Atlas / MCP servers / harmitx7 / tribunal-kit

tribunal-kitBLOCK

Name: tribunal-kit
Author: harmitx7

mcp/harmitx7/tribunal-kit

MCP server by harmitx7 · ai agents · ai coding assistant · aider · Audited 2026-10-12

A universal, native-Rust governance layer for AI coding agents. Enforces Subagent-Driven Development (SDD) to stop hallucinations before they hit disk.

Verdict

BLOCK

Grade

Trust score

22 /100

Exposed tools

64 52r · 11w · 1d

Transport

—

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Tribunal Kit

The minimal-dependency governance layer for AI coding agents. Real-time hallucination prevention, parallel AST review, and cross-session memory via a compiled Rust core.

Read from source at commit f35a08a05e9aOBSERVED · 2026-10-12

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add tribunal-kit --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env ANTHROPIC_AUTH_TOKEN=${ANTHROPIC_AUTH_TOKEN} --env AWS_SECRET=${AWS_SECRET} --env GEMINI_API_KEY=${GEMINI_API_KEY} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "tribunal-kit": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
        "ANTHROPIC_AUTH_TOKEN": "${ANTHROPIC_AUTH_TOKEN}",
        "AWS_SECRET": "${AWS_SECRET}",
        "GEMINI_API_KEY": "${GEMINI_API_KEY}"
      }
    }
  }
}

Exposed tools (64)

