Dispatching parallel agents

Skill funky-eyes/best-copilot/skills/dispatching-parallel-agents

Use when multiple frozen subtasks have non-overlapping write sets and can be implemented or reviewed in parallel.From its SKILL.md

Install

npx -y skills add funky-eyes/best-copilot --skill dispatching-parallel-agents

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

12 stars12 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.4 KB, as published. Nobody here has run it

Dispatching Parallel Agents

Preconditions

Each subtask has goal, files, dependencies, material assumptions or explicit unknowns, acceptance checks, and verification.
Write sets do not overlap.
Shared foundation work is already done or assigned as a dependency.
Fan-in owner is identified.

Fan-out Packet

Each parallel subtask uses the six-block PM dispatch packet from core-workflow-contract. At minimum include:

task_intent: goal, task_type, work_mode
frozen_scope: files_involved, write_set, dependencies, owner_lane, reviewer_lanes, review_package_ref, diff_ref
execution_contract: assumptions, tradeoffs, simpler_option_considered, constraints, acceptance_checks, verification_budget, model_policy, model_cost_boundary, cost_boundary, reviewer_input_boundary, reviewer_permission_boundary, forbidden_approaches, read_before_write_targets when code edits are planned
output_contract: required skills, required_artifacts

Add sub_task_id for fan-in tracking.

Model, Permission, And Token Boundaries

Every parallel dispatch records the intended owner_lane, reviewer_lane, permission_mode, write_set, review_package_ref or diff_ref, model_policy, and model_cost_boundary.
When the runtime supports subagent model selection, choose the cheapest adequate explicit model/tier for each lane instead of inheriting the current session default. If model selection is not enforceable, state that in model_policy and model_cost_boundary.enforcement.
Batch fan-out that uses many reviewers, high-tier models, or repeated large context requires PM-visible cost acknowledgement or a cheaper explicit model/context strategy before dispatch.
Review-only lanes use read-only permission mode and must not run mutating git/workspace commands. Reclassify to fix/implementation before allowing edits.
Shared diffs, specs, logs, and review packages should be referenced once and reused by path or shard, not pasted into every subagent prompt.

Fan-in

Adjudicate fan-in using the priority order from core-workflow-contract Fan-In Arbitration. Accept only structured handbacks with the required fields. Merge results by dependency order. Re-run integration verification after combining parallel work.