2026-09 · agent-host-integration

DSH × LoopX Replan

A real DSH session changes a logging-library decision after a serverless constraint while preserving the original evidence and successor work.

The user can change a material constraint without losing the decision trail, while DSH remains the execution host and LoopX keeps the work frontier visible.

Reproducible demoreal_recording_with_reproducible_fixtureDSHreal recordingreproducibleAppendix case
LoopX DSH × LoopX Replan native skill activation reproducible synthetic d goal / trigger todo / gate / evidence public outcome
01

Case context

A new material constraint can invalidate an agent's earlier decision, but ordinary chat correction can erase why the plan changed and what still needs validation.

02

Repository evidence

Proof

A host-native agent can change plans without erasing why the old plan was superseded.

LoopX intervention

native skill activation, Replan, successor Todo, evidence, GoalBar

60sreal recording
12→4dependency packages
3/3behavior tests
2/2GoalBar Todos
Proof

A host-native agent can change plans without erasing why the old plan was superseded.

DSH × LoopX demo cover showing explicit skill activation, durable work, and a real Replan from Pino to Roarr
Public cover derived from the edited DSH recording; the linked case provides the video, fixture, and evidence boundary.
03

LoopX behavior

04

What the user sees

The user can change a material constraint without losing the decision trail, while DSH remains the execution host and LoopX keeps the work frontier visible.

05

Repository sources

Case notedocs/showcases/cases/dsh-loopx-replan-demo.md

Evidence boundary. Edited real DSH recording plus a synthetic public fixture and tests; credentials, provider configuration, raw reasoning, setup retries, local LoopX state, and the unedited recording are excluded.

Case note
Demopython3 examples/dsh-loopx-demo-smoke.py