ucsahinn/agentchef / plugins/agentchef/skills/adaptive-agent-routing
adaptive-agent-routing
VerifiedSelect the narrowest useful agent, skill, MCP, and config route for non-trivial work. Use when delegation, context placement, specialist ownership, duplicate skill aliases, or routing visibility materially affects execution; avoid for trivial single-surface work.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.1 KB |
| agents/openai.yaml | config | 310 B |
| references/global-working-agreements.md | markdown | 10.9 KB |
SKILL.md
View on GitHub---
name: adaptive-agent-routing
description: Select the narrowest useful agent, skill, MCP, and config route for non-trivial work. Use when delegation, context placement, specialist ownership, duplicate skill aliases, or routing visibility materially affects execution; avoid for trivial single-surface work.
---
# Adaptive Agent Routing
Select routes by expected value, not by catalog match alone.
## Workflow
1. Classify the task shape and identify the smallest owning surface.
2. Read [global-working-agreements.md](references/global-working-agreements.md) only when the detailed route, specialist map, alias policy, or MCP boundary is relevant.
3. Preserve the active session's profile and model. Role files carry only the catalog worker `model`; they never pin reasoning effort, and spawns pass no model override.
4. Spawn an agent only when one of the four catalog conditions holds: an autoVerify routing profile matched and files changed, so its verifier runs before the task is reported done; independent parallel work exists; noisy logs or research should be isolated from the main thread; the user explicitly requests delegation. Skip trivial, strictly sequential, tightly coupled, and single-file work where delegation adds coordination cost.
5. When a request matches a routing profile, load the profile's auto-skill first (suggest it instead when it is explicit-only), and run the required verifier for autoVerify profiles after file changes. Start at most 2 agents per task without the user naming them; treat ten threads as capacity, not a target, and size the workers per task to the work and to the live capacity recommendation (`beyin kapasite`, hard ceiling 8; a coordinator is not counted); a request above the recommendation is refused with the measured reason. "Takım koşusu" means `beyin takim` (the `team-conductor` skill), not the Workflow tool.
6. Pick one of two routes, both at most two levels deep: Direct (the main session briefs the specialists the task needs) or Team (for a user-created board task, the main session briefs its coordinator, which briefs only its own cataloged workers). Workers never spawn; they name a needed role under Open questions as `needs: <role> - <why>`. Brief and check handoffs with the `agent-brief` skill.
7. Emit one `Routing plan:` update and one `Routing result:` summary instead of per-surface narration.
8. Keep destructive, credentialed, account, database, publish, deploy, and external-write actions approval-gated.
## Effort scaling
Use one worker for a single file or strictly sequential work. Use a team only
for independent file groups. For an explicitly requested substantial team run,
plan 3-5 workers with 5-6 bounded tasks per worker when the scope warrants it;
the worker count never exceeds live capacity. These are sizing guidelines,
not a minimum or a reason to invent tasks. The automatic spawn cap still applies.
Keep one active owner per file. `team-run plan --serialize-overlaps` records
ordered dependencies for overlapping groups; otherwise planning refuses them.
## Output Contract
Return selected and skipped surfaces, summarized evidence, commands run, and blockers. Do not paste raw agent transcripts or large logs.
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: 472k | 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 |