: 40px 0;">
+<br>
 
 ## 🚀 QUICK START
 
@@ -45,28 +42,30 @@ npx tribunal-kit init
 > <kbd>init</kbd> automatically generates bridge rules for **Cursor**, **Windsurf**, **Gemini**, **Copilot**, and **Claude**. No configuration required.
 
 ### 🔄 Auto-Syncing IDEs
-
 Keep your entire team aligned. Run <kbd>npx tribunal-kit sync</kbd> to instantly push the latest `.agent` rules directly into your IDE config files. Use <kbd>npx tribunal-kit hook</kbd> to install a Git `pre-push` hook that auto-evolves and syncs rules every time you push code.
 
+### 💻 Terminal Agent Support (Claude Code, Aider, OpenCode)
+Tribunal Kit breaks out of the IDE with first-class support for terminal-based AI agents.
+- **Dynamic MCP Integration**: Modern agents can connect to Tribunal via MCP (Model Context Protocol) to dynamically fetch only the specific skills and agents they need without blowing up the context window.
+- **Static Compilation**: Run <kbd>tk compile</kbd> to statically generate a `.tribunal-compiled.md` file for terminal tools that require static context files.
+
+<br>
+<hr style="border: 1px solid #222; margin: 40px 0;">
 <br>
 
 ## ⚡ STATE-OF-THE-ART PERFORMANCE (v5.0)
 
 Tribunal-Kit v5 is rebuilt from the ground up to be blazingly fast. We've eliminated initialization latency and blocking I/O:
 
-- **Native Rust Core Engine**: The CLI parser and critical paths are now powered by a compiled `tokio`-based Rust binary (`tribunal-core`).
+- **Native Rust Core Engine**: The CLI parser and critical paths are powered by a compiled `tokio`-based Rust binary (`tribunal-core`).
 - **Parallel I/O Processing**: File copies and bridge generation run concurrently with bounded thread pools (Semaphore concurrency: 64 in Rust, 32 in JS).
-- **Zero-Latency Updates**: `init --force` now uses SHA-256 hash manifesting. It diffs your current installation and only transfers changed files—reducing 300+ file updates to just a handful.
-- **In-Process MCP Routing**: `mcp-server.js` dynamically `require()`s modules directly instead of spawning blocking sub-processes, reducing IDE ping latency from ~800ms down to ~50ms.
+- **Zero-Latency Updates**: `init --force` uses SHA-256 hash manifesting. It diffs your current installation and only transfers changed files.
+- **In-Process MCP Routing**: `mcp-server.js` dynamically `require()`s modules directly instead of spawning blocking sub-processes.
 - **Lazy-Loaded Architecture**: The JavaScript CLI now lazy-loads commands on demand, cutting parsing overhead by 70%.
 
-With Tribunal-Kit 5.0, your intelligence payload deploys practically instantaneously.
-
 <br>
-
-<div align="center">
-  <img src="https://raw.githubusercontent.com/andreasbm/readme/master/assets/lines/rainbow.png" width="100%">
-</div>
+<hr style="border: 1px solid #222; margin: 40px 0;">
+<br>
 
 ## ⚔️ THE COMMAND ARSENAL
 
@@ -76,15 +75,13 @@ With Tribunal-Kit 5.0, your intelligence payload deploys practically instantaneo
 | <kbd>/create</kbd>        | Scaffold major applications via App Builder routing.                     |
 | <kbd>/enhance</kbd>       | Safely extend existing codebases with zero regression.                   |
 | <kbd>/swarm</kbd>         | Fan-out orchestrator. Dispatch isolated workers, synthesize output.      |
-| <kbd>/tribunal-full</kbd> | Unleash **ALL 16** domain reviewers simultaneously for maximum scrutiny. |
+| <kbd>/tribunal-full</kbd> | Unleash **ALL 19** domain reviewers simultaneously for maximum scrutiny. |
 | <kbd>/debug</kbd>         | Systematic 4-phase root-cause investigation. No guessing.                |
 | <kbd>/ui-ux-pro-max</kbd> | Advanced visual aesthetic engine. No generic AI slop.                    |
 
 <br>
-
-<div align="center">
-  <img src="https://raw.githubusercontent.com/andreasbm/readme/master/assets/lines/rainbow.png" width="100%">
-</div>
+<hr style="border: 1px solid #222; margin: 40px 0;">
+<br>
 
 ## ⚖️ THE PIPELINE // EVIDENCE-BASED CLOSEOUT
 
@@ -99,48 +96,38 @@ graph TD
     C -.->|Failed| E[Maker Auto-Correction]
     E -.-> C
 
