Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Assign work to a specific workspace agent or external runtime. Picks the right adapter (Claude for reasoning, Codex for bulk changes, Gemini for multimodal). Creates tasks with parent chain tracking for full delegation lineage. Triggers on: "delegate", "assign", "hand off", "send to agent"
.claude/skills/miosa-osa-delegate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 90% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -7% | 0% |
> Assign work to a specific agent or runtime with adapter selection.
Route a task to the best available agent or runtime. Unlike /spawn (which creates new agents), /delegate assigns work to existing agents or selects the optimal adapter for the task type. Maintains a parent chain so every delegated task traces back to its origin. Handles adapter selection automatically: Claude for reasoning, Codex for bulk file changes, Gemini for multimodal inputs.
bash# Delegate to a specific running agent /delegate pricing-researcher "Also check competitor free tier limits" # Delegate with auto-adapter selection /delegate --auto "Refactor all API endpoints to use new auth middleware" # Delegate to a specific adapter /delegate --adapter codex "Update all import statements from v1 to v2 across 47 files" # Delegate with priority /delegate analyst "Revenue report needed for board meeting" --priority high # Delegate with dependencies /delegate writer "Draft announcement" --after task-123,task-456 # Delegate and wait for result /delegate researcher "Find the answer" --sync
| Flag | Type | Default | Description | |------|------|---------|-------------| | <target> | positional | — | Agent name or ID to delegate to | | <task> | positional | required | Task description | | --auto | flag | false | Auto-select the best agent/adapter for the task | | --adapter | enum | auto | Force adapter: claude, codex, gemini, ollama | | --priority | enum | normal | critical, high, normal, low | | --after | string] | — | Task IDs that must complete before this starts | | --deadline | duration | — | Soft deadline for completion | | --sync | flag | false | Block until task completes and return result | | --context | path] | — | Files to include as context for the delegate | | --output-format | string | — | Expected output format (guides the delegate) | | --parent | string | auto | Parent task ID (auto-set from current context) |
--auto, match task requirements against available agents and adapters:--after specified, verify prerequisite tasks exist. Queue until they complete./board status. Start deadline timer if set.--sync mode, poll for completion and return result. Otherwise, return task ID for tracking.markdown## Task Delegated | Field | Value | |-------|-------| | Task ID | task-2026-03-20-143500-d4e5 | | Target | pricing-researcher (agent-a1b2) | | Adapter | claude-sonnet | | Priority | high | | Parent | task-2026-03-20-140000-a1b2 | | Dependencies | none | | Deadline | 2h | | Status | delivered | Track progress: `/board` or `/inbox peek pricing-researcher`
/spawn — May spawn a new agent if no suitable one exists/inbox — Delivers task messages/checkout — Locks task to prevent double-work/budget — Checks budget availability before delegating| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,681 | 4,156 | -27% | 1 | 1 | 0% | 1,032 | 1,755 | +70% | 0 | 0 | — |
case-02 | fail→pass | 21,446 | 5,639 | -74% | 1 | 1 | 0% | 2,832 | 1,851 | -35% | 0 | 0 | — |
case-03 | fail→pass | 6,223 | 5,812 | -7% | 1 | 1 | 0% | 1,115 | 2,113 | +90% | 0 | 0 | — |
case-04 | fail→pass | 13,102 | 10,566 | -19% | 1 | 1 | 0% | 2,163 | 2,438 | +13% | 0 | 0 | — |
case-05 | fail→pass | 6,400 | 2,756 | -57% | 1 | 1 | 0% | 961 | 1,433 | +49% | 0 | 0 | — |
case-06 | fail→pass | 9,337 | 3,033 | -68% | 1 | 1 | 0% | 1,531 | 1,419 | -7% | 0 | 0 | — |
case-07 | fail→pass | 43,239 | 5,708 | -87% | 1 | 1 | 0% | 6,276 | 1,752 | -72% | 0 | 0 | — |
case-08 | fail→pass | 14,134 | 4,496 | -68% | 1 | 1 | 0% | 1,904 | 1,748 | -8% | 0 | 0 | — |
case-09 | fail→pass | 6,085 | 2,599 | -57% | 1 | 1 | 0% | 961 | 1,391 | +45% | 0 | 0 | — |
case-10 | fail→pass | 10,274 | 2,737 | -73% | 1 | 1 | 0% | 1,471 | 1,404 | -5% | 0 | 0 | — |
case-11 | fail→fail | 9,201 | 2,853 | -69% | 1 | 1 | 0% | 1,413 | 1,370 | -3% | 0 | 0 | — |
case-12 | pass→pass | 12,230 | 5,016 | -59% | 1 | 1 | 0% | 1,888 | 1,710 | -9% | 0 | 0 | — |
case-13 | fail→pass | 9,825 | 2,809 | -71% | 1 | 1 | 0% | 1,264 | 1,328 | +5% | 0 | 0 | — |
case-14 | pass→pass | 6,835 | 2,480 | -64% | 1 | 1 | 0% | 1,044 | 1,280 | +23% | 0 | 0 | — |
case-15 | pass→pass | 9,958 | 2,913 | -71% | 1 | 1 | 0% | 1,481 | 1,410 | -5% | 0 | 0 | — |
case-16 | fail→pass | 26,823 | 2,621 | -90% | 1 | 1 | 0% | 1,935 | 1,412 | -27% | 0 | 0 | — |
case-17 | fail→pass | 3,940 | 4,840 | +23% | 1 | 1 | 0% | 656 | 1,658 | +153% | 0 | 0 | — |
case-18 | fail→pass | 6,169 | 4,103 | -33% | 1 | 1 | 0% | 928 | 1,635 | +76% | 0 | 0 | — |
case-19 | fail→pass | 11,031 | 2,451 | -78% | 1 | 1 | 0% | 1,837 | 1,388 | -24% | 0 | 0 | — |
case-20 | fail→pass | 13,625 | 3,530 | -74% | 1 | 1 | 0% | 2,251 | 1,562 | -31% | 0 | 0 | — |
case-21 | fail→pass | 5,205 | 3,502 | -33% | 1 | 1 | 0% | 815 | 1,572 | +93% | 0 | 0 | — |
case-22 | fail→pass | 12,258 | 3,974 | -68% | 1 | 1 | 0% | 1,712 | 1,539 | -10% | 0 | 0 | — |
case-23 | pass→pass | 12,454 | 14,107 | +13% | 1 | 1 | 0% | 2,217 | 2,979 | +34% | 0 | 0 | — |
case-24 | fail→pass | 12,411 | 5,436 | -56% | 1 | 1 | 0% | 1,974 | 1,689 | -14% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +75 percentage points is the difference between those two pass rates over the 24 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.