loam

4-skill academic pipeline: deep research (13-agent), academic paper (LaTeX), peer review (7-agent), pipeline orchestrator (10-stage). CC BY-NC 4.0 — non-commercial use only.

Analyzes existing workflows and processes to find inefficiencies, bottlenecks, and optimization opportunities. Goes beyond automation to improve the process itself. Use when user says "optimize my process", "find bottlenecks", "why is this slow", "improve this workflow", "process audit", or "efficiency review".

Takes a process description and generates a complete Standard Operating Procedure document with numbered steps, decision points, checklists, and troubleshooting guidance. Use when user says "write an SOP", "create SOP", "standard operating procedure", "document this process", "write a procedure for", or asks to formalize any business process.

Interactive weekly review assistant that helps business owners reflect on their week, measure progress against stated goals, and plan the next week with clarity. Use when user says "weekly review", "review my week", "what did I accomplish", "plan next week", "week in review", "how did my week go", or asks to reflect on recent work.

Maps business processes from natural language descriptions into structured workflows, identifies automation opportunities using the GREEN/YELLOW/RED framework, and generates visual workflow diagrams. Use when user says "map my workflow", "describe my process", "where should I automate", "workflow audit", or describes any business process.

GraphRAG-powered code review with blast-radius analysis via Tree-sitter AST. MCP server. Recommended for 500+ file codebases.

Autonomous multi-source research agent with MCP server. Planner/executor architecture, 5 MCP tools. Requires 3 API keys (Tavily + Anthropic + embeddings). Research-flavor.

Use when you have a free-form draft prompt and want it rewritten into Claude Opus 4.6 or 4.8 conventions before pasting it into a new planning or execution session. Trigger phrases — "align prompt", "rewrite for opus", "tune prompt for claude". NOT for planning itself, code implementation, or validation. Pre-planning / pre-execution only.

Structured decision-making tool that helps users evaluate options systematically using weighted scoring, criteria analysis, and practical recommendations. Use when user says "decision matrix", "help me decide", "compare options", "which should I choose", "evaluate options", "pros and cons", or "decision framework".

Adversarial research interrogation before launching an eval or making a published claim. Use before /overnight-eval, before adding a claim to the paper, or when a reviewer-style stress test is needed. 4 waves — basics, clarifications, deep probes, assessment. Stops you from running expensive batches on shaky premises. NOT for implementation plan review (use plan-reviewer agent) or general topic research (use /researcher).

Advisor for selecting optimal Claude model tier (Opus/Sonnet/Haiku) for a specific task. Use before launching an agent team, before a multi-file refactor, or when deciding whether to switch from Opus to Haiku for transactional work (commits, formatting). Analyzes reasoning depth, blast radius, domain expertise, output length, correctness cost.

Tool recommender — takes freeform intent and returns the best skill/agent/command to use. Use when unsure which tool to invoke next. NOT for: executing tasks directly — only recommending which skill, agent, or command to use next.

Use when invoked automatically by the UserPromptSubmit hook after a prompt is flagged as vague. Researches context, then generates 1-6 targeted clarifying questions before executing. Do NOT invoke this manually — it is triggered by the hook in .claude/hooks/improve-prompt.py.

Design polish with 27 deterministic anti-pattern rules + 12-rule LLM critique. By Paul Bakaus (jQuery UI creator). Commands: polish, audit, critique, typeset, animate, colorize, layout, delight. Apache-2.0.

Continuously improve Claude's effectiveness by recognizing patterns, saving memory, creating skills, and refining project knowledge. Use when Claude notices repeated workflows, encounters a problem it solved before, wants to save something for future sessions, needs to create a reusable skill, or when the user asks Claude to improve itself, learn, remember, or get smarter over time.

9 Nature-journal skills: figures, polishing, writing, citations, data availability, paper reading, reviewer responses, paper-to-PPT, academic search (MCP). MIT license.

Manus-style persistent markdown planning with task_plan.md, findings.md, progress.md. Hook-based re-injection after /compact.

Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce → minimise → hypothesise → instrument → fix → regression-test. Use when user says "diagnose this" / "debug this", reports a bug, says something is broken/throwing/failing, or describes a performance regression.

Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (DOMAIN.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.

Find deepening opportunities in a codebase, informed by the domain language in DOMAIN.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.

Build a throwaway prototype to flesh out a design before committing to it. Routes between two branches — a runnable terminal app for state/business-logic questions, or several radically different UI variations toggleable from one route. Use when the user wants to prototype, sanity-check a data model or state machine, mock up a UI, explore design options, or says "prototype this", "let me play with it", "try a few designs".

