godmode

分类: 工具与效率 | 上传者: arbazkhan971 | 下载: 0 | 版本: v1.0（最新）

打开上帝模式。 126个技能，7个子代理，零配置。自动路由到正确的技能。

更新日志: Source: GitHub https://github.com/arbazkhan971/godmode

目录结构

当前层级: tree/master/

📁 .claude-plugin/
- 📄 marketplace.json 85.1 KB
- 📄 plugin.json 427 B
📁 .codex/
- 📁 agents/
  - 📄 builder.toml 986 B
  - 📄 explorer.toml 1.2 KB
  - 📄 optimizer.toml 1.2 KB
  - 📄 planner.toml 1.4 KB
  - 📄 reviewer.toml 952 B
  - 📄 security.toml 948 B
  - 📄 tester.toml 1.0 KB
- 📄 config.toml 71 B
📁 .github/
- 📄 FUNDING.yml 21 B
📁 .godmode/
- 📄 measure-quality-v2.sh 1.8 KB
- 📄 measure-quality-v3.sh 3.2 KB
- 📄 measure-quality-v4.sh 1.7 KB
- 📄 measure-quality.sh 1.3 KB
- 📄 optimize-results.tsv 1.1 KB
📁 adapters/
- 📁 codex/
  - 📄 codex-config.md 6.8 KB
  - 📄 install.sh 7.9 KB
  - 📄 README.md 9.5 KB
  - 📄 verify.sh 3.3 KB
- 📁 cursor/
  - 📄 .cursorrules 12.1 KB
  - 📄 install.sh 5.9 KB
  - 📄 README.md 7.8 KB
  - 📄 verify.sh 3.1 KB
- 📁 gemini/
  - 📄 gemini-config.md 4.2 KB
  - 📄 install.sh 6.3 KB
  - 📄 README.md 4.4 KB
  - 📄 verify.sh 2.9 KB
- 📁 opencode/
  - 📄 install.sh 6.4 KB
  - 📄 plugin.json 3.0 KB
  - 📄 README.md 7.8 KB
  - 📄 verify.sh 3.1 KB
- 📁 shared/
  - 📄 sequential-dispatch.md 7.7 KB
  - 📄 verify-common.sh 9.5 KB
📁 agents/
- 📄 builder.md 6.3 KB
- 📄 code-reviewer.md 2.6 KB
- 📄 explorer.md 7.5 KB
- 📄 optimizer.md 7.7 KB
- 📄 planner.md 8.2 KB
- 📄 reviewer.md 7.7 KB
- 📄 security.md 8.8 KB
- 📄 spec-reviewer.md 2.6 KB
- 📄 tester.md 8.8 KB
📁 commands/
- 📁 godmode/
  - 📄 a11y.md 2.5 KB
  - 📄 agent.md 3.2 KB
  - 📄 analytics.md 3.6 KB
  - 📄 angular.md 2.9 KB
  - 📄 api.md 2.1 KB
  - 📄 apidocs.md 3.2 KB
  - 📄 architect.md 2.3 KB
  - 📄 auth.md 2.8 KB
  - 📄 automate.md 2.8 KB
  - 📄 backup.md 2.4 KB
  - 📄 build.md 1.5 KB
  - 📄 cache.md 2.7 KB
  - 📄 changelog.md 2.4 KB
  - 📄 chaos.md 2.3 KB
  - 📄 chart.md 3.1 KB
  - 📄 cicd.md 2.1 KB
  - 📄 cli.md 2.6 KB
  - 📄 comply.md 2.3 KB
  - 📄 concurrent.md 2.6 KB
  - 📄 config.md 2.0 KB
  - 📄 cost.md 2.1 KB
  - 📄 cron.md 2.8 KB
  - 📄 crypto.md 2.4 KB
  - 📄 ddd.md 2.6 KB
  - 📄 debug.md 1.8 KB
  - 📄 deploy.md 2.4 KB
  - 📄 designsystem.md 2.7 KB
  - 📄 devsecops.md 2.5 KB
  - 📄 distributed.md 2.8 KB
  - 📄 django.md 2.7 KB
  - 📄 docker.md 2.5 KB
  - 📄 docs.md 2.0 KB
  - 📄 e2e.md 2.1 KB
  - 📄 edge.md 2.8 KB
  - 📄 email.md 2.5 KB
