Generate structured project specifications, user stories, database schemas, and phased implementation plans from developer input and code analysis, with automated context documentation and deployment safety hooks.
Generate or refresh the canonical AGENTS context tree (AGENTS.md, CLAUDE.md, docs/agents/*.md) from the implemented code. Idempotent upsert β safe to re-run; updates only what drifted. Documents reality only; never reads .spec.
Show the status of the init spec chain (.spec/init/*) β present, absent, or stale β and run the next init:* command in the chain. Writes nothing itself; all authoring lives in the invoked init:* commands.
Read the project description and user stories, then derive a suggested database schema in DBML at .spec/init/database-schema.md
Interview the developer, discover the stack, and produce a structured project description at .spec/init/project-description.md
Read the project description, user stories, database schema, and any design specs, then plan the build into numbered, agent-ready phases with tasks, acceptance criteria, and feature tests at .spec/init/project-phases.md
Writer agent for the /ai-context pipeline. Generates or refreshes AGENTS.md and CLAUDE.md in the target repo from the inspector digest. Idempotent β diffs against on-disk bytes, writes only when content changed, respects the ownership contract. Use only as step 4 of /ai-context.
Writer agent for the /ai-context pipeline. Generates or refreshes the 8 docs/agents/*.md files describing the implemented code as it is (AS IS only β no prescriptive layers). Idempotent β diffs against on-disk bytes, writes only when content changed, respects the ownership contract. Use only as step 4 of /ai-context.
Read-only repo inspector for the /ai-context pipeline. Sweeps the implemented code (never .spec) and returns a structured digest β stack, commands, layout, signals, artifact ownership β that the writer agents consume verbatim. Use only as step 3 of /ai-context.
Adversarial requirements QA for the /plan pipeline (step 6). Two modes β analyze (find ambiguities, gaps, contradictions in SPEC.md and return prioritized questions) and resolve (apply developer answers in-place, increment version). Never talks to the developer directly; the router owns the conversation. Use only as step 6 of /plan.
Read-only planning lead for the /plan pipeline (step 7). Produces PLAN.md β architecture-aware task decomposition with dependency phases, risks, validation criteria β plus PHASES.md (ralph.sh-executable view) and conditionally formal contracts (OpenAPI/gRPC/AsyncAPI) from SPEC RIGID. Use only as step 7 of /plan.
Executes bash commands
Hook triggers when Bash tool is used
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
bc-harness)π§π· DocumentaΓ§Γ£o em portuguΓͺs
A Claude Code plugin with commands, agents, and scripts that take a project from idea to implementation in a structured way: formal specification, phased planning, and autonomous execution with mechanical validation β while keeping a human in control at every decision point.
The harness is stack-agnostic: language, framework, commands, and conventions are defined by the project's own documents (AGENTS.md, CLAUDE.md, the .spec/ chain), never by the harness.
IDEA CODE
β β²
βΌ β
/init:project-description βββ β
/init:user-stories β init chain β
/init:database-schema β (.spec/init/) β
/init:project-phases βββ β
β β
β /plan "<feature description>" β
β (.spec/features/<slug>/) β
βΌ β
project-phases.md or PHASES.md βββββββββΊ scripts/ralph.sh
(autonomous execution
with 4 gates)
/ai-context ββΊ AGENTS.md + docs/agents/* (documents the ALREADY implemented
code; feeds /plan and ralph)
Three independent pipelines that fit together:
/init β from zero to a project build plan (description β user stories β schema β phases)./plan β from a feature description to a formal SPEC + phased plan, ready for execution.ralph.sh β executes any phase document autonomously, one fresh agent session per phase, with mechanical gates and one commit per completed phase.Cross-cutting: /ai-context keeps the context tree (AGENTS.md, CLAUDE.md, docs/agents/*.md) in sync with the real code.
This repository is a Claude Code plugin (.claude-plugin/plugin.json). Install it via marketplace/local path according to your plugin setup:
/plugin install bc-harness
Commands are namespaced: /bc-harness:init, /bc-harness:plan, etc. (abbreviated without the namespace throughout this document).
ralph.sh is a standalone bash script β copy or reference scripts/ralph.sh and run it directly in the target project's repository.
ralph.sh prerequisites:
npm install -g @openai/codex + OPENAI_API_KEYnpm install -g @anthropic-ai/claude-code + ANTHROPIC_API_KEY/init β init chain routerShows the state of the .spec/init/ artifacts (present / absent / stale) and invokes the next command in the chain (one hop per run β re-run /init to advance). Writes nothing itself; all authoring lives in the invoked init:* command.
The chain, in order:
| # | Artifact | Command | Inputs |
|---|---|---|---|
| 1 | .spec/init/project-description.md | /init:project-description | β (head of chain) |
| 2 | .spec/init/user-stories.md | /init:user-stories | project-description |
| 3 | .spec/init/database-schema.md | /init:database-schema | description + stories |
| 4 | .spec/init/project-phases.md | /init:project-phases | description + stories + schema |
| β | .spec/init/design/ | manual (optional) | β |
Every generated artifact carries a stamp of its inputs on line 3 (file@sha256:<12 chars>). If an input changes later, /init detects it and reports the downstream artifact as stale β re-running the corresponding command is upsert-safe: it interviews only about the deltas and refreshes the stamp.
/init:project-description β interviews the developer, discovers the stack, and produces a structured project description./init:user-stories β derives structured, testable user stories from the description./init:database-schema β derives a suggested database schema in DBML./init:project-phases β plans the build into numbered, agent-ready phases with tasks, acceptance criteria, and feature tests. This is ralph.sh's default input. Reads .spec/init/design/ when present (screen/component refs)./plan β feature planning pipeline/plan "<feature description or path to a description file>"
Produces, under .spec/features/<slug>/:
npx claudepluginhub beerandcodeteam/beer-and-code-harnessFull-cycle project development - brainstorm ideas, create specifications, plan architecture, initialize projects, and execute implementation with integrated workflows from superpowers and spec-kit
Lightweight AI-assisted build harness: scaffold + tasks + acceptance + living roadmap + decision log + mandatory plan/spec gate + verify gate + deploy/rollback. Six-phase loop (roadmap β plan β build β test β reflect β deploy) with ANSI-colored phase banners, .harness/conventions.md for code-quality rules, and explicit memory confirmation after every action.
Context-driven development framework that enables structured lifecycle for software projects: Context β Spec & Plan β Implement
Production-ready Claude Code configuration with role-based workflows (PMβLeadβDesignerβDevβQA), safety hooks, 44 commands, 19 skills, 8 agents, 43 rules, 30 hook scripts across 19 events, auto-learning pipeline, hook profiles, and multi-language coding standards
Specification-driven development workflow: specify β plan β tasks β implement
AI-powered cascading development framework with design document system and multi-agent collaboration. Breaks down projects into Features (Mega Plan), Features into Stories (Hybrid Ralph), with auto-generated technical design docs, dependency-driven batch execution, Git Worktree isolation, and support for multiple AI agents (Codex, Amp, Aider, etc.).