Agent Workflow Amplifiers (AWAs)
A Claude Code plugin of high-leverage skills that improve orchestration, integration, and one-shot task completion with minimal prompt overhead.
What This Is
This repo contains compact skills for agentic workflows.
These skills are not meant to add net-new intelligence. They are meant to extract more value from capabilities the agent already has — planning, execution, sub-agents, validation, and tool use.
The goal is simple: small prompts, disproportionate workflow improvement.
Why This Exists
AI coding agents are good out of the box. They read files, write code, run tests. But left to their own defaults, they work linearly — one hypothesis at a time, one file at a time, one pass. The failure mode isn't stupidity. It's underutilization. The agent has sub-agent dispatch, parallel execution, isolated worktrees, web search, and browser automation available. It almost never uses them together without being told to.
That's the gap this plugin fills. Each skill here is a small, structured prompt that changes the shape of how the agent works — not what it knows, but how it moves. Pre-flight recon before editing. Parallel research across web and codebase simultaneously. One-prompt API integrations that chain sub-agents through a full build pipeline. Friction detection that turns recurring pain into permanent solutions.
The design constraint is intentional: high leverage per token. A skill that takes 200 tokens of prompt and saves 30 minutes of manual orchestration is worth keeping. A skill that reminds the agent to write tests is not. The qualify agent enforces this bar on every addition.
Core Idea
A good skill is not just a reminder. It should do one of three things:
- act as a force multiplier by unlocking more of the agent's existing capabilities
- improve one-shot completion by increasing the odds a large task is fully completed in one pass
- do both
This repo is focused on high leverage per token.
Included Skills & Agents
Every skill is a single SKILL.md. The description field controls when Claude auto-invokes it — see each file for the full body.
Skills
| Skill | What it does |
|---|
agentify | Bootstraps a dispatchable sub-agent skill for every coding agent CLI installed on the machine (codex, aider, cursor, etc.). |
appmap | Maps a web UI and generates a Claude skill to automate it via parallel orchestration. |
automate | Sets up a cron/launchd job to run Claude headlessly and send results to Telegram. |
contract | Reference convention for sub-agent I/O schemas. Loaded by orchestrator skills via /agent-workflow-amplifiers:contract and into agents via skills: field. |
forge-friction | Surfaces recurring friction patterns detected across sessions and identifies actionable skill opportunities. |
ground-state | Before any non-trivial implementation, dispatches a parallel recon wave (git + infrastructure + memory) that produces a 5-line ground-truth snapshot to prevent wrong-branch edits, assumed-no-CI, and missed memory context. |
integrate | One-prompt API integration pipeline. Any API (docs URL, name, or description) → working CLI wrapper + skill + tests. |
provideme | One-prompt provider bridge. Any coding agent CLI → local Anthropic-compatible API server (POST /v1/messages) with a {provider}-claude launcher. |
research | Dispatches two sub-agents in parallel — one to search the web, one to inspect the local codebase — and returns a merged research brief. |
resolve | Resolves PR code review feedback by dispatching one parallel sub-agent per issue. |
ship | Release pipeline for already-done work: /ground-state pre-flight → tests → commit (user-approved) → push → PR with structured verification summary. --verify triggers an adversarial verifier wave on the diff. |
spec | Transforms a loose idea into a structured, actionable spec ready for implementation. |
web | Orchestrates parallel browser automation across independent Chrome tabs, one sub-agent per tab. |
Agents
| Agent | What it does |
|---|
qualify | Gate-keeps proposed skills. Approves only real force multipliers; rejects reminders, checklists, and best-practice nudges. |
Installation
From the marketplace
Inside a Claude Code session:
/plugin marketplace add griffinwork40/agent-framework
/plugin install agent-workflow-amplifiers@griffinwork40-agent-framework
/reload-plugins
Local installation