  - 📄 eval.md 3.1 KB
  - 📄 event.md 2.6 KB
  - 📄 experiment.md 4.4 KB
  - 📄 fastapi.md 2.9 KB
  - 📄 feature.md 2.9 KB
  - 📄 fix.md 1.6 KB
  - 📄 forms.md 2.5 KB
  - 📄 ghactions.md 3.1 KB
  - 📄 git.md 2.4 KB
  - 📄 graphql.md 2.7 KB
  - 📄 grpc.md 2.6 KB
  - 📄 i18n.md 2.3 KB
  - 📄 incident.md 2.4 KB
  - 📄 infra.md 2.1 KB
  - 📄 integration.md 2.3 KB
  - 📄 k8s.md 2.0 KB
  - 📄 laravel.md 2.9 KB
  - 📄 legacy.md 3.1 KB
  - 📄 lint.md 2.6 KB
  - 📄 loadtest.md 2.4 KB
  - 📄 logging.md 3.1 KB
  - 📄 micro.md 2.7 KB
  - 📄 migrate.md 2.9 KB
  - 📄 migration.md 3.0 KB
  - 📄 ml.md 2.6 KB
  - 📄 mlops.md 3.2 KB
  - 📄 mobile.md 2.3 KB
  - 📄 monorepo.md 2.6 KB
  - 📄 network.md 2.3 KB
  - 📄 nextjs.md 2.6 KB
  - 📄 node.md 2.7 KB
  - 📄 nosql.md 3.0 KB
  - 📄 notify.md 3.0 KB
  - 📄 npm.md 2.5 KB
  - 📄 observe.md 2.0 KB
  - 📄 onboard.md 1.8 KB
  - 📄 opensource.md 2.8 KB
  - 📄 optimize.md 2.2 KB
  - 📄 orm.md 2.6 KB
  - 📄 pattern.md 2.3 KB
  - 📄 pay.md 2.4 KB
  - 📄 pentest.md 2.4 KB
  - 📄 perf.md 2.4 KB
  - 📄 pipeline.md 3.1 KB
  - 📄 plan.md 1.4 KB
  - 📄 postgres.md 3.1 KB
  - 📄 pr.md 2.1 KB
  - 📄 prompt.md 3.1 KB
  - 📄 query.md 2.6 KB
  - 📄 queue.md 2.5 KB
  - 📄 rag.md 2.8 KB
  - 📄 rails.md 2.8 KB
  - 📄 ratelimit.md 3.1 KB
  - 📄 rbac.md 2.8 KB
  - 📄 react.md 2.6 KB
  - 📄 realtime.md 2.6 KB
  - 📄 redis.md 3.1 KB
  - 📄 refactor.md 2.2 KB
  - 📄 reliability.md 3.1 KB
  - 📄 resilience.md 3.3 KB
  - 📄 responsive.md 2.6 KB
  - 📄 rfc.md 1.8 KB
  - 📄 scale.md 2.6 KB
  - 📄 schema.md 2.9 KB
  - 📄 search.md 2.5 KB
  - 📄 secrets.md 1.9 KB
  - 📄 secure.md 1.9 KB
  - 📄 seed.md 2.9 KB
  - 📄 seo.md 2.5 KB
  - 📄 ship.md 1.7 KB
  - 📄 slo.md 3.4 KB
  - 📄 spring.md 2.8 KB
  - 📄 state.md 2.7 KB
  - 📄 storage.md 2.4 KB
  - 📄 svelte.md 3.0 KB
  - 📄 tailwind.md 3.1 KB
  - 📄 think.md 1.3 KB
  - 📄 type.md 2.6 KB
  - 📄 ui.md 2.3 KB
  - 📄 upload.md 3.2 KB
  - 📄 vue.md 2.7 KB
  - 📄 webhook.md 3.1 KB
  - 📄 webperf.md 2.9 KB
- 📄 godmode.md 2.6 KB
📁 docs/
- 📁 blog/
  - 📄 godmode-meets-autoresearch.md 16.7 KB
- 📁 examples/
  - 📄 api-feature.md 4.1 KB
  - 📄 bug-fix.md 4.4 KB
  - 📄 optimization.md 6.5 KB
- 📁 languages/
  - 📄 csharp.md 17.6 KB
  - 📄 elixir.md 22.8 KB
  - 📄 go.md 13.6 KB
  - 📄 java.md 18.9 KB
  - 📄 kotlin.md 16.6 KB
  - 📄 php.md 17.1 KB
  - 📄 python.md 14.6 KB
  - 📄 react.md 18.4 KB
