tribunal-kit 9.2.4 → 9.2.6

package/.agent/skills/tribunal-instincts-memory/SKILL.md

@@ -1,31 +1,86 @@
 ---
 name: tribunal-instincts-memory
-description: "Use when extracting lessons from session logs, resolving repeated failures, creating persistent context, and validating AI memory constraints."
-version: 1.0.0
-last-updated: 2026-09-25
+description: "Use when capturing, distilling, querying, and applying durable architectural lessons and failure post-mortems across sessions. Prevents repeated mistakes and resolves conflicting learned behaviors."
+version: 6.0.0
+last-updated: 2026-09-29
 skills:
   - fabel-protocol
+  - clean-code
 tools: Read, Grep, Glob, Bash, Edit, Write
 scripts-binding:
-  - .agent/scripts/memory_archivist.js
+  - .agent/scripts/memory_engine.js
+  - .agent/scripts/verify_all.js
+  - .agent/scripts/lint_runner.js
 ---
 
-# Tribunal Instincts (Continuous Memory)
+# Tribunal Instincts & Durable Learning Memory
 
-## Overview
-Tribunal Instincts allows the agent harness to learn from its past mistakes and successful resolutions without blowing up the context window. It extracts concrete lessons from the Event Log and injects them selectively into future sessions.
+## Activation Boundaries
+- **Activate when:** Use when capturing, distilling, querying, and applying durable architectural lessons and failure post-mortems across sessions. Prevents repeated mistakes and resolves conflicting learned behaviors.
+- **DO NOT activate when:** The task falls outside the `tribunal-instincts-memory` domain or is managed by a different dedicated specialist agent.
 
-## Memory Lifecycle
-1. **Candidate Extraction**: The `memory-archivist` scans the durable JSONL session log for anomalies, repeated errors, or successful recoveries.
-2. **Evidence Validation**: A memory CANNOT be created without a specific Event ID or stack trace proving it happened. Subjective complaints are dropped.
-3. **Storage**: Approved lessons are saved to `instincts.json` under specific scopes (e.g., `database`, `ui`).
-4. **Context Injection**: During workflow initialization, `context_broker.js` reads only the instincts relevant to the current task.
 
-## Rules of Instinct Retrieval
-- **Token Budget**: Never inject more than 5 instincts at a time.
-- **Conflict Resolution**: If a newer instinct contradicts an older one, the older one is archived.
-- **System Rule Supremacy**: Instincts can NEVER override hardcoded system rules (e.g., `GEMINI.md`). They serve only as technical implementation hints.
+## 🔁 Multi-Pass Execution Protocol
 
