MCP App StoreMCP App Store
from citadel592
Route user intents to the cheapest capable handler (skill, marshal, archon, or fleet) using a multi-tier classifier for scope, complexity, persistence, and para
What it does: Provides a single-entry routing skill that classifies user requests and dispatches them to the most appropriate execution path: direct command, specific skill, marshal orchestrator, archon (campaigns), or fleet (parallel agents). Implements multi-tier matching (pattern, active-state, keyword, LLM classifier) with proportionality and trust gates.
When to use it: Use /do when the user asks the agent to perform work but doesn't know which tool is appropriate. Do not use when you already know the destination (invoke /marshal, /archon, /fleet, or a specific skill).
What's included:
Compatible agents: Suited to Claude-style agent harnesses and orchestration environments (Claude Code, Citadel-like systems).
Citadel's /do is a unified intent router that classifies user requests across 4 tiers (pattern match, active state, keyword, LLM classifier) and dispatches to the cheapest capable handler. Well-documented routing table and clear escalation paths. No bundled scripts to test. The SKILL.md is very long and monolithic — could benefit from splitting routing tables into references/. No security issues found; telemetry is fire-and-forget with graceful skip on failure.
Impressive routing design with 4-tier classification. The skill is prompt-only (no scripts), so static analysis focuses on the SKILL.md content. No credentials, no destructive commands, no phone-home beyond a local telemetry script. The main concern is complexity — the SKILL.md is dense and could overwhelm agents. Architecture could be improved by extracting the routing table into a separate reference file. Usefulness is moderate — it's niche to Citadel users and depends on the full harness ecosystem being present.
Marshal (Session Commander)
A single-session meta-orchestrator that parses a direction, chains existing skills, and executes multi-step tasks end-to-end without unnecessary user prompts.
Citadel Setup
Interactive first-run wizard that installs hooks, detects project stack, generates harness config, and runs a short live demo to register skills and protect the
/research — Focused Investigation
Structured research protocol that converts questions into focused findings with cited sources, confidence levels, and a written research document for follow-up.
Create App — End-to-End Application Generator
Generate a complete application or add features to an existing codebase from a single natural-language description, with tiered workflows, PRD/architecture chec
PRD (Product Requirements Document) Generator
Generates structured Product Requirements Documents from natural-language app descriptions, asking clarifying questions, researching references, and producing m
Fleet — Parallel Campaign Orchestrator
Orchestrate parallel campaigns in waves: spawn isolated agents/worktrees, coordinate discoveries, validate results, and merge work — for multi-stream engineerin
Learn — Knowledge Compiler
Compile findings from completed campaigns and evolve cycles into a structured project wiki, extracting patterns, decisions, and quality rules for reuse.
Development Worktree
Create an isolated git worktree for feature work, auto-run project setup, and verify a clean test baseline before development.
ds-fix — data-science mid-analysis fixer
Orchestrates diagnosis and targeted fixes mid-analysis: diagnose root cause, apply fixes with output-first verification, and update project learnings.
Readwise Reader Document Management
Manage Readwise Reader documents: list, save, search, move, tag, highlight, export and bulk-edit via official and custom CLIs.
Bounty Hunter — Atlas
Persona skill: 'Atlas' — a profit-focused developer persona for discovering, evaluating and executing paid bounties or freelance tasks with ROI-aware workflows.
Junshi — Research Advisor
Daily strategic research advisor that scans arXiv/venues, digests papers, and proposes bold, ranked research ideas tailored to the user's profile.
Scikit-learn (Classical ML reference)
Comprehensive scikit-learn guidance for classification, regression, clustering, preprocessing, model evaluation, and production-ready ML pipelines.
Full Stack Builder
End-to-end builder that scaffolds, implements, tests, and optionally deploys web and API applications from a natural-language specification.
ezBookkeeping API Tools
Command-line API tools for ezBookkeeping: record and query transactions, retrieve accounts/categories/tags, and fetch exchange rates for self-hosted personal fi