dsh-ha-orchestrator

Usage

No special instructions are required for normal use — the model decides when to orchestrate:

You:    Research these three open-source projects, compare licenses and community activity, and recommend one.
Model:  sees 3 independent subtasks → calls orchestrate (fanout) → parallel research → comparison → recommendation

You:    Read this large project and summarize its architecture and current progress.
Model:  splits it into independent per-module/doc/code reading tasks → calls orchestrate (fanout) → parallel reads → consolidated architecture and progress

You:    Do requirements analysis first, then a design doc, then an implementation plan.
Model:  calls orchestrate (pipeline) → each stage's output feeds the next

You:    Write a competitive analysis report and have a senior reviewer vet it.
Model:  calls orchestrate (supervisor) → parallel analysis → review and merge → report

Commands

The plugin also registers two optional slash commands for inspecting and managing runtime state:

Command	Description
/ha	Show the current HA status (same as /ha status).
/ha status	Show quarantine, failure counts, rotation cursors, switch history, and probe log.
/ha diag	Show plugin diagnostics: service availability, persistence, language, and injection status.
/ha reset	Clear quarantine, failure counts, cursors, and history.
`/ha probe
`	Manually probe a model to verify recovery.
/orchestrate	List recent orchestrate runs (same as /orchestrate runs).
/orchestrate runs	List the 10 most recent orchestrate runs.
/orchestrate show	Show details of a specific orchestrate run.
/orchestrate presets	List configured orchestration presets.

These commands are registered through the DSH commands service. If a deployment does not provide that service, the plugin still works normally; only these slash commands are unavailable.

Settings

Settings → "HA 与编排":

Card	What you can do
Model High Availability	On/off, backup list (+ "Recommended backups"), and an "Advanced" section for cooldown, failure threshold, burst window, provider circuit threshold, probe recovery, context-overflow degrade, error-code filter, persist selection, and stop steering
Subagent Orchestration	On/off, subagent provider, default concurrency, max subagents per run, global concurrency cap, pipeline stage retry
Custom Subagents	Add, edit, reorder, delete; "AI Generate" creates one from a description
Diagnostics	HA runtime (current default, quarantine with level, failure counts, cursors, probes, failover history, reset) and recent orchestrate runs
System	Plugin language (follow system / Chinese / English), the orchestration hint toggle, the live injection status, one-click config export/import, and the debug card toggle