  - 📄 ruby.md 17.8 KB
  - 📄 rust.md 14.8 KB
  - 📄 swift.md 16.8 KB
  - 📄 typescript.md 11.3 KB
- 📁 platforms/
  - 📄 aws.md 11.8 KB
  - 📄 azure.md 12.4 KB
  - 📄 gcp.md 11.2 KB
  - 📄 supabase.md 17.0 KB
  - 📄 vercel.md 14.5 KB
- 📁 recipes/
  - 📄 ai-chatbot.md 21.6 KB
  - 📄 ai-product.md 20.9 KB
  - 📄 api-gateway.md 15.3 KB
  - 📄 auth-system.md 19.4 KB
  - 📄 cli-tool.md 17.2 KB
  - 📄 compliance-journey.md 18.5 KB
  - 📄 data-pipeline.md 14.2 KB
  - 📄 design-system.md 17.2 KB
  - 📄 docker-k8s.md 17.8 KB
  - 📄 event-system.md 20.4 KB
  - 📄 fullstack-feature.md 20.8 KB
  - 📄 graphql-api.md 17.0 KB
  - 📄 greenfield-saas.md 18.4 KB
  - 📄 incident-response.md 14.3 KB
  - 📄 legacy-modernization.md 17.5 KB
  - 📄 microservices-migration.md 22.4 KB
  - 📄 mobile-app.md 15.5 KB
  - 📄 monitoring-setup.md 16.9 KB
  - 📄 open-source-release.md 17.9 KB
  - 📄 performance-optimization.md 19.2 KB
  - 📄 realtime-chat.md 18.8 KB
  - 📄 refactoring-legacy.md 21.0 KB
  - 📄 security-hardening.md 20.8 KB
  - 📄 startup-mvp.md 12.4 KB
  - 📄 team-onboarding.md 23.5 KB
- 📁 testing/
  - 📄 cypress.md 31.6 KB
  - 📄 go-testing.md 32.3 KB
  - 📄 jest-vitest.md 21.0 KB
  - 📄 playwright.md 27.6 KB
  - 📄 pytest.md 31.3 KB
- 📁 tutorials/
  - 📄 first-skill.md 16.4 KB
- 📄 architecture.md 8.8 KB
- 📄 auto-generated-index.md 22.2 KB
- 📄 autoresearch-integration.md 9.1 KB
- 📄 chaining.md 5.4 KB
- 📄 ci-cd.md 7.0 KB
- 📄 COMPLETE-SKILL-LIST.md 21.2 KB
- 📄 decision-tree.md 10.8 KB
- 📄 domain-guide.md 7.1 KB
- 📄 FAQ.md 16.1 KB
- 📄 getting-started.md 5.1 KB
- 📄 godmode-design.md 557.1 KB
- 📄 index.md 4.5 KB
- 📄 PHILOSOPHY.md 9.0 KB
- 📄 platform-comparison.md 11.0 KB
- 📄 quick-reference.md 13.6 KB
- 📄 quick-start.md 13.5 KB
- 📄 results-log.tsv 113 B
- 📄 SECURITY.md 10.1 KB
- 📄 skill-chaining.md 5.5 KB
- 📄 skill-chains.md 12.1 KB
- 📄 skill-dependencies.md 16.4 KB
- 📄 skill-index.md 15.1 KB
- 📄 skill-matrix.md 25.8 KB
- 📄 troubleshooting.md 16.8 KB
- 📄 tsv-schemas.md 14.8 KB
📁 hooks/
- 📄 hooks.json 1.4 KB
- 📄 init.sh 5.1 KB
- 📄 session-start 3.0 KB
📁 scripts/
- 📄 count-skills.sh 8.0 KB
- 📄 generate-index.sh 6.8 KB
📁 skills/
- 📁 a11y/
  - 📄 SKILL.md 8.9 KB
- 📁 agent/
  - 📄 SKILL.md 11.5 KB
- 📁 analytics/
  - 📄 SKILL.md 11.0 KB
- 📁 angular/
  - 📄 SKILL.md 9.3 KB
- 📁 api/
  - 📁 references/
    
