builder

builder makes your coding agent answer to you. Every piece of work ends with a short report: is it done, what you have to do, what isn't proven yet, what left your machine, and every call the agent made on its own, with the reason. Nothing leaves the machine (no push, no PR, no message) without a send rule or your yes.

It's a plugin for Claude Code, Codex CLI, Cursor CLI and opencode. Skills and one session hook; no server, no account, no network call at session start.

▶ Needs you: a customer who moved can now change their email at checkout, in one step.

▶ You (1): push feat/change-email and open the PR. I ask below.

⚠ Not proven: the confirmation email in a real inbox.

✓ Proven: go test ./... → 14 passed (example).

↗ Outside: nothing sent.

Calls: the old address gets a notice, because a stolen account changes the email first.

The marks mean one thing each: ✓ done or proven, ✗ failed, ▶ you act, ⚠ not proven, ↗ left this machine. After the report the agent asks each ▶ You item as a choice, its recommendation first.

Why builder

superpowers and gstack make the agent better at a job: planning, debugging, a browser, QA. builder's report is the same whichever skill ran: every finished piece of work ends with what's done, what's yours to do, and what isn't proven. builder also covers ground neither pack does: messages to people, rules that remember their incident, many sessions at once, and your usage limit.

	builder	superpowers	gstack	your own CLAUDE.md
Starts from the person and the non-goals	every piece of work	in brainstorming	in /spec, /office-hours	if you wrote it
Each solo call recorded with its reason	every report, under Calls	while executing a plan	durable decisions, in a decision log	if you wrote it
Unproven work marked as unproven	every report	when verification runs	in QA and review	if you wrote it
Messages to people: send rules, duplicate check, 6 an hour	reach	no	no	no
A rule carries its incident and a check that fails	canon	no	no	no
Many sessions, one lane each, hand-offs as briefs	fleet	no	no	no
Weekly usage-limit forecast	budget	no	no	no

builder doesn't replace either pack. It routes to them: planning, TDD and debugging go to superpowers; the browser, QA and shipping go to gstack. Install all three.

Install

builder runs on macOS and Linux. It needs Bash, and it offers to install one small helper binary in your first session.

Claude Code

/plugin marketplace add Lutherwaves/builder
/plugin install builder@builder

Codex CLI. Trust the plugin's hook when Codex asks.

codex plugin marketplace add Lutherwaves/builder
codex plugin add builder@builder

Cursor CLI and opencode: see docs/install.md.

In your first session the agent does your work first, then offers one setup step per reply: superpowers, gstack, a signed-in gh, signed commits. Answer yes, not now (quiet for 7 days) or never. Answers live in ~/.config/builder/ledger.json; set BUILDER_HOME to move them. After setup, a session shows at most one recommendation, and only when something calls for it, such as your weekly usage passing 80%.

The first session also asks to install the helper binary. It runs the setup steps, the send checks and the forecasts. The installer downloads the binary of your plugin's version from builder's GitHub release, checks the checksum and, with a signed-in gh, the build provenance. It installs to ~/.config/builder/bin. To install it yourself, run bash <plugin root>/scripts/install.sh.

Harness support

A harness counts as supported once every eval fixture passes on it.

Harness	Version	Eval date	Status
Claude Code	2.1.308 (Claude Code)	2026-10-22	experimental: canon 5/6; fleet 3/4; harness 4/6; reach 3/4; shape 2/6; 5 not graded

Skills

Skill	What you get
builder	The thirteen tenets, routing to superpowers and gstack, and the report.
shape	A feature name becomes an issue: the person, the moment, anti-goals, open questions with a recommendation.
canon	A repo rule with its incident and a check that fails when someone breaks it.
reach	Messages to teammates wait for your yes, with a duplicate check and 6 sends an hour.
fleet	Many sessions side by side, one lane each; work outside a lane goes to its owner.
intake	Today's captured tasks reconciled against work in motion; nothing filed without your yes.
budget	A forecast of your weekly usage limit, and which idle session to compact.
harness	A local code graph that answers "what calls this?" in one query.
review	A review that proves each harm with evidence, a fix and a test. Posts nothing without your yes.
security	A full security stance: every layer, standards-mapped findings with CVSS scores, and what changed since last time.

Ask for a skill by name, or just describe the job. The agent loads the right one.

Works with

builder runs on its own and says once when a partner is missing.

• superpowers: planning, TDD, debugging, review.
• gstack: a real browser, QA, security and ship workflows.
• gh: intake reads open PRs and issues through it.
• matchblox: budget reads your plan windows and fleet your live sessions from it (docs).
• Your team channel's MCP server: reach sends through it. For Slack, the claude.ai Slack connector or Slack's own MCP server.
• codebase-memory-mcp: harness sets it up; it runs locally with no API key.

Contributing

Every change lands through a pull request that passes bash scripts/validate.sh. See CONTRIBUTING.md, and SECURITY.md to report a vulnerability.

License

MIT. The harness skill takes its framing from Ryan Lopopolo's Harness Engineering (CC BY 4.0); builder adapts the ideas and copies no prose.