52 read · 11 write · 1 destructive. Blast radius: 1 tool can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`HARD_TO_REVERSE`	read	Modifies public API contracts, shared library signatures, or alters complex database tables.
`IRREVERSIBLE`	destructive	Destructive database operations (DROP TABLE, DELETE records), production secret rotation, or distributed data loss.
`Over-Engineering`	read	Using complex patterns (Kafka, K8s, event sourcing) for simple problems.
`PARTIALLY_REVERSIBLE`	read	Requires minor migration rollback or cache clearing (e.g. additive schema column, internal caching layer).
`REVERSIBLE`	read	Can be rolled back trivially with git checkout or simple code reversion (e.g. CSS, formatting, comments).
`Under-Engineering`	read	Missing critical engineering for the requirements (no transactions for payments, no auth for sensitive data).
`align_output`	read	Align model outputs to Fabel-5 constraints: strips conversational introductions and conclusions, collapses single/double bullet items to prose, and checks for code traps (unawaited dynamic functions in Next.js 15, deprecated hooks in React 19, or non-existent models).
`auth_before_business_logic`	read	All API endpoints enforce authentication and authorization before executing business logic
`auth_boundary`	read	Authentication and authorization enforcement points
`clean-code`	read	Clean code mastery. Naming, SOLID, refactoring.
`concurrent_duplicates`	read	10 identical requests simultaneously → single result, no duplicates
`data_persistence`	read	Database and persistent storage access points
`exit_plan_mode`	read	Present your finished plan for human reviewed designing your approach. Execution will suspend until the user approves or requests changes.
`external_integration`	read	Outbound connections to third-party services
`get_sparse_context`	read	Get a JIT, token-optimized context prompt tailored to the active task and files. Uses the Context Density Broker to score and select relevant skills, stripping duplicate boilerplate and saving up to 85% in prompt tokens.
`get_tribunal_agent`	read	Get the full markdown rules for a specific Tribunal agent.
`get_tribunal_skill`	read	Get the full markdown instructions for a specific Tribunal skill.
`list_tribunal_agents`	read	List all available Tribunal Kit agents.
`list_tribunal_skills`	read	List all available Tribunal Kit skills.
`module_boundary_enforcement`	read	Modules access each other only through defined interfaces, not direct imports
`no_circular_dependencies`	read	No circular dependency chains between modules/services
`no_forbidden_dependencies`	read	No module imports from forbidden layers (e.g., infrastructure importing domain)
`no_missing_timeouts`	read	All external HTTP/RPC calls have explicit timeout configuration
`no_shared_database`	read	Each service owns its database; no direct cross-service DB access
`no_synchronous_chain`	read	No synchronous call chain exceeding 3 services deep
`no_unbounded_retries`	read	All retry logic has maximum attempt limits and backoff
`public_api`	read	HTTP API endpoints exposed to clients
`query_semantic_graph`	read	Query the Ahead-of-Time (AOT) Semantic Context Graph to quickly resolve function signatures, dependencies, and file structures without inflating context windows.
`react-specialist`	read	React 19+ specialist. Components, hooks, state management.
`recall_memory`	read	Budget-constrained memory recall from the 4-Type Taxonomy Persistent Memory Engine. Returns the most relevant memories that fit within the token budget, ranked by relevance × recency × priority. Use this BEFORE writing code to recall project guidelines without bloating context.
`run_tribunal_audit`	write	Runs a full anti-hallucination audit across the workspace.
`search_case_law`	read	Search historical code rejections and legal precedent. Use this before writing code to avoid past mistakes.
`single_retry`	read	Request A sent once → state S1; Request A sent again → state S1 unchanged
`sql-pro`	read	SQL queries, PostgreSQL, MySQL, optimization
`store_memory`	read	Store a new memory entry in the 4-Type Taxonomy Persistent Memory Engine. Memories are schema-validated and persisted across sessions. Types: semantic (permanent facts), procedural (how-to recipes), episodic (30-day TTL events), working (session scratch).
`sync_ide_bridges`	write	Synchronize IDE bridge files with the current GEMINI.md rules.
`task`	write	The task or target file to execute the workflow against
`testing-patterns`	read	Testing mastery. Jest, Vitest, Playwright, mocking.
`timeout_retry`	read	Request A times out → client retries with same idempotency key → single execution
`tk_browser_audit`	read	Live web quality, accessibility, console error, and Core Web Vitals auditor. Navigates headlessly to a URL, checks WCAG 2.2 contrast/aria, network failures, uncaught exceptions, and security headers. Returns structured audit report.
`tk_browser_compare`	read	Visual regression differ. Captures synchronized screenshots of two URLs (e.g. localhost dev server vs production) and computes pixel mismatch percentage to verify layout integrity before release.
`tk_browser_navigate`	read	Token-efficient browser navigation. Navigates headlessly to a URL, prunes HTML using htmltrim (< 4,000 bytes / ~1,000 tokens), enforces IDPI prompt injection defense, and returns compact semantic markdown with an interactive elements tree.
`tk_browser_screenshot`	read	Captures a viewport screenshot of a target URL as base64 PNG data.
`tk_codify_browser_audit`	read	Audits a target URL and automatically codifies detected WCAG accessibility failures, missing security headers, and console errors into binding Case Law precedents.
`tk_deconstruct_component`	read	Reverse-engineers a live web element into a production React TSX component with Tailwind CSS classes by inspecting computed CSSOM (geometry, typography, elevation, transitions).
`tk_heal_runtime_errors`	read	Runtime Sentinel error hunter. Scans a live web page or dev server for uncaught exceptions, React boundary errors, and framework overlays, localizing them to workspace source files.
`tribunal_analyze_finding`	write	Initializes a correction run for a finding.
`tribunal_approve_write`	write	Approve a transaction requiring human approval (Simulated via token).
`tribunal_commit_write`	write	Commit an approved write transaction.
`tribunal_dispatch_dag`	read	Dispatch a Dynamic Directed Acyclic Graph (DAG) for parallel, staged execution.
`tribunal_execute_correction`	write	Executes the correction plan via ControlledWriteService.
`tribunal_generate_review_plan`	read	Generate an intelligence-driven review plan.
`tribunal_get_case`	read	Retrieve full case structured data.
`tribunal_get_context`	read	Generate or retrieve an elite Flight Data HUD context dossier for any file, multi-file pair, or directory. Automatically extracts public API contracts, callers, Chesterton
`tribunal_get_resolution_certificate`	write	Get the cryptographically signed resolution certificate for a run.
`tribunal_plan_correction`	read	Generates a correction plan and proposals.
`tribunal_preview_write`	write	Preview an approved or pending write transaction.
`tribunal_request_write`	write	Request a repository mutation. Generates a preview and token.
`tribunal_run_validation`	write	Run a specific validation command securely.
`tribunal_search_cases`	read	Search historical cases by tier, files, or categories.
`tribunal_verify_correction`	read	Verifies the executed correction.
`triple_retry`	read	Request A sent 3 times → state S1; verify single DB record, single side effect
`verify_contracts`	read	Verify proposed code changes against project behavioral contracts. Use BEFORE writing code to ensure compliance with team conventions.
`vulnerability-scanner`	read	Security vulnerability analysis. OWASP, injection.

Trust audit

BLOCKgrade F · trust 22/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	FAIL
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	FAIL
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (11 observation(s))
Network: declared (19 observation(s))
Shell: declared (13 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

CRITICALPrompt injection · prompt.transfer_instruction · CWE-94, CWE-1427

.agent/skills/huggingface-lora-space-builder/SKILL.md:398

**Step 2 — Verify the endpoint signature.** `gradio info {repo_id} --token {hf_token}` returns the exposed endpoints and their parameter types. Read the output and confirm: (a) the endpoint exists (de

Why it matters. an instruction to move sensitive data to an outside destination

Fix. remove; a skill never needs the user's secrets off the machine

HIGHInventory / provenance · inv.suspicious_name · CWE-1104

scripts/validate-payload.js

validate-payload.js

Why it matters. member named after an attack tool

Fix. remove or justify