templates

Ringer Template Library

A kit is a reusable Ringer starter: a manifest skeleton, check skeletons, and a short README that capture one proven or planned task shape. At manifest-writing time, make one of three sanctioned moves: choose one kit and fill in the placeholders; mix and match by lifting a round, task shape, or check from one kit into another; or roll your own when nothing fits. Task specs and checks are designed to be lifted whole, so prefer moving complete pieces over rewriting them from memory. If you roll your own, skim the two nearest kits first so your manifest still inherits the prior art.

Catalog

Kit	What it does	Reach for it when	Status
review-swarm	Runs N read-only scouts, one surface each, with structured findings.	You need broad review coverage before deciding what to fix.	Proven in a recorded run
fix-swarm	Runs N isolated workers in worktrees to apply independent fixes, then exports patches.	You have confirmed fixes that can be split by file or surface.	Proven in a recorded run
focus-group	Runs isolated persona workers against a product, pitch, prompt, or workflow and collects parseable verdicts.	You need product or messaging feedback without persona bleed.	Proven in a recorded run
bakeoff	Runs scenarios by candidate models through the per-task model field.	You need evidence for choosing a model, prompt, or configuration.	Proven in a recorded run
research-with-proof	Runs research tasks plus a proof task whose check executes the claim.	You need the final answer to be verifiable, not just plausible.	Proven in a recorded run
launch-kit	Runs three go-to-market rounds: research/candidates/build, persona panel, then final assembly.	You need a launch package from rough direction to assembled output.	Proven in a recorded run, 2026-07-06
asset-swarm	Runs media production lanes with render-as-check animations, idempotent image batches, diagrams, and captures.	You need many visual or media assets produced with executable validation.	Proven in a recorded run, 2026-07-06
adversarial-review	Sends the same artifact to N different models, collects structured findings, then synthesizes.	You want model diversity on one artifact before deciding what is real.	Proven in a recorded run
repo-feature	Lets sandboxed workers edit a real repo through writable_roots, then runs a build check and git-porcelain allowlist.	You know what to build and need a delivery lane into an actual repo.	Proven in a recorded run, 2026-07-06
migration-swarm	Splits mechanical codebase transforms across worktrees and exports patches.	You have repetitive edits that can be partitioned safely.	Blueprint
doc-swarm	Assigns documentation by module with executed examples and no-invented-API checks.	You need docs that prove the APIs and commands they describe.	Blueprint
test-hardening	Adds tests by module with count-increase and assertion-density checks while keeping src/ off-limits.	You need stronger tests without letting workers change production code.	Blueprint
competitive-teardown	Runs N scouts by competitor with verbatim-citation allowlist checks and a synthesis phase.	You need grounded competitor analysis with traceable source material.	Blueprint
data-pipeline	Splits fetch, transform, and validate stages with executed validators and honesty rules.	You need a data workflow that proves each stage produced what it claims.	Blueprint
probe	Provides a one-task manifest for smokes, probes, and post-mortems.	You need a visible, logged check before trusting a new engine, model, harness, or diagnosis.	Proven as a practice

Kit Anatomy

Standard kit files are manifest.json, README.md, and usually one or more executable helpers under checks/. Placeholders use bare identifiers like {{NAME}} for required values and freeform prompts like {{THING — inline guidance}} when the placeholder itself explains the choice to make. After fill-in, every manifest must pass ./ringer.py lint before a run. Kit checks are skeletons to adapt to the real task; never turn them into exit 0, true, or any other check that cannot fail.

Composition Guide

• review-swarm findings feed fix-swarm: scouts identify the confirmed work, then fix workers own disjoint patches.
• launch-kit round 2 is focus-group specialized: use the same isolated persona pattern anywhere a draft needs reaction.
• asset-swarm lanes drop into any kit that needs media: images, animations, diagrams, and captures can be validated as task outputs.
• probe is the pre-flight for any new engine or model: prove the harness path once before giving it a real batch.
• repo-feature is the delivery lane after research kits decide what to build: keep discovery separate from repo mutation.

Craft Floor

Every spec opens with the worker's role and boundary, names what it owns, and states what it must not touch. Checks print why they fail, not just that they failed. Declare deliverables in expect_files so the artifact page has the real outputs. When worktrees are doomed after success, deliverables and patch exports land outside those worktrees before the check passes.