    📄 api-design-patterns.md 21.4 KB
  - 📄 SKILL.md 9.9 KB
- 📁 apidocs/
  - 📄 SKILL.md 10.0 KB
- 📁 architect/
  - 📁 references/
    
    📄 architecture-patterns.md 27.8 KB
    
    📄 system-design-patterns.md 42.4 KB
  - 📄 SKILL.md 7.4 KB
- 📁 auth/
  - 📁 references/
    
    📄 auth-flows.md 22.5 KB
  - 📄 SKILL.md 11.2 KB
- 📁 automate/
  - 📄 SKILL.md 6.5 KB
- 📁 backup/
  - 📄 SKILL.md 10.1 KB
- 📁 build/
  - 📄 SKILL.md 8.0 KB
- 📁 cache/
  - 📄 SKILL.md 8.0 KB
- 📁 changelog/
  - 📄 SKILL.md 7.7 KB
- 📁 chaos/
  - 📄 SKILL.md 10.0 KB
- 📁 chart/
  - 📄 SKILL.md 11.6 KB
- 📁 cicd/
  - 📄 SKILL.md 9.8 KB
- 📁 cli/
  - 📄 SKILL.md 10.9 KB
- 📁 comply/
  - 📄 SKILL.md 11.0 KB
- 📁 concurrent/
  - 📁 references/
    
    📄 concurrency-patterns.md 23.0 KB
  - 📄 SKILL.md 9.0 KB
- 📁 config/
  - 📄 SKILL.md 11.5 KB
- 📁 cost/
  - 📄 SKILL.md 10.4 KB
- 📁 cron/
  - 📄 SKILL.md 10.8 KB
- 📁 crypto/
  - 📄 SKILL.md 7.5 KB
- 📁 ddd/
  - 📄 SKILL.md 11.7 KB
- 📁 debug/
  - 📁 references/
    
    📄 debug-workflow.md 8.4 KB
  - 📄 SKILL.md 7.0 KB
- 📁 deploy/
  - 📄 SKILL.md 9.7 KB
- 📁 designsystem/
  - 📄 SKILL.md 8.7 KB
- 📁 devsecops/
  - 📄 SKILL.md 10.5 KB
- 📁 distributed/
  - 📄 SKILL.md 10.1 KB
- 📁 django/
  - 📄 SKILL.md 10.8 KB
- 📁 docker/
  - 📄 SKILL.md 10.6 KB
- 📁 docs/
  - 📄 SKILL.md 9.3 KB
- 📁 e2e/
  - 📄 SKILL.md 6.0 KB
- 📁 edge/
  - 📄 SKILL.md 6.9 KB
- 📁 email/
  - 📄 SKILL.md 5.8 KB
- 📁 eval/
  - 📄 SKILL.md 6.4 KB
- 📁 event/
  - 📁 references/
    
    📄 event-patterns.md 24.9 KB
  - 📄 SKILL.md 6.2 KB
- 📁 experiment/
  - 📄 SKILL.md 6.4 KB
- 📁 fastapi/
  - 📄 SKILL.md 6.9 KB
- 📁 feature/
  - 📄 SKILL.md 5.9 KB
- 📁 finish/
  - 📄 SKILL.md 4.5 KB
- 📁 fix/
  - 📁 references/
    
