Animus Sprint Loops

Animus Project's lightweight protocol for autonomous software development — for Claude Code, Codex CLI, and open harnesses like Animus_Ferric, OpenClaw and OpenCode.

Sprint Loops decomposes long-horizon coding work into numbered sprints, each a five-phase sequence — Research → Plan → Build → Test → Loop — with persistent state on disk and a clean separation between working memory and the rolling backlog. The filesystem IS the state machine. Trust the disk.

Three install paths

This repo ships the same protocol three ways. Pick the directory that matches your agent runtime — each is self-contained and ready to use:

The core protocol — filesystem layout, phase exit conditions, schemas — is identical across all three. What changes is only how the agent discovers and routes through the phases. Each directory is a complete, self-contained unit: installing one does not require the others, and each carries the full core protocol. open-harnesses/ additionally doubles as the canonical cross-harness reference and the Oovra-particle source.

Quick start

Claude Code (recommended — install as a plugin; this repo is a marketplace):

/plugin marketplace add crussella0129/sprint-loops
/plugin install sprint-loop@sprint-loops

Or manual (skill-only — the skill is the /sprint-loop command):

cp -r claude-code/skills/sprint-loop ~/.claude/skills/
chmod +x ~/.claude/skills/sprint-loop/scripts/*.sh

Then say "start a sprint" or run /sprint-loop start "<goal>". See claude-code/README.md.

Codex CLI:

cp -r codex-cli/skills/sprint-loops ~/.codex/skills/
chmod +x ~/.codex/skills/sprint-loops/scripts/*.sh
cat codex-cli/skills/sprint-loops/AGENTS.md.fragment >> AGENTS.md

See codex-cli/README.md.

Open harness:

cp -r open-harnesses/scripts /path/to/your/project/scripts
chmod +x /path/to/your/project/scripts/*.sh

Index open-harnesses/particles/ into your retrieval store and wire the invocation loop. See open-harnesses/README.md.

Repository layout

sprint-loops/
├── README.md            # this file
├── LICENSE              # MIT
├── open-harnesses/      # Section 1 — canonical spec: protocol, particles, schemas, scripts
├── claude-code/         # Section 2 — the sprint-loop Claude Code skill (is the /sprint-loop command) + plugin manifest
└── codex-cli/           # Section 3 — drop-in Codex CLI skill + AGENTS.md fragment

Cross-harness compatibility

A repo using Sprint Loops can be worked on by any of the three harnesses interchangeably. The filesystem state is the contract:

• sprints/sN/ and agent-tasks/ mean the same thing everywhere.
• Phase files are content-identical across harnesses (only the routing layer differs).
• decisions.md and confidence.txt are universal.
• Git history is the source of truth for what actually happened.

You can start a sprint on Claude Code at your desk, continue it on Codex in a CI run overnight, and resume it on an OpenClaw node the next morning — without any state translation. The project itself is the protocol.

Distribution

Each adapter directory can ship as a separate release targeted at its harness:

• sprint-loops-protocol — the spec, schemas, and reference shell scripts (open-harnesses/).
• sprint-loops-claude-code — drop-in ~/.claude/skills/sprint-loop/ bundle.
• sprint-loops-codex — drop-in ~/.codex/skills/sprint-loops/ bundle.
• sprint-loops-particles — Oovra-ready individual .md particles (open-harnesses/particles/).

If folded into GECK, "GECK Loops" becomes a fourth adapter directory — without changing anything in the first three.

License

MIT © 2026 Charles Russella.