Test-driven development with red-green-refactor loop. Use when user wants to build features or fix bugs using TDD, mentions "red-green-refactor", wants integration tests, or asks for test-first development.

Break a plan, spec, or PRD into independently-grabbable issues on the project issue tracker using tracer-bullet vertical slices. Use when user wants to convert a plan into issues, create implementation tickets, or break down work into issues.

Turn the current conversation context into a PRD and publish it to the project issue tracker. Use when user wants to create a PRD from the current context.

Triage issues through a state machine driven by triage roles. Use when user wants to create an issue, triage issues, review incoming bugs or feature requests, prepare issues for an AFK agent, or manage issue workflow.

Tell the agent to zoom out and give broader context or a higher-level perspective. Use when you're unfamiliar with a section of code or need to understand how it fits into the bigger picture.

Stanford STORM — LLM research pipeline for Wikipedia-quality articles with citations. Multi-perspective research, Co-STORM collaborative mode. Python package, no native Claude Code integration.

Creates and launches coordinated agent teams using TeamCreate for multi-teammate tasks requiring cross-talk, shared task lists, and lifecycle management. Use when 2+ workers need to communicate findings to each other, not just report to parent. NOT for independent parallel tasks (use dispatching-parallel-agents instead).

Multi-agent deliberation council for complex tasks. Spawns 4 specialist agents with distinct thinking styles who independently analyze the problem, then 2 cross-examination agents who argue and synthesize. Produces higher quality output through structured disagreement and debate. Use when user says "council", "deliberate", "multi-agent", "debate this", "think deeply about", "I want multiple perspectives on", or when a task clearly benefits from adversarial analysis.

Use when a session's implementation work is complete and needs adversarial critique and code quality review before committing. Triggers — after multi-file changes, after feature implementation, before final commit on significant work, when independent review of own work is needed. NOT for single-file trivial edits (use /validate directly) or standard code review without fixes (use /multi-review).

Design system generator with curated color palettes, font pairings, UI styles, product-type rules, and UX guidelines across multiple tech stacks.

Multi-agent codebase analysis with interactive React knowledge graph dashboard. Best for onboarding and visual architecture exploration.

Creates and launches coordinated agent teams using TeamCreate for multi-teammate tasks requiring cross-talk, shared task lists, and lifecycle management. Use when 2+ workers need to communicate findings to each other, not just report to parent. NOT for independent parallel tasks (use dispatching-parallel-agents instead).

Use when you have a free-form draft prompt and want it rewritten into Claude Opus 4.6 or 4.8 conventions before pasting it into a new planning or execution session. Trigger phrases — "align prompt", "rewrite for opus", "tune prompt for claude". NOT for planning itself, code implementation, or validation. Pre-planning / pre-execution only.

Autonomous multi-phase release execution. Reads a PHASE-PLAN.md with stages 1-N and executes them end-to-end, running validation + critique per stage, committing with clean scope, and advancing automatically. Halts only when a validation gate fails twice or a critique surfaces an unresolvable issue. Use for multi-stage release plans, phased restructures, or any numbered-phase execution. NOT for single-stage work (use /ship), plan creation (use brainstorming), or work without a written phase plan. Use when you have a numbered-phase plan ready for execution.

Fast 30s session bootstrap briefing. Use when resuming work after any break, at the start of a fresh session, or when unsure of current project state. Reports git status, recent commits, env state, memory staleness, pending tasks, and red flags (uncommitted changes, detached HEAD, stale memory). NOT for: deep codebase exploration, implementing changes, or running tests — only a read-only status snapshot.

Create a conventional commit. Use when user says "commit", "make a commit", or "save changes". NOT for: drafting code, running tests, or anything that changes files — this skill only stages and commits already-finished work.

Create high-quality Claude Code custom skills and slash commands. Use when the user wants to create a new skill, build a custom command, make a slash command, or add a reusable workflow.

Parallel adversarial critique with 4 specialized agents plus synthesis. Use before merging a feature branch, before committing a complex plan, or when you want independent multi-angle review of a plan or implementation. Complements /multi-review (which reviews code quality) by reviewing plan/scope quality. NOT for single-file trivial changes (use /validate), standard code review (use /multi-review), or post-implementation shipping (use /ship).

Generate diagrams and visuals from code or descriptions. Use when the user needs a rendered diagram, code visualization, or architecture image. Subcommands — excalidraw (hand-drawn), drawio (formal XML), paper (academic illustrations), gitdiagram (repo architecture), yoshida (atmospheric Hiroshi Yoshida woodblock hero art).