    📄 fix-workflow.md 6.1 KB
  - 📄 SKILL.md 4.2 KB
- 📁 forms/
  - 📄 SKILL.md 6.1 KB
- 📁 ghactions/
  - 📄 SKILL.md 5.9 KB
- 📁 git/
  - 📄 SKILL.md 6.6 KB
- 📁 godmode/
  - 📄 SKILL.md 6.6 KB
- 📁 graphql/
  - 📄 SKILL.md 7.1 KB
- 📁 grpc/
  - 📄 SKILL.md 7.4 KB
- 📁 i18n/
  - 📄 SKILL.md 6.4 KB
- 📁 incident/
  - 📄 SKILL.md 6.8 KB
- 📁 infra/
  - 📄 SKILL.md 6.2 KB
- 📁 integration/
  - 📄 SKILL.md 6.3 KB
- 📁 k8s/
  - 📁 references/
    
    📄 k8s-patterns.md 23.6 KB
  - 📄 SKILL.md 6.1 KB
- 📁 laravel/
  - 📄 SKILL.md 7.2 KB
- 📁 legacy/
  - 📄 SKILL.md 7.0 KB
- 📁 lint/
  - 📄 SKILL.md 5.9 KB
- 📁 loadtest/
  - 📄 SKILL.md 6.1 KB
- 📁 logging/
  - 📄 SKILL.md 8.0 KB
- 📁 micro/
  - 📁 references/
    
    📄 microservices-patterns.md 27.1 KB
  - 📄 SKILL.md 3.9 KB
- 📁 migrate/
  - 📄 SKILL.md 4.2 KB
- 📁 migration/
  - 📄 SKILL.md 4.3 KB
- 📁 ml/
  - 📁 references/
    
    📄 ml-evaluation.md 21.3 KB
  - 📄 SKILL.md 3.8 KB
- 📁 mlops/
  - 📄 SKILL.md 3.8 KB
- 📁 mobile/
  - 📄 SKILL.md 4.0 KB
- 📁 monorepo/
  - 📄 SKILL.md 3.6 KB
- 📁 network/
  - 📄 SKILL.md 3.9 KB
- 📁 nextjs/
  - 📄 SKILL.md 4.2 KB
- 📁 node/
  - 📄 SKILL.md 3.9 KB
- 📁 nosql/
  - 📄 SKILL.md 3.4 KB
- 📁 notify/
  - 📄 SKILL.md 4.1 KB
- 📁 npm/
  - 📄 SKILL.md 3.8 KB
- 📁 observe/
  - 📄 SKILL.md 3.9 KB
- 📁 onboard/
  - 📄 SKILL.md 3.9 KB
- 📁 opensource/
  - 📄 SKILL.md 3.8 KB
- 📁 optimize/
  - 📁 references/
    
    📄 autonomous-loop-protocol.md 7.4 KB
    
    📄 core-principles.md 7.4 KB
    
    📄 results-logging.md 5.7 KB
  - 📄 SKILL.md 5.7 KB
- 📁 orm/
  - 📄 SKILL.md 4.0 KB
- 📁 pattern/
  - 📄 SKILL.md 3.5 KB
- 📁 pay/
  - 📄 SKILL.md 4.0 KB
- 📁 pentest/
  - 📄 SKILL.md 4.4 KB
- 📁 perf/
  - 📄 SKILL.md 3.9 KB
- 📁 pipeline/
  - 📄 SKILL.md 3.7 KB
- 📁 plan/
  - 📄 SKILL.md 3.1 KB
- 📁 postgres/
  - 📄 SKILL.md 5.3 KB
- 📁 pr/
  - 📄 SKILL.md 3.0 KB
- 📁 predict/
  - 📄 SKILL.md 3.4 KB
- 📁 prompt/
  - 📄 SKILL.md 4.3 KB
- 📁 query/
  - 📄 SKILL.md 3.9 KB
- 📁 queue/
  - 📄 SKILL.md 4.5 KB
- 📁 rag/
  - 📄 SKILL.md 4.7 KB
- 📁 rails/
  - 📄 SKILL.md 4.4 KB
- 📁 ratelimit/
  - 📄 SKILL.md 4.1 KB
- 📁 rbac/
  - 📄 SKILL.md 4.0 KB
- 📁 react/
  - 📄 SKILL.md 3.9 KB
- 📁 realtime/
  - 📄 SKILL.md 4.3 KB
- 📁 redis/
  - 📄 SKILL.md 4.6 KB
- 📁 refactor/
  - 📄 SKILL.md 3.7 KB
- 📁 reliability/
  - 📄 SKILL.md 3.9 KB
- 📁 resilience/
  - 📁 references/
    
