hashgraph-online/awesome-codex-plugins / plugins/boshu2/agentops/skills/orchestrate
orchestrate
VerifiedDispatch parallel workers without write collisions and decide what idle or finished ones do next. Use when: managing agents, or running or planning agents in parallel, even two.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 12.4 KB |
| agents/bulk-reader.toml | config | 2.1 KB |
| agents/code-writer.toml | config | 2.1 KB |
| references/context-budget-delegation.md | markdown | 9.6 KB |
| references/decide-once-apply-in-parallel.md | markdown | 2.3 KB |
| references/launch-and-isolation.md | markdown | 5.3 KB |
| references/recover-and-resume.md | markdown | 2.6 KB |
| references/session-associations.md | markdown | 4.0 KB |
| references/status-block.md | markdown | 698 B |
references/launch-and-isolation.md
View on GitHub# Launch and isolation
Conditional detail for [Orchestrate](../SKILL.md): what a refused dispatch
means, where a worker's isolation comes from, what disjoint files leave shared,
how a worker confirms its starting content, how to read an
authentication-shaped launch error, and what a worker does with a choice its
packet leaves open.
## Refused launches
Classify a rejected dispatch by whether a worker started. When the runtime
refuses the call itself, nothing ran:
- A malformed invocation, such as a bad argument, may be corrected once. That
is not a worker retry.
- An active-agent, concurrency or capacity limit is backpressure. Hold the
packet in a native wait until a slot frees, and launch the validated batch in
waves no larger than the limit; what each freed slot runs first stays the
caller's decision ([Orchestrate](../SKILL.md#before-any-dispatch)).
Waves under a limit are not a partial launch, a failed worker or a reason to
reroute, and holding a packet inside the caller's window is a wait, not an
AgentOps queue.
- A packet still unstarted when the caller's window ends is reported as
`not launched`, never as failed. A required judgment leg becomes unavailable
only then, not because a slot was busy.
- Any other failure to start is `launch failed`. Running that packet inline or
on another executor or model is the caller's decision, disclosed in its
result, and a substitute from another family never satisfies a required one
([model dispatch](../../validate/references/model-dispatch.md#selected-adapters)); running inline in a
context that is not fresh never counts as a fresh judgment leg.
A worker has failed only when it launched and then failed.
## Workspace isolation
Create a writer's workspace through the selected runtime's own isolation option
where it has one. A worktree made by hand is invisible to the runtime's tracking
and cleanup; it stays the caller's to remove, under the integration proof in
[the removal rule](recover-and-resume.md#prove-integration-before-removal).
Before creating one, detect isolation that already exists. Resolve the git
directory and the common git directory to absolute paths and compare them: when
they differ, the session is in a linked worktree, unless its git directory
belongs to a submodule, which a superproject check tells apart. A session
already in a linked worktree works there and never nests another worktree inside
it. Workers it launches still need their own isolation, as peers of that
worktree rather than inside it.
## Shared runtime
Disjoint write scopes do not make workers independent. Concurrent workers,
read-only ones included, contend when they or their checks and services write,
install into or hold exclusively the same:
- port or dev server;
- local database, cache or global config;
- browser profile, browser session or device;
- package install;
- an external API quota their checks call.
A shared model-provider limit is not one of these: it is backpressure under
[refused launches](#refused-launches), launched in waves, not serialized.
Name such resources among the packet's effects. Give each worker its own
instance, such as a port or path taken from its environment or a copied
database, or serialize only the packets that contend. A resource only one
process can hold serializes every packet that needs it.
## Starting content
A runtime-made workspace may be cut from another base than the caller's
checkout, such as the primary checkout's branch or the default branch, and
uncommitted changes never cross into it. Pass the intended starting commit with
the packet. Before its first edit the worker compares its HEAD with that commit
and stops on a mismatch; the result is a launch failure, never a candidate.
Resetting or rebasing the workspace to repair it is outside the packet's
authority.
## Authentication-shaped errors
From a sandboxed or network-restricted launch, an authentication-shaped error
does not show that the account is logged out: blocked egress can produce the
same text. Report the error verbatim. Attribute it to credentials, or suggest a
login, only once the launch is known to have reached the provider.
## Open choices
A headless worker cannot ask. When its packet leaves a choice open, it takes no
irreversible or out-of-scope step on a guess: it finishes the independent parts,
leaves existing content in place and returns the choice under `gaps` with its
recommendation. The caller answers a choice that accepted intent already
settles through the runtime's follow-up turn on that worker, and escalates only
a choice nothing settles. When a choice can be framed only after a scan,
dispatch the scan, settle the choice, then dispatch the action with the scan
result passed by reference.
## Recording identity
For parent and resume links, a session that spans several work items, or
frozen source bounds, follow
[session associations](session-associations.md#work-to-session-associations).
## Many items in one packet
When a packet carries many items (files, records, issues), give each a stable
ID before launch and join the returned results back to that list. An item with
no result is missing, not done; an output with no matching input is
unexplained. Report both by ID. Name each return field and its allowed values
in the packet when results will be counted, sorted, deduplicated or compared;
parsed prose is where counts drift.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |