iterative-development

iterative-development

A Claude Code plugin that drives an autonomous, audited implementation loop for projects with large, comprehensive, or ambiguous specs. Pairs with superpowers.

Fork notice: This is the System Level fork of prime-radiant-inc/iterative-development (by Jesse Vincent), repackaged with a self-contained marketplace manifest so it installs directly from this repository. Licensed under Apache-2.0.

When to use it

Use this plugin when:

• Your spec is large, comprehensive, or ambiguous (10+ files, 100+ requirements).
• You want the product in a working, testable state at every iteration boundary.
• You want an autonomous audited loop rather than a single upfront plan.
• The superpowers:writing-plans → superpowers:subagent-driven-development flow has lost the plot on this project before.

Do not use it for small, bounded projects — superpowers:writing-plans is simpler and more appropriate.

What it does

The top-level iterative-development skill orchestrates the full lifecycle:

1. Extract requirements — chunks the spec, dispatches parallel extraction subagents, and produces stories with proof obligations plus behavior scenarios. Aggregates into per-epic files, a behavior-scenarios corpus, and a coverage ledger.
2. Scope the simplest core — defines the walking skeleton iteration (ITER-0000) and an ordered roadmap of follow-on iterations. The walking skeleton must close at least one real journey scenario, not just compile.
3. Run iterations — sentinel baseline → scope review → decompose code + evidence tasks → implement → impacted + sentinel scenario runs → wrap up.
4. Audit progress — paired auditors run a three-tier review (deep evidence, impacted behavior, sentinel corpus) using parallel adversarial review (PAR). Gaps feed back into the backlog.
5. Terminate — when the roadmap is drained and a final behavior-evidence audit passes, the loop exits.

Completion means the product has passing behavior evidence at the correct seam for every externally observable requirement — not just that stories are marked done.

Skills

Skill	Role
iterative-development	Top-level orchestrator for the autonomous loop.
extracting-requirements	Chunk + parallel-extract + aggregate stories and scenarios from spec collateral.
scoping-the-simplest-core	Define walking skeleton and roadmap.
running-an-iteration	Execute one sprint end-to-end.
implementing-tasks	Per-task implementer with evidence discipline.
auditing-progress	PAR-based audits at iteration and corpus level.

Installation

This repository ships its own marketplace manifest, so it installs directly — no separate marketplace repo required.

Claude Code (desktop app or CLI):

/plugin marketplace add systemlevel/iterative-development
/plugin install iterative-development@systemlevel-plugins

Restart Claude Code after installing.

Codex (CLI or desktop app):

codex plugin marketplace add systemlevel/iterative-development
codex plugin add iterative-development@systemlevel-plugins

Maintainer note: the plugin entry in .claude-plugin/marketplace.json pins a commit sha. Claude Code installs fine without one, but Codex silently ignores remote plugin sources that are not pinned. Bump that sha whenever you publish new plugin content.

Running the tests

./scripts/run_validation_suite.sh

Individual test files can be run with pytest tests/.

Relationship to superpowers

This plugin depends conceptually on skills from the superpowers plugin (brainstorming, TDD, parallel adversarial review, verification-before-completion). Install superpowers alongside iterative-development for the intended experience.

License

Apache License 2.0. See LICENSE.