    📄 resilience-patterns.md 26.4 KB
  - 📄 SKILL.md 4.8 KB
- 📁 responsive/
  - 📄 SKILL.md 4.1 KB
- 📁 review/
  - 📄 SKILL.md 3.6 KB
- 📁 rfc/
  - 📄 SKILL.md 3.5 KB
- 📁 scale/
  - 📁 references/
    
    📄 scalability-patterns.md 33.7 KB
  - 📄 SKILL.md 5.2 KB
- 📁 scenario/
  - 📄 SKILL.md 4.7 KB
- 📁 schema/
  - 📄 SKILL.md 5.7 KB
- 📁 search/
  - 📄 SKILL.md 5.1 KB
- 📁 secrets/
  - 📄 SKILL.md 4.8 KB
- 📁 secure/
  - 📁 references/
    
    📄 security-workflow.md 9.5 KB
  - 📄 SKILL.md 4.2 KB
- 📁 seed/
  - 📄 SKILL.md 5.1 KB
- 📁 seo/
  - 📄 SKILL.md 5.0 KB
- 📁 setup/
  - 📄 SKILL.md 5.0 KB
- 📁 ship/
  - 📁 references/
    
    📄 ship-workflow.md 7.0 KB
  - 📄 SKILL.md 3.5 KB
- 📁 slo/
  - 📄 SKILL.md 5.4 KB
- 📁 spring/
  - 📄 SKILL.md 5.7 KB
- 📁 state/
  - 📄 SKILL.md 5.7 KB
- 📁 storage/
  - 📄 SKILL.md 4.5 KB
- 📁 svelte/
  - 📄 SKILL.md 5.0 KB
- 📁 tailwind/
  - 📄 SKILL.md 5.0 KB
- 📁 test/
  - 📄 SKILL.md 4.1 KB
- 📁 think/
  - 📄 SKILL.md 4.4 KB
- 📁 type/
  - 📄 SKILL.md 5.5 KB
- 📁 ui/
  - 📄 SKILL.md 5.2 KB
- 📁 upload/
  - 📄 SKILL.md 5.3 KB
- 📁 verify/
  - 📄 SKILL.md 3.8 KB
- 📁 vue/
  - 📄 SKILL.md 5.7 KB
- 📁 webhook/
  - 📄 SKILL.md 6.9 KB
- 📁 webperf/
  - 📄 SKILL.md 5.7 KB
📁 tests/
- 📄 validate-skills.sh 12.7 KB
- 📄 validate-structure.sh 10.7 KB
📄 .DS_Store 8.0 KB
📄 .gitignore 9 B
📄 AGENTS.md 20.5 KB
📄 CHANGELOG.md 3.2 KB
📄 CONTRIBUTING.md 30.7 KB
📄 GEMINI.md 10.6 KB
📄 LICENSE 1.1 KB
📄 OPENCODE.md 9.7 KB
📄 package.json 1.5 KB
📄 README.md 22.7 KB
📄 SKILL.md 8.6 KB

SKILL.md

---
name: godmode
description: "Turn on Godmode. 126 skills, 7 subagents, zero configuration. Routes to the right skill automatically."
---

@./skills/godmode/SKILL.md

---

# Universal Protocol — governs ALL 126 skills

## 1. The Loop

Every iterative skill follows one loop. No exceptions.

```
round = 0
baseline = measure(metric_cmd)

WHILE goal_not_met AND budget_not_exhausted:
    round += 1
    REVIEW   — read state: in-scope files, last 10 of results.tsv, git log -5
    IDEATE   — propose ONE change (or N parallel agents, each ONE change)
    MODIFY   — implement, then commit immediately
    VERIFY   — run guard (test_cmd && lint_cmd); run metric_cmd 3x, take median
    DECIDE   — keep or discard (rules below)
    LOG      — append to .godmode/<skill>-results.tsv

Do NOT pause to ask. Do NOT wait for confirmation. Loop until goal, budget, or stuck.
```