-## Hallucination Traps
-- ❌ Trusting a memory that says "Always use library X" without checking `package.json`.
-- ❌ Creating an instinct for a typo. (Only systemic or architectural lessons deserve permanent memory).
+| Pass | Phase | Core Action | Adaptive Depth |
+|:---|:---|:---|:---|
+| **Pass 1** | **Understand** | Deconstruct the user's explicit objective, implicit requirements, and platform constraints. | Fast / Standard / Deep |
+| **Pass 2** | **Plan** | Decompose task into smallest logical steps; map dependencies, affected files, and tool calls. | Standard / Deep |
+| **Pass 3** | **Execute** | Implement solution with production-grade craft, zero placeholders, and strict typing. | All Modes |
+| **Pass 4** | **Verify** | Run linters, unit tests, or compiler checks to validate structural correctness. | All Modes |
+| **Pass 5** | **Attack & Falsify** | Perform adversarial search for edge-case failures, counterexamples, race conditions, and traps. | Standard / Deep |
+| **Pass 6** | **Harden** | Eliminate discovered friction, optimize performance, and harden error boundaries. | Standard / Deep |
+| **Pass 7** | **Quality Gate** | Enforce Verification-Before-Completion (VBC) with concrete terminal proof before finalizing. | All Modes |
+
+
+## Mandatory Pre-Flight Context Inspection
+1. **Trigger Identification**: Check if current error or task matches a known failure pattern in durable memory.
+2. **Instinct Extraction Criteria**: An instinct is created ONLY when a non-trivial failure is diagnosed and verified with a fix.
+3. **Conflict Resolution**: If two learned instincts conflict, prioritize active codebase code over historical instincts.
+4. **Generalization Boundary**: Ensure learned rules are scoped to the relevant framework/domain, avoiding over-generalization.
+5. **Decay & Pruning**: Remove obsolete instincts when underlying framework versions or libraries are updated.
+
+## 🛠️ Technical Architecture & Reference Recipes
+
+### 1. Instinct Schema Definition
+```json
+{
+  "id": "instinct-042-fastapi-blocking-io",
+  "domain": "backend",
+  "trigger": "async def with blocking IO",
+  "root_cause": "Calling time.sleep() or sync DB inside async def freezes event loop",
+  "remediation": "Use def for sync libraries or await asyncio.sleep()",
+  "evidence_test": "test/integration/event_loop.test.py",
+  "confidence": 0.95
+}
+```
+
+## 🏛️ Tribunal Verification & Guardrails
+
+### 🔬 Evidence Standard (Tri-State Verification)
+Every finding, audit statement, or completion claim must classify its factual certainty:
+- **`[OBSERVED]`**: Directly confirmed in the codebase or verified via executed terminal command.
+- **`[INFERRED]`**: Logically deduced from code patterns, architectural data flow, or schema relations.
+- **`[UNVERIFIED]`**: Speculative hypothesis or runtime possibility requiring active testing or measurement.
+
+- **Active Reviewers:** `memory-archivist`, `orchestrator`
+- **Evidence Standard:** Every stored instinct must link to concrete reproduction code and verified test proof.
+
+## 🚨 Edge-Case & Failure Mode Matrix
+
+| Scenario | Risk | Production Mitigation |
+|:---|:---|:---|
+| **Empty or Null Inputs** | Unhandled exception or unexpected rendering collapse | Enforce fallback guards, optional chaining, and explicit empty state handlers |
+| **Network Timeout / Latency** | Hanging operations or duplicate side-effects | Implement bounded abort controllers, exponential backoff, and idempotency keys |
+| **Concurrency / Race Conditions** | Stale state overwrite or inconsistent data mutations | Use atomic transactions, mutex locking, or cancel-on-resubmit controls |
+| **Invalid Schema / Malformed Payload** | Downstream runtime errors or security injection | Validate boundary payloads with Zod/Pydantic schemas prior to execution |
+| **Resource / Memory Saturation** | OOM errors, frame drops, or memory leaks | Clean up listeners, cancel active timers, and enforce pagination/virtualization |
+
+
+## 🤖 LLM-Specific Traps Table
+
+| Anti-Pattern | What AI Commonly Does Wrong | What Is Actually Correct |
+|:---|:---|:---|
+| **Hallucinated Tool Capabilities** | Assuming an external library or CLI command exists without verification | Run a verification check or verify package.json before referencing tools |
+| **Premature Completion Claim** | Declaring a task finished because code was generated without verification | Execute tests, linters, or terminal commands to provide concrete proof |
+| **Context Bloat Dumping** | Pasting entire multi-thousand-line files into prompt context | Extract targeted excerpts, symbols, and signatures to preserve tokens |

package/.agent/skills/trl-training/SKILL.md

@@ -1,18 +1,43 @@
----
-name: trl-training
-description: Train and fine-tune transformer language models using TRL (Transformers Reinforcement Learning). Supports SFT, DPO, GRPO, KTO, RLOO and Reward Model training via CLI commands.
-license: Apache-2.0
-metadata:
-  version: "1.0.0"
-  author: huggingface
-  commands: trl sft, trl dpo, trl grpo, trl kto, trl rloo, trl reward
-  categories: machine-learning, llm-training, reinforcement-learning
-  tags: rlhf, supervised-fine-tuning, dpo, grpo, huggingface, transformers
-  documentation: https://huggingface.co/docs/trl/en/clis
----
-
-# TRL Training Skill
-
+---
+name: trl-training
+description: "Use when loading, evaluating, fine-tuning, and deploying trl training machine learning models, datasets, and inference pipelines."
+version: 6.0.0
+last-updated: 2026-09-29
+tools: Read, Grep, Glob, Bash, Edit, Write
+scripts-binding:
+  - .agent/scripts/lint_runner.js
+  - .agent/scripts/verify_all.js
+---
+
+# TRL Training Skill
+
+## Mandatory Pre-Flight Context Inspection
+Before reading, generating, or refactoring code in the `trl-training` domain, inspect these 5 critical parameters:
+1. **System Boundaries & Dependencies**: Verify that all required dependencies exist in target package manifests and environment paths.
+2. **Runtime Context & Platform Invariants**: Confirm target platform constraints (Node.js, Browser, Mobile OS, Edge runtime) before applying APIs.
+3. **Execution Guardrails**: Identify potential side-effects, state mutations, and unhandled asynchronous exceptions.
+4. **Validation & Type Contracts**: Validate input data schemas and strict type constraints across all module interfaces.
+5. **Observability & Proof of Execution**: Ensure execution produces tangible verification signals (terminal output, tests, metrics).
+
+
+## Activation Boundaries
+- **Activate when:** Use when loading, evaluating, fine-tuning, and deploying trl training machine learning models, datasets, and inference pipelines.
+- **DO NOT activate when:** The task falls outside the `trl-training` domain or is managed by a different dedicated specialist agent.
+
+
+## 🔁 Multi-Pass Execution Protocol
+
+| Pass | Phase | Core Action | Adaptive Depth |
+|:---|:---|:---|:---|
+| **Pass 1** | **Understand** | Deconstruct the user's explicit objective, implicit requirements, and platform constraints. | Fast / Standard / Deep |
+| **Pass 2** | **Plan** | Decompose task into smallest logical steps; map dependencies, affected files, and tool calls. | Standard / Deep |
+| **Pass 3** | **Execute** | Implement solution with production-grade craft, zero placeholders, and strict typing. | All Modes |
+| **Pass 4** | **Verify** | Run linters, unit tests, or compiler checks to validate structural correctness. | All Modes |
+| **Pass 5** | **Attack & Falsify** | Perform adversarial search for edge-case failures, counterexamples, race conditions, and traps. | Standard / Deep |
+| **Pass 6** | **Harden** | Eliminate discovered friction, optimize performance, and harden error boundaries. | Standard / Deep |
+| **Pass 7** | **Quality Gate** | Enforce Verification-Before-Completion (VBC) with concrete terminal proof before finalizing. | All Modes |
+
+
 You are an expert at using the TRL (Transformers Reinforcement Learning) library to train and fine-tune large language models.
 
