fhhs-skills
An all-in-one workflow toolkit for Claude Code and pi.dev — 33 skills and 31 specialized agents covering planning, spec-driven development, building, reviewing, debugging, design quality, security, web research, startup validation, and autonomous execution. One install, no other plugins required.
What You Get
Plan before you build. /fh:plan-work researches your problem using FPF-lite (hypothesis diversity with epistemic tagging), brainstorms conservative and ambitious approaches, generates a SPEC.md for complex plans (architecture, data flow, failure modes via the spec-architect agent), locks decisions, and outputs an execution-ready plan. /fh:plan-review stress-tests it against business alignment, engineering rigor, and blast radius before a single line of code is written.
Build with quality gates. /fh:build turns plans into code using parallel subagents — each task runs in fresh context with TDD, LSP navigation, SPEC.md awareness, and framework-specific best practices. Task state persists for crash resume. Phase completion triggers integration checks, goal verification, and architecture refresh. Complex builds get a post-build reflection pass — an adversarial critic that catches biases, grades against spec, and surfaces friction before review.
Review that actually catches things. /fh:review runs static analysis, spec verification, goal verification against must-haves, and conditional quality refinement — dispatching design agents (polish, harden, normalize, security) based on what the diff actually touches. Recurring patterns across sessions are surfaced and escalated.
Research anything. /fh:research routes queries through content-type-specific patterns — documentation sites, GitHub repos, news articles, research papers, YouTube transcripts — using Firecrawl for clean markdown extraction. Dispatches subagents for multi-source synthesis.
Fix bugs systematically. /fh:fix triages by complexity, writes a failing test first, then patches. Complex bugs get parallel debugger subagents or persistent debug sessions.
Run autonomously. /fh:auto chains plan -> review -> build -> review for every phase without intervention. Decisions are logged with confidence levels. Crashes resume from state. Cross-phase reflection findings feed into subsequent planning to avoid repeating mistakes.
Cross-session memory. Skills persist findings between sessions — root causes, architectural decisions, vulnerability patterns. claude-mem is the primary context access layer: skills query by intent before reading files, and large outputs are indexed instead of flooding the context window.
Validate startup ideas. Five dedicated skills cover market research, competitive analysis, positioning, pitch scripts, and strategic advising — all feeding into the build pipeline when you're ready to code.
Install
Claude Code
claude plugin marketplace add cinjoff/fhhs-skills
claude plugin install fh@fhhs-skills
Then run /fh:setup in a Claude Code session. It detects your platform and walks you through dependencies.
pi.dev
pi install /path/to/fhhs-skills
# or: pi install git:https://github.com/cinjoff/fhhs-skills
The repository now ships pi-compatible adapters under .pi/skills/ (declared in package.json via the pi.skills manifest). In pi, use /skill:<name> commands, for example:
/skill:fh-setup
/skill:fh-plan-work
/skill:fh-build
If you use a pi subagent extension/tooling, this repo also generates project-local subagent adapters in .pi/agents/*.md with fh:* names (for example fh:code-reviewer, fh:gsd-debugger).
Full command mapping: .pi/README.md.
Subagent mapping: .pi/agents/README.md.
OpenAI Codex
Codex supports Agent Skills format (SKILL.md). This repository ships Codex adapters under .codex/skills/.
Use the adapters in-repo (or keep .codex/ and .claude/ together in your own distribution). The adapters read canonical skill instructions from .claude/skills/.
In Codex, invoke by skill name (for example fh-setup, fh-plan-work, fh-build).
Full command mapping: .codex/README.md.
Maintainers: adapter sync
.claude/skills/ is the single source of truth for skills, and agents/*.md is the source of truth for subagent personas. .pi/skills/, .codex/skills/, and .pi/agents/ are generated adapters.
npm run sync:adapters
npm run check:adapters
Quick Start
New Project
/fh:new-project scaffold a project with vision, stack, design language, and roadmap
/fh:plan-work design the first feature
/fh:build execute the plan
/fh:review verify and promote
(Claude command surface shown above. pi.dev and Codex equivalents are listed in .pi/README.md and .codex/README.md.)