## 2. Keep / Discard Rules

```
KEEP    if  metric improved  AND  guard passed
DISCARD if  metric worse     OR   guard failed
DISCARD if  lines_added > 5  AND  metric_delta < 0.5%   # complexity tax
DISCARD if  complexity_cost > improvement_magnitude      # simplicity wins
KEEP    if  same metric + fewer lines                    # free simplification
```

On discard: `git reset --hard HEAD~1`. Never leave broken commits.
On keep: update baseline, continue.

## 3. Simplicity Criterion

Concrete thresholds — apply universally:

| Added lines | Required improvement |
|-------------|---------------------|
| 1-5         | any positive delta  |
| 6-20        | >= 1%               |
| 21-50       | >= 3%               |
| 51+         | >= 5%               |

Same metric + fewer lines = always keep. Readability > marginal gains.

## 4. Stopping Conditions

Stop when ANY is true:

- **target_reached** — metric hit goal
- **budget_exhausted** — max rounds/iterations consumed
- **diminishing_returns** — last 3 keeps each < 1% improvement
- **stuck** — 5+ consecutive discards (after recovery attempts)

Log `stop_reason` in session-log.tsv. Always print final summary.

## 5. Stuck Recovery (4-step escalation)

```
Step 0: DIAGNOSE — read last 3 commit diffs + test output.
        Write a 2-sentence diagnosis: what pattern the failures share,
        and what constraint they all violate.
        Example: "Last 3 attempts all added caching layers.
        The metric is I/O-bound, not CPU-bound. Switch to async I/O."
Step 1: try OPPOSITE of last approach (informed by diagnosis)
Step 2: try RADICAL rewrite of hotspot (informed by diagnosis)
Step 3: accept defeat — stop, log "stuck", report best result
```

The diagnosis turns blind retries into informed pivots.
Never skip Step 0. The 2-sentence diagnosis must reference specific code or test output.
Never repeat a failed approach. Never loop without changing strategy.

## 6. Logging

All skills log to `.godmode/`:

**Per-skill:** `.godmode/<skill>-results.tsv`
```
round	change	metric_before	metric_after	delta%	status	lines_changed
```

**Session:** `.godmode/session-log.tsv`
```
timestamp	skill	rounds	kept	discarded	final_metric	stop_reason
```

Append only. Never overwrite. Create on first write.

## 7. Execution Rules

1. Detect stack FIRST (see orchestrator). Cache `stack`, `test_cmd`, `lint_cmd`, `build_cmd`.
2. Read `skills/<skill>/SKILL.md` — follow it literally. This protocol overrides on conflict.
3. Commit BEFORE verify. Revert on failure. Zero broken commits in history.
4. Multi-agent: <=5 agents/round, worktree isolation. Merge sequentially, test after each.
   For skills that support it, parallel hypothesis mode dispatches N agents
   on different approaches to the same problem. Best wins, rest discarded.
   This is different from multi-agent task dispatch (different parts of code).
5. No worktrees? Sequential branches: `godmode-{skill}-{round}`, merge winner, delete rest.
6. Metric = shell command outputting a single number. No subjective judgment. Ever.
7. `Iterations: N` = exactly N rounds. No number = loop until stopped. Never ask to continue.
8. Chain: think -> plan -> [predict] -> build -> test -> fix -> review -> optimize -> secure -> ship.

## 8. Failure Classification

Every DISCARD must be classified. Append to `.godmode/<skill>-failures.tsv`:

```
round	change	delta%	failure_class	reason	files_touched
```

**Failure classes (use exactly one per discard):**

| Class | When to use |
|--|--|
| measurement_error | Metric command flaky or non-deterministic (stdev > delta) |
| noise | Delta within variance threshold (<0.5%) |
| regression | Change broke something unrelated |
| scope_drift | Change touched files outside assigned scope |
| complexity_tax | Improvement too small for lines added |
| infrastructure | Docker/env/dependency/tooling issue |
| already_tried | Similar approach discarded in last 10 rounds |
| overfitting | Improvement specific to one case, not generalizable |

**Before each IDEATE step, read the last 10 rows of failures.tsv:**
- If >3 failures in the same class: stop trying that approach category.
- If last 2 discards share a class: switch strategy before next attempt.
- Discarded runs still provide learning signal. Never delete failures.tsv.

