Quick Start
That's it. decapod init creates (and backs up existing) CLAUDE.md, AGENTS.md, and GEMINI.md with methodology your agents follow.
What agents get:
- Persistent memory across sessions
- Standard interface (no guessing)
- Proof requirements before claiming "done"
What you get:
- Proofs validate correctness automatically
- Intent tracking shows what changed and why
- Confidence to merge agent work
The Problem
AI lowers the barrier to writing code—but shipping code is still hard.
Every agent session starts from scratch. Context evaporates. You can't trust it to:
- Remember what it built yesterday
- Follow your standards without drift
- Prove the code works before claiming "done"
- Ship without you checking every line
Shipping is a system. Decapod is that system for agentic development.
What It Does
You wouldn't run microservices without coordination. You wouldn't run a database without ACID. But we're handing agents write access to production repos and hoping "good prompting" substitutes for discipline.
Decapod turns agent output into an engineering pipeline:
- Shared state that survives sessions — work doesn't reset on handoff
- One agent-first interface (CLI + schemas) — agents don't poke internals
- One authority chain — Intent → Spec → Code → Proof → Merge
- Proof gates — "sounds right" can't ship without evidence
- Full traceability — what changed, who changed it, why
Before / After
Before Decapod:
You: "Add auth"
AI: *writes code*
You: "Add logout"
AI: "What auth?" 🤦
With Decapod:
You: "Add auth"
AI: *code → intent logged → tests pass*
You: "Add logout"
AI: *extends auth system → proves it works*
You: "Ship it" ✅
The One Big Idea: Proofs
A proof is any runnable check that can falsify a claim.
Tests? Proof. Schema validation? Proof. Linter passing? Proof.
If you can't name the proof, you can't claim it's verified.
That's the whole trick: make correctness cheap to verify and expensive to fake.
Repo Layout
.decapod/
├── data/ # state that survives sessions
└── constitution/ # the operating contract (authority + workflow + proof doctrine)
Local-first by design:
- No daemon
- No hosted service
- No "agent memory SaaS"
- Just a repo-native control plane you version, review, and enforce
Who This Is For
✅ You're building real products with AI agents ✅ You want CI/CD discipline, not "vibes-based" shipping ✅ You need multiple agents working without chaos ✅ You merge to production (not just prototyping)
If you want a one-off script, Decapod is overkill. If you want agents to ship production code, Decapod is the missing layer.
Get Involved
- Ship a subsystem — adapters, proof harnesses, connectors
- Harden the constitution — if you found a rule that stops drift, PR it
- Break it — open issues with repros (they become proof gates)