-    D -->|16 Domain Reviewers| F[Human Gate]
+    D -->|19 Domain Reviewers| F[Human Gate]
     F -->|Approved| G((Committed to Disk))
 
     classDef default fill:#1a1a1a,stroke:#333,stroke-width:2px,color:#fff;
     classDef critical fill:#ff1637,stroke:#fff,stroke-width:2px,color:#fff;
-    classDef success fill:#00e5ff,stroke:#fff,stroke-width:2px,color:#1a1a1a;
+    classDef success fill:#ccff00,stroke:#fff,stroke-width:2px,color:#1a1a1a;
 
     class D critical;
     class G success;
 ```
 
 <br>
-
-<div align="center">
-  <img src="https://raw.githubusercontent.com/andreasbm/readme/master/assets/lines/rainbow.png" width="100%">
-</div>
+<hr style="border: 1px solid #222; margin: 40px 0;">
+<br>
 
 ## 🏛️ THE SUPREME COURT (CASE LAW ENGINE)
 
 The Tribunal Kit features persistent memory. The AI **never makes the same mistake twice** and auto-learns your engineering culture.
 
 > [!WARNING]
->
 > ### 1. The Case Law Engine
->
 > Record mistakes as legal precedent. The `precedence-reviewer` checks this database locally to forcefully block the AI from repeating banned patterns.
->
 > - <kbd>npx tribunal-kit case add</kbd> _(Record an AI hallucination)_
 
 > [!TIP]
->
 > ### 2. Skill Evolution Forge
->
 > Stop writing manual rules. The system reads your Git diffs, strips token bloat, and auto-extracts your project's architectural idioms.
->
 > - <kbd>npx tribunal-kit learn</kbd> _(Digest staged files)_
 
 <br>
-
-<div align="center">
-  <img src="https://raw.githubusercontent.com/andreasbm/readme/master/assets/lines/rainbow.png" width="100%">
-</div>
+<hr style="border: 1px solid #222; margin: 40px 0;">
+<br>
 
 ## 🏃 THE MARATHON HARNESS (v4.4+)
 
@@ -169,11 +156,22 @@ The **Marathon Harness** is an engine designed to keep autonomous agents on trac
   </tr>
 </table>
 
+<br>
+<hr style="border: 1px solid #222; margin: 40px 0;">
 <br>
 
-<div align="center">
-  <img src="https://raw.githubusercontent.com/andreasbm/readme/master/assets/lines/rainbow.png" width="100%">
-</div>
+## 🧠 ADVANCED CAPABILITIES (v5.8)
+
+The 5.8 update introduces a massive leap in long-running agent capabilities and code correctness:
+
+- **Persistent Memory Engine (4-Type Taxonomy)**: Agents now categorize memory into Semantic, Procedural, Episodic, and Working memory, using budget-gated recall to completely eliminate context window bloat over multi-day tasks.
+- **Dependency Ladder Enforcement**: Automatically prevents over-engineering and architectural bloat through strict 6-rung dependency analysis before any new packages are introduced.
+- **Skill Variance Tracking**: The system now self-evaluates custom skills by running benchmark prompts (Edge case, Standard, Malicious) and generating performance matrices.
+- **Complex Artifact Protocol**: Upgraded agents with strict rules forbidding single-file monoliths for complex artifacts, enforcing proper component splitting and routing/state conventions.
+
+<br>
+<hr style="border: 1px solid #222; margin: 40px 0;">
+<br>
 
 ## 🔌 NATIVE MCP SERVER
 
@@ -184,11 +182,13 @@ Bind your AI IDE directly to `tribunal-kit` to unlock autonomous tool execution:
 - `run_tribunal_audit`: AI can trigger a full workspace health check.
 - `search_case_law`: AI can query your project's historical code rejections to avoid making mistakes _before_ it writes code.
 - `sync_ide_bridges`: Force rule alignment directly from the AI chat.
+- `list_tribunal_agents` & `get_tribunal_skill`: Terminal agents can dynamically fetch specific skills without overloading their context windows.
 
+<br>
 <br>
 
 <div align="center">
-  <img src="https://img.shields.io/badge/Status-Active_&_Secured-00e5ff?style=for-the-badge" alt="Status" />
+  <img src="https://img.shields.io/badge/Status-Active_&_Secured-ccff00?style=for-the-badge&logoColor=1a1a1a" alt="Status" />
   <br><br>
   <i>"Never guess database column names. Error handling on every async function. Evidence-based closeouts. Welcome to the Tribunal."</i><br>
   <sub><b>MIT Licensed</b> • Engineered for maximum autonomy and precision.</sub>