## 9. Overfitting Prevention

Before every KEEP decision, apply these tests:

1. **Variance test:** Run metric_cmd 3 times. If stdev > |delta|, classify as `noise` and DISCARD.
2. **Generalization test:** "If this exact file/test disappeared, would this change still improve the project?" If NO → DISCARD as `overfitting`.
3. **No task-specific hacks:** Never add hardcoded logic for one specific test case, input, or benchmark. Fix the class of problems, not one instance.
4. **Simplification wins:** Equal metric + fewer lines = always KEEP. This is never overfitting.

## 10. Learning from Discards

Discarded runs still provide learning signal. Never waste a failure.

After every DISCARD:
1. Classify the failure (see Failure Classification above).
2. Append to `.godmode/<skill>-failures.tsv` with reason.
3. Check: has this failure class occurred 3+ times? If yes, that approach category is exhausted — switch strategy entirely.

Before every IDEATE:
1. Read last 10 rows of `<skill>-failures.tsv`.
2. Count failures per class.
3. Avoid the most common failure class. Try the LEAST common instead.
4. If all 8 classes have 2+ failures: escalate to stuck recovery.

The failures log is append-only. It persists across sessions. It is the memory of what NOT to try.

## 11. Environment Isolation

For reproducible metrics, optionally wrap metric_cmd in Docker:

```
metric_cmd_docker: "docker run --rm -v $(pwd):/app -w /app node:20 npm run benchmark"
```

When Docker is available AND metric variance > 5%:
- Build image once, cache dependencies.
- Run each metric measurement in a fresh container.
- Eliminates: cache drift, background process noise, dependency skew.

When Docker is NOT available (default):
- Run metric_cmd 3x, take median.
- Flag warning if variance > 5% between runs.
- Log variance in results.tsv as additional column.

Docker is recommended but never required. The loop works identically either way.

## 12. Session Resume

After every iteration, atomically save state to `.godmode/session-state.json`:

```json
{
  "skill": "optimize",
  "round": 7,
  "baseline": 847,
  "current_best": 198,
  "last_kept_commit": "abc1234",
  "consecutive_discards": 0,
  "approach_history": ["index", "gzip", "eager_load", "pool", "cache"],
  "failure_classes": {"noise": 2, "regression": 1},
  "stop_reason": null,
  "timestamp": "2026-04-04T12:30:00Z"
}
```

On session start:
1. Check `.godmode/session-state.json`. If exists and `stop_reason` is null:
   - Print: "Resuming from round {round}. Best so far: {current_best}."
   - Verify `last_kept_commit` matches HEAD. If not, warn and ask.
   - Continue the loop from round N+1.
2. If `stop_reason` is set: previous session completed. Start fresh.
3. If file doesn't exist: first run. Start fresh.

On session end (normal or interrupted):
- If loop completed: set `stop_reason` in state file.
- If interrupted: state file has `stop_reason: null` → next session resumes.

## 13. Lessons

Persistent learning across sessions. File: `.godmode/lessons.md`

**After each session**, append 1-3 lessons learned:
```
### Round N — {skill} — {date}
- Lesson: {concrete, reusable insight}
- Context: {what happened that taught this}
```

**Before each session**, read lessons.md:
- Apply relevant lessons to the current task.
- Never repeat a mistake that has a lesson entry.

**Format rules:**
- One lesson per bullet. Concrete and actionable.
- Bad: "Be careful with caching."
- Good: "Redis TTL must match DB write frequency. 60s TTL with 5min writes = stale reads."
- Lessons are append-only. Never delete. Mark obsolete lessons with `[OBSOLETE]`.

登录后下载/点赞/收藏 ❤ 7 | ★ 0

godmode

目录结构

SKILL.md

举报内容

提示