 ## Overview
@@ -315,4 +340,49 @@ When helping users with TRL:
 - Verify dataset format matches the expected schema
 - Recommend starting with smaller models for testing
 - Suggest LoRA for resource-constrained environments
-- Point to specific documentation sections for advanced features
+- Point to specific documentation sections for advanced features
+
+## 🚨 Edge-Case & Failure Mode Matrix
+
+| Scenario | Risk | Production Mitigation |
+|:---|:---|:---|
+| **Empty or Null Inputs** | Unhandled exception or unexpected rendering collapse | Enforce fallback guards, optional chaining, and explicit empty state handlers |
+| **Network Timeout / Latency** | Hanging operations or duplicate side-effects | Implement bounded abort controllers, exponential backoff, and idempotency keys |
+| **Concurrency / Race Conditions** | Stale state overwrite or inconsistent data mutations | Use atomic transactions, mutex locking, or cancel-on-resubmit controls |
+| **Invalid Schema / Malformed Payload** | Downstream runtime errors or security injection | Validate boundary payloads with Zod/Pydantic schemas prior to execution |
+| **Resource / Memory Saturation** | OOM errors, frame drops, or memory leaks | Clean up listeners, cancel active timers, and enforce pagination/virtualization |
+
+
+## 🤖 LLM-Specific Traps Table
+
+| Anti-Pattern | What AI Commonly Does Wrong | What Is Actually Correct |
+|:---|:---|:---|
+| **Platform-Dependent Shell Commands** | Hardcoding $(cat ~/.cache/huggingface/token) or piping to jq | Use platform-agnostic token checks via env var HF_TOKEN and native JSON parsing |
+| **VRAM OOM Crash** | Loading 70B model weights on a consumer 16GB GPU without quantization | Calculate parameter bounds (fp16 = VRAM/2, Q4 = VRAM*2) and load with bitsandbytes/GGUF |
+| **Blocking Batch Generation** | Generating large batches synchronously without streaming or yield | Use TextIteratorStreamer or asynchronous generate workers to prevent request timeouts |
+
+
+## 🏛️ Tribunal Verification & Guardrails
+
+**Active Reviewers:** `ai-code-reviewer` · `python-pro` · `performance-optimizer`
+**Slash Command:** `/review` or `/tribunal-full`
+
+### 🔬 Evidence Standard (Tri-State Verification)
+Every finding, audit statement, or completion claim must classify its factual certainty:
+- **`[OBSERVED]`**: Directly confirmed in the codebase or verified via executed terminal command.
+- **`[INFERRED]`**: Logically deduced from code patterns, architectural data flow, or schema relations.
+- **`[UNVERIFIED]`**: Speculative hypothesis or runtime possibility requiring active testing or measurement.
+
+### ✅ Pre-Flight Self-Audit Checklist
+```
+✅ Are model architectures and weights verified to fit within target hardware VRAM budgets?
+✅ Are pipeline requests guarded with bounded timeouts and retry backoffs?
+✅ Are dataset loading scripts operating in streaming mode to prevent out-of-memory errors?
+✅ Are API tokens and cache paths handled portably without assuming UNIX shell environments?
+✅ Are tokenizers and generation parameters (max_new_tokens, temperature) strictly bounded?
+```
+
+### 🛑 Verification-Before-Completion (VBC) Protocol
+**CRITICAL:** You must follow a strict "evidence-based closeout" state machine.
+- ❌ **Forbidden:** Declaring a task complete because the output "looks correct."
+- ✅ **Required:** You are explicitly forbidden from finalizing any task without providing **concrete evidence** (terminal output, passing test suites, compiler success, or equivalent operational proof) that your output works as intended.