specialist-dispatch
Skill0 starsChanged 4 months agoMITView on GitHub
Selects scoped architecture, backend, frontend, security, data, DevOps, QA, docs, performance, or product roles.
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 4 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Use this skill when a task benefits from a specialist lens but does not require installing a large external agent marketplace.
- Distinctive: line 40
Do not spawn a specialist without a concrete question.
- Distinctive: line 42
Prefer read-only specialist review unless implementation ownership is clear.
- Very specific: line 43
Use agent-squad when multiple specialists will work in parallel.
- Very specific: line 44
Use subagent-result-merge to consolidate outputs.
- Very specific: line 62
When /multimodel is active, use budget models for patch proposals and stronger models for independent review.
- Distinctive: line 68
Use a native custom subagent only when the installed host documents and functionally verifies a separate provider/model and its effective …
- Very specific: line 78
Use agent-squad for concurrency limits and context-pack for exported data.
- Very specific: line 86
Start with synthetic data, explicit iteration/output/request limits and an outer process deadline.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Specialist Dispatch
For parallel author workflows, select
superpowers.dispatching-parallel-agents or
superpowers.subagent-driven-development through be skills route ID --json
and read the complete exposed provider. This skill retains Bible's specialist
selection and worker authorization policy separately. Missing providers or
delegation capabilities stay unresolved until the selected profile supplies
them and host exposure is verified.
Use this skill when a task benefits from a specialist lens but does not require installing a large external agent marketplace.
This skill turns broad agent catalogs into a small Codex-native dispatch layer. It can be used with subagents when available, or as a checklist for the main agent when no subagent tool exists.
Specialist Profiles
Choose the smallest set:
- architect: boundaries, dependency direction, module shape, migration seams;
- backend: APIs, domain logic, persistence, concurrency, errors;
- frontend: UI behavior, state, accessibility, visual quality;
- security: auth, authz, secrets, parser risk, supply chain;
- data: schemas, migrations, analytics, integrity, retention;
- DevOps: CI, deploy, runtime config, observability;
- QA: test plan, regression coverage, edge cases;
- docs: user-facing and operator-facing documentation;
- performance: evidence-led bottleneck analysis;
- product: user workflow, scope, acceptance criteria.
Dispatch Rules
- Do not spawn a specialist without a concrete question.
- Give each specialist bounded files and expected output.
- Prefer read-only specialist review unless implementation ownership is clear.
- Use
agent-squadwhen multiple specialists will work in parallel. - Use
subagent-result-mergeto consolidate outputs.
External Worker Roles
Map portable roles to verified providers and models in machine-local
configuration. Default: keep the current selected model; multimodel and quorum
are off. Preserve the parent's model, provider, and authorization. Only the
user's /multimodel command selects the budget-author/strong-review workflow.
/quorum selects its fixed reviewer roster and voting while keeping the author's
model unchanged. Same-model specialist lanes activate neither mode.
The following roles do not themselves authorize model changes:
- scout: inspect bounded context and return a map of facts;
- reviewer: review an allowed diff and related files with evidence;
- tests-docs: propose tests or documentation as an artifact, without applying it;
- patch-proposer: return a bounded code patch for the parent to inspect and apply;
- local-helper: perform a bounded task in the locally verified inference mode.
When /multimodel is active, use budget models for patch proposals and stronger
models for independent review. Role names do not prove availability, price or quality. Record actual
routes and usage, exclude the proposer and integrator from approval votes, and
never silently upgrade the implementation route. See
Cross-Provider Review.
Use a native custom subagent only when the installed host documents and functionally verifies a separate provider/model and its effective permissions. Otherwise use an explicit, locally verified external CLI capability. A config entry is not evidence of a route. No hidden fallback: an unavailable requested model returns blocked or failed; a replacement must be explicitly selected and recorded. An unverified tool-calling model is only a text helper.
The main agent owns decisions and applies changes. External workers may not
edit source, run arbitrary shell, merge, commit, push, install dependencies, or
spawn other workers. Enforce limits in the runtime, not only in role text.
Use agent-squad for concurrency limits and context-pack for exported data.
Alternative Runtime Pilots
Swarms or another programmatic runtime may be evaluated as an alternative
external worker for fixed roles, not a provider switch inside the native
parent's subagent tool. Export only the required operations and keep the
existing subagent-result-merge contract, source scope and parent ownership.
Start with synthetic data, explicit iteration/output/request limits and an
outer process deadline; an async timeout alone may leave work running.
A function exported over MCP proves that function path, not model delegation. Separate server initialization, tool discovery, function invocation, actual model execution and parent-to-worker execution in the result. No automatic loops, global registration, new router or hidden provider fallback. Keep library import side effects, credentials and telemetry inside reviewed private runtime configuration. Do not stack frameworks without a measured benefit.
Specialist Prompt Shape
Specialist:
Question:
Scope:
Allowed files:
Forbidden files:
Evidence required:
Expected output:
Validation:
Task ID and base commit:
Snapshot state and hash:
Requested route and required runtime evidence:
Allowed tools and enforced permissions:
Timeout and retry budget:
Result: subagent-result-merge contractOutput
Report:
- selected specialists;
- why each was needed;
- scope boundaries;
- result or next routing skill.
In this repository62 other files it gives its agents.
- AGENTS.mdAGENTS.md
- accessibility-ui-reviewSkill
- agent-retrospectiveSkill
- agents-md-retrospectiveSkill
- agent-squadSkill
- architecture-mapSkill
- architecture-normalizerSkill
- architecture-principlesSkill
- authz-boundary-reviewSkill
- c-cppSkill
- code-qualitySkill
- code-reviewSkill
- code-to-figmaSkill
- code-wiki-ruSkill
- context-packSkill
- core-engineeringSkill
- debuggingSkill
- dependency-advisory-auditSkill
- deserialization-parser-reviewSkill
- design-system-extractorSkill
- documentationSkill
- engineering-standardsSkill
- esp32Skill
- esphomeSkill
- external-agent-pack-auditSkill
- fastSkill
- figma-to-codeSkill
- fix-security-findingSkill
- goSkill
- homeassistantSkill
- mcp-tool-routerSkill
- migration-plannerSkill
- mobile-qaSkill
- multi-agent-pr-reviewSkill
- performanceSkill
- playwright-visual-qaSkill
- pythonSkill
- quality-gatesSkill
- refactoringSkill
- responsive-breakpoint-checkSkill
- review-routerSkill
- rustSkill
- secrets-and-config-reviewSkill
- security-diff-reviewSkill
- security-routerSkill
- securitySkill
- session-memorySkill
- subagent-result-mergeSkill
- supply-chain-reviewSkill
- tdd-guardSkill
- testing-tddSkill
- threat-modelSkill
- typescriptSkill
- ui-buildSkill
- ui-business-appsSkill
- ui-concept-firstSkill
- ui-figmaSkill
- ui-qaSkill
- ui-researchSkill
- ui-routerSkill
- The first 60 shown.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

