iplan-runner

OSS · MIT · pre-1.0 — the execution plane. AIDoc Flow ends at the IPLAN; iplan-runner is what turns an approved IPLAN into auditable execution.

The framework is the control plane — BRD → PRD → … → IPLAN, "done when committed + green." iplan-runner is the operations plane on the other side of that line: it takes the IPLAN as input and never modifies the framework repo.

The model

IPLAN ─▶ Ledger ─▶ Gate ─▶ Monitor
plan    evidence   proof    observe
(from   (append-   (indep.  (OTel,
 SDD)    only,      completion SLOs)
         hashed)    check)

Ledger — an append-only, hash-chained, isolation-scoped record of what agents actually did: tasks, leases, evidence, blockers, reconciliation.
Gate — independent verification that a completion claim is actually valid, not self-attested.
Monitor — post-implementation OpenTelemetry signals and SLOs, bound to the same IPLAN / ledger identity.

Local-first, sync optional

iplan-runner runs an approved IPLAN with a local, append-only signed ledger, an independent gate, and a handover receipt — no network required. A single sync toggle (iplanic.sync, off by default) flips it from standalone to online, relaying signed events to iplanic (the hosted control plane) as the system-of-record. A standalone run can flush its offline ledger to iplanic at any time; canonical-JSON signing lets iplanic verify events produced offline.

The model transport ships an offline deterministic stub, so the full standalone path runs with zero external dependencies. Real LLM clients are optional extras.

Source

github.com/vladm3105/iplan-runner — MIT-licensed, pre-1.0. The full pipeline (IPLAN intake → run loop → gate → land → handover → monitor) runs end-to-end on both engines; expect breaking changes between minor versions.

Found a rough edge? The Support page — this is the lab, not the showroom.