功能描述

Route mid-session work to the right spawned model without changing the fixed main session. Use for coding, architecture, math, algorithms, web development, b...

使用说明 (SKILL.md)

Advanced Dispatcher

Name: Advanced Dispatcher
Author: omaression

Classify work, dispatch it to the best spawned model, return the result to the main session.

Non-negotiable rules

Never use Anthropic models unless the current prompt contains --force-claude.
Reject --use-claude, --force-opus, --no-opus.
--force-claude is prompt-scoped, not session-scoped.
Deterministic routing over ad hoc judgment.
openai-codex/* → long cache retention.
opencode-go/* → short cache retention.
Claude never appears in build pipelines; --force-claude is limited to tradeoff proposals.
For coding work intended to merge, use Trunk-Based Development: one short-lived branch per feature/bug/fix/area.
Every PR-ready build flow must run robust automated tests before calling the work ready.
Keep git-visible diffs minimal, atomic, and limited to necessary files.

Routing table

Domain	Model
Code & architecture	`openai-codex/gpt-5.4`
Math & algorithms	`opencode-go/glm-5`
Web dev & brainstorming	`opencode-go/minimax-m2.5`
Research & long context	`opencode-go/kimi-k2.5`
Quick scripts & formatting	`openai-codex/gpt-5.3-codex-spark`

Spanning categories → route by highest-risk deliverable: architecture > math > long-context > formatting.

Tradeoff protocol

Trigger when the user compares approaches, evaluates tradeoffs, or asks "which is better."

Default proposals: opencode-go/glm-5 + openai-codex/gpt-5.3-codex With --force-claude: anthropic/claude-sonnet-4-6 + anthropic/claude-opus-4-6 Judge (always): openai-codex/gpt-5.4

Build pipelines

Trigger on buildq:, build:, or buildx: prefix.

Pre-step: create or verify one short-lived branch for the scoped work.

buildq: (quick, 5 steps)

plan → gpt-5.4
implement → gpt-5.4
test → glm-5
simplify → gpt-5.3-codex
retest → glm-5

build: (standard, 10 steps)

parallel-plan-a → gpt-5.4
parallel-plan-b → glm-5
judge-plan → gpt-5.4
boilerplate → gpt-5.3-codex-spark
implement → gpt-5.4
test → glm-5
simplify → gpt-5.3-codex
retest → glm-5
review-resolve → gpt-5.4
final-test → glm-5

buildx: (strict, 12 steps)

parallel-plan-a → gpt-5.4
parallel-plan-b → glm-5
judge-plan → gpt-5.4
boilerplate → gpt-5.3-codex-spark
implement → gpt-5.4
test → glm-5
simplify → gpt-5.3-codex
retest → glm-5
review-resolve-a → gpt-5.4
test-a → glm-5
review-resolve-b → kimi-k2.5
final-test → glm-5

Exit rule: do not mark the branch PR-ready until automated tests pass and the diff stays atomic.

Judge output contract

Judge-plan must emit:

Selected architecture
Why it won
Project/file structure
Implementation order
Branch plan (name, scope boundary)
Test plan
PR/CI test gates
Simplification targets
Done criteria

For buildx:, also include:

Risk list
Likely failure modes
Review checklist

Simplify contract

Must:

Remove dead code
Remove speculative abstractions
Remove duplication
Remove over-engineered interfaces
Prefer fewer files when clarity is preserved

Must not:

Rewrite architecture
Add abstractions
Expand scope

Implementation contract

dispatcher.py must produce deterministic RoutePlan objects, expose route choice + cache retention + rationale + pipeline steps, support tradeoff/buildq/build/buildx, reject empty prompts and legacy flags.

Validation checklist

Run test_dispatcher.py — all pass.
Smoke test tradeoff, buildq:, build:, buildx:.
Confirm Claude unreachable without --force-claude.
Confirm Claude never in build pipelines.
Confirm cache retention correctness.
Confirm judge output includes branch plan and PR/CI test gates.

安全使用建议

This skill appears coherent and implements only routing logic (no network calls or credentials). Before installing: (1) review dispatcher.py yourself or run the included unit tests (test_dispatcher.py) to confirm behavior in your environment; (2) verify your platform enforces provider credentials and policy separately (the skill references Anthropic but does not request keys—platform will need keys to actually call those models); (3) if you are concerned about autonomous agent activity, remember autonomous invocation is allowed by default—restrict or audit skill invocation in agent policy if needed; (4) consider restricting or monitoring any use of the --force-claude flag if Anthropic usage is disallowed in your organization.

功能分析

Type: OpenClaw Skill Name: advanced-dispatcher-skill Version: 1.0.1-alpha The advanced-dispatcher skill is a well-structured routing and workflow management tool designed to delegate tasks to specific AI models based on domain and command triggers (e.g., build:, buildx:). The Python logic in dispatcher.py is focused on deterministic route planning and input validation, while SKILL.md provides clear operational constraints and multi-step pipeline definitions for the agent. No evidence of data exfiltration, malicious execution, or harmful prompt injection was found; the code and instructions are entirely consistent with the stated purpose of model dispatching and software development orchestration.

