π¦
cargo binstall decapod && decapod init
Decapod
Repo-native governance for AI coding agents.
Intent Β· Custody Β· Trajectory Β· Proof β fleet coherence
Work in Cursor, Claude Code, Codex, Antigravity, Grok, or any harness you prefer. Decapod is the shared governance layer inside the repository β so agent work stays bounded, attributable, and recoverable.
cargo binstall decapod
decapod initOne command installs the kernel. One command prepares the repository. You keep speaking naturally. Agents call Decapod at the points that matter β before acting, before inference, and before claiming done.
What was asked, what was understood, what changed, and what was proven is written into the project itself.
Independently launched agents need one project authority: distinct claims, a selected governance record, and a shared proof boundary for completion.
| Intent | Outcome, constraints, and completion standard β carried in plans, todos, and specs. |
| Custody | Exclusive claims and isolated workspaces so agents do not silently collide. |
| Trajectory | Selected events so a later process can reorient without replaying a vendor session. |
| Proof | Validation bound to governed state. Completion is a verified transition. |
The hard case is concurrent work that is similar but distinct: shared modules, different outcomes. Worktrees stop file trampling. Claims, context, trajectory, and publication gates keep the fleet coherent.
You speak to whichever agent is available. The agent calls Decapod. Humans keep natural language. Agents hold a stable machine contract.
Decapod sits between the harness and the model as a governance kernel β not another agent, and not a model router.
flowchart LR
subgraph HumanGroup["Human"]
UserIn["User"]
UserOut["User"]
end
subgraph HarnessGroup["Agent Harness"]
HarnessNode["Harness"]
end
subgraph IntelligenceGroup["Intelligence"]
ModelNode["Model"]
end
subgraph GovernanceGroup["Governance"]
DecapodNode["Decapod"]
end
subgraph AgentGroup["Agent"]
AgentNode["Agent"]
end
UserIn ==>|intent| HarnessNode
HarnessNode ==>|request| AgentNode
AgentNode -.->|pre-inference| DecapodNode
DecapodNode -.->|context, gates| AgentNode
AgentNode ==>|inference| ModelNode
ModelNode ==>|response| AgentNode
AgentNode -.->|post-inference| DecapodNode
DecapodNode -.->|proof| AgentNode
AgentNode ==>|verified result| UserOut
AgentNode -.->|clarify| UserIn
style UserIn fill:#9f1239,stroke:#4c0519,color:#ffffff
style UserOut fill:#9f1239,stroke:#4c0519,color:#ffffff
style HarnessNode fill:#1e40af,stroke:#172554,color:#ffffff
style AgentNode fill:#6b21a8,stroke:#3b0764,color:#ffffff
style ModelNode fill:#155e75,stroke:#083344,color:#ffffff
style DecapodNode fill:#111827,stroke:#f59e0b,color:#fbbf24,stroke-width:3px
style HumanGroup fill:#fff1f2,stroke:#9f1239,color:#4c0519,stroke-width:2px
style HarnessGroup fill:#dbeafe,stroke:#1e40af,color:#172554,stroke-width:2px
style IntelligenceGroup fill:#cffafe,stroke:#155e75,color:#083344,stroke-width:2px
style GovernanceGroup fill:#fef3c7,stroke:#b45309,color:#78350f,stroke-width:2px
style AgentGroup fill:#f3e8ff,stroke:#6b21a8,color:#3b0764,stroke-width:2px
linkStyle default stroke:#334155,stroke-width:1.5px
The model stays fixed. The corridor around it changes: resolve project context before implementation, gate completion on proof. A later agent can re-enter the same work without an ad hoc briefing.
- Intent β Vague requests become explicit, versioned specifications.
- Context β Only the relevant code and docs, with provenance.
- Coordination β Exclusive claims and isolated workspaces for concurrent agents.
- Trajectory β Durable events for handoff across sessions and harnesses.
- Boundaries β Protected branches and sensitive modules stay protected.
- Adaptation β Instruction changes go through explicit review.
- Proof β Completion requires deterministic verification.
.decapod/ holds the durable state of governed agent work:
.decapod/
managed/specs/ # Living project specifications
managed/sessions/ # Session custody
generated/ # Context capsules and proof artifacts
data/ # Local project store
governance/ # Trajectory, claims, and validation
workspaces/ # Isolated worktrees (containers when configured)
config.toml # Project shape
OVERRIDE.md # Local authority, in plain Markdown
- Intent becomes specs and todos.
- Context becomes scoped capsules.
- Custody becomes claimed tasks and workspaces.
- Trajectory becomes event-backed records.
- Boundaries become rules, overrides, and gates.
- Validation becomes proof artifacts.
- Completion becomes a verified state transition.
OVERRIDE.md is yours to edit. Write instructions the way you write documentation.
Decapod binds each directive to its source and body with fail-closed resolution.
Decapod does not make agents smarter with longer chats. It makes agent work shippable by turning governance into repository state.
Decapod ships with an embedded engineering constitution β architecture, security, performance, and testing as queryable doctrine.
Agents consult it, cite claims, follow gates, and produce proof. Judgment remains human. Authority is explicit: baseline constitution, project override, task-scoped projection.
- Daemonless β Invoked like
gitorgrep. - Local-first β Governance runs on your machine by default.
- Repo-native β State stays with the repository.
- Provider-agnostic β Any model, any harness. The project is the coordination surface.
- Proof-gated completion β
VERIFIEDrequires passed proof-plan gates. - Branch isolation β Protected paths and branches stay protected.
Decapod is a governance kernel β not an agent framework, prompt pack, or model router.
git clone https://github.com/DecapodLabs/decapod
cd decapod
cargo build && cargo testCONTRIBUTING Β· SECURITY Β· Issues