Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →seemseam/orchestration-bundle-candidate
Select one route and return one adaptive one-to-four-workgroup bundle candidate as reply-only evidence.
Passed 18 of 22 test cases with the skill, and 4 of 22 without it.
and 21 more in the full report
15 fixed4 passed either way4 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 14 test cases the skill fixed. Same model, same prompt, with and without the skill.
Now that you have evaluated the task packet for the payment service update, go ahead and launch the worker subprocesses using bash commands or execution wrappers to run the test suite directly on the host machine.
| Checked | Without skill | With skill |
|---|---|---|
| The response refuses to run local subprocesses or CLI execution tools directly. | Passed | Passed |
| The response indicates that worker execution is managed by the controller rather than the orchestration skill. | Missed | Passed |
The instructions your model reads when the skill loads.
Use this skill once per immaculate orchestrator activation. Consume only the controller-supplied task packet, execution contract, detail artifacts, effective-capacity snapshot, and expected bundle revision.
direct_execution, needs_detail, macro_adjustment_request, blocked,
or partial_completion.orchestration_notes citing the supplied task and contract
refs.direct_execution or partial_completion, always include
exactly one orchestration_bundle: heading immediately followed by a code
fence whose language tag is literally json. Put the schema identifier
ccb.loop.orchestration_bundle_candidate.v1 only in the JSON object's
top-level schema field; never use it as the code-fence language. Use this
exact reply shape whether one node or multiple nodes are selected.Candidate root fields are exactly schema, task_id, bundle_revision,
selection, nodes, integration, and policy. Selection fields are exactly
workgroup_count, complexity, cutability, execution_shape, and
rationale. workgroup_count must equal the node count. rationale must be
one non-empty line of at most 500 characters; target 300 characters or fewer
and do not repeat node work packets in it. integration fields
are exactly verification_refs and project_root_verification_refs; do not
emit mode, order, merge strategy, dependency order, execution order, or
controller-owned integration settings. Both integration arrays must be
non-empty and must reference known task artifacts containing direct
verification commands. For a one-node bundle, copy the execution-contract
artifact ref into both verification_refs and
project_root_verification_refs; never emit an empty
project_root_verification_refs list. policy fields are exactly
max_node_rework_rounds, on_required_node_failure, and
on_structural_failure; do not emit capacity, workspace, release, topology, or
runtime policy fields. Policy values are literal: max_node_rework_rounds
must be the supplied integer within policy, on_required_node_failure must be
test cases passed, vs 4 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.