能力评估

✓ Purpose & Capability

Name/description match the implementation: dispatcher.py and tests implement deterministic routing rules, tradeoff and build pipelines, and explicit refusal of legacy flags. The models referenced in the routing table are appropriate for a routing skill and there are no unexpected credentials or system access requests.

✓ Instruction Scope

SKILL.md confines runtime behavior to classification/routing, pipeline composition, and policies about which models to prefer or forbid. It does not instruct reading files, environment variables, network endpoints, or transmitting data to third parties. The implementation enforces non-empty prompts, domain validation, and flag rejection as described.

✓ Install Mechanism

No install spec is provided (instruction-only style). Code files are included but there are no downloads, brew/npm installs, or archive extraction. This is low-risk from an install vector perspective.

✓ Credentials

The skill declares no required environment variables or credentials. Although it references provider-specific model IDs (openai-codex, opencode-go, anthropic), the skill only returns routing decisions and does not itself perform API calls or ask for keys, which is proportionate to its purpose.

✓ Persistence & Privilege

always is false and the skill does not request system-wide persistence or modify other skills. Autonomous invocation is allowed (platform default) but the skill's actions are confined to returning RoutePlan objects.

版本历史

v1.0.1-alpha

- Expanded "Judge output contract" with a detailed list of required outputs for judge-plan, especially for `buildx:`, including risk lists and failure modes. - Added a new "Simplify contract" section specifying what simplification steps must and must not do. - No changes to routing rules, pipelines, or main contract logic. - Documentation in SKILL.md is now more explicit on output expectations.

v1.0.0-alpha

Initial alpha release with structured dispatch, model selection, and build pipeline support: - Added skill documentation (SKILL.md, README.md) and initial tests (test_dispatcher.py). - Defined strict model routing rules with prompt-scoped Claude enablement and cache retention policies. - Detailed routing table, tradeoff evaluation flow, and three build pipeline protocols (`buildq:`, `build:`, `buildx:`). - Enforced minimal atomic git diffs and required automated testing before work is PR-ready. - Centralized all contracts and step outputs for reproducible pipeline runs and validation.

v0.5.0

Advanced build pipeline routing and strict Claude isolation. - Added structured build pipeline support for `buildq:`, `build:`, and `buildx:` prompt triggers, each with detailed, multi-step routing. - Enforced that Claude models are never used in build pipelines, even with `--force-claude`; Claude remains exclusive to tradeoff proposal generation. - Expanded and formalized contracts for judge and simplify steps within build workflows. - Enhanced implementation contract and validation checklist to cover new pipeline modes and stricter model constraints.

v0.3.0

Rebuild as a stricter ClawHub-friendly dispatcher. Update routing to GPT-5.4 for code/architecture, GLM-5 for math/algorithms, Minimax M2.5 for web dev/brainstorming, Kimi K2.5 for research/long context, and GPT-5.3 Codex Spark for quick scripts/formatting. Tradeoff evaluation now uses GLM-5 + GPT-5.3-Codex with GPT-5.4 as judge, and Claude is only allowed through prompt-scoped --force-claude.

v0.2.2

- Added explicit data egress guardrails: external model dispatch now requires user consent via `--allow-external` or runtime consent. - Updated dispatcher to block all outbound provider runs unless egress permission is given, improving privacy and compliance. - Expanded runtime requirements section: self-hosted deploys must set provider credentials in environment variables, and `external_data_egress` is now declared in skill manifest. - Minor: Skill trigger pattern extended to detect `--allow-external` flag. Implementation and tests revised to enforce consent requirements across all routing logic. - Updated docs and security notes to reflect new consent and egress controls.

v0.2.1

**Changelog v0.2.1** - Added dispatcher logic (dispatcher.py) and tests (tests/test_dispatcher.py) to the repository. - Scoped activation: Dispatcher now triggers only on specific flags, routing, or tradeoff keywords, not on all messages. - Updated SKILL.md to document credential needs, narrow permission usage, and explicitly limit Anthropic model usage to controlled cases. - Code enforces dispatcher scope via new trigger matching and gating in `should_dispatch(...)` and route logic. - Tightened security posture: No persistent writes required; permissions limited to model spawning and parallelization. - Expanded test coverage for routing scenarios, protocol adherence, and flag/override handling.

v0.1.0

**Major upgrade: Introduces dynamic, flag-based task routing and Mixture of Experts protocols for cost-effective, granular model selection.** - Adds granular routing logic: sub-tasks are dispatched to specialized models based on intent (coding, research, creative, utility). - Explicit user flags (`--use-claude`, `--force-opus`, `--no-opus`) now dynamically override standard routing, enabling strict Anthropic/Opus gating. - Implements a multi-model “Tradeoff Evaluation Protocol”: parallel model generation with cost/labor escalation only on user demand or complexity threshold. - Introduces session transition checks, memory state handling, and repository fetch rules for improved performance and reduced context bloat. - Robust fallback, validation, and failure modes included for resilient operation. - Comprehensive test coverage for routing, override logic, tradeoff flows, and error handling.

元数据

Slug advanced-dispatcher-skill

版本 1.0.1-alpha

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 7

常见问题