benchmark
Runs, supervises, changes, and publishes ttsc's performance, graph, and evidence benchmarks under benchmarks/. Use before running, changing, or publishing any benchmark or its fixtures.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 21
- Forks
- 2
01 前言 Front Matter
- name
- benchmark
- description
- Runs, supervises, changes, and publishes ttsc's performance, graph, and evidence benchmarks under benchmarks/. Use before running, changing, or publishing any benchmark or its fixtures.
02 资源文件 8
- 参考文档evidence/aggregate.md1.3 KB
- 参考文档evidence/boundary.md1.2 KB
- 参考文档evidence/campaign.md2.1 KB
- 参考文档evidence/integrity.md2.1 KB
- 参考文档evidence/recovery.md1.9 KB
- 参考文档evidence/review.md1.2 KB
- 参考文档graph.md1.3 KB
- 参考文档performance.md926 B
03 SKILL.md 正文
Benchmark
| Work | Read |
|---|---|
| Performance harness and its fixture branches | performance.md |
| Graph harness, prompts, comparators, traces, and measuring a graph change | graph.md |
| Evidence campaign: launch, supervision, dashboard | evidence/campaign.md |
| Evidence: whether a cell's edit is allowed, and warning a cell | evidence/integrity.md |
| Evidence: a Plain cell's Review boundary | evidence/review.md |
| Evidence: fixing a repository defect mid-campaign | evidence/boundary.md |
| Evidence: a stopped, hung, or failed cell, or cancellation | evidence/recovery.md |
| Evidence: publishing results and closing the cohort | evidence/aggregate.md |
The rules below cover the performance and graph harnesses; the evidence benchmark has its own documents.
- Measure the real product on the full workload, with every comparator set up as its own documentation prescribes. Never restrict an agent or skip input to improve a number.
- Inspect the raw report or trace behind any surprising result before accepting or explaining it.
- A CLI entry in
src/executableonly imports oneTtscBenchmark*owner and callsmain(), within 12 lines. Every other module exports oneTtscBenchmark*class or namespace, or oneITtscBenchmark*interface with its companion namespace, from a file of that name, and never a standalone function, constant, enum, or type alias. Document each member's role and each field's unit and default. Run the README's strict type check before committing. - Give each run its own temporary root, including
GOCACHEandGOTMPDIR, and delete it afterwards unless--keep-...or the user says otherwise. Never clean a shared Go cache or temp directory. - Edit fixtures in their own repositories, and push a fixture branch only when its build, format, and lint pass and it holds no generated benchmark input.
- Update the sticky result comment only with pull-request authority; otherwise put the table in the report.
04 同仓库更多 Skills
contracts
Defines self-acknowledgments for production declarations and tests. Use when implementing or reviewing maintained source, unit tests or E2E tests, or selecting Evidence checklists.
prompting
Writes and trims ttsc's prompts for agents, namely AGENTS.md, CLAUDE.md, and the skills under .agents/skills. Use before changing any of them, together with the documentation skill's writing rules.
contracts
Defines the @evidence checklists every maintained ttsc declaration and test answers, and their evidence.config.json wiring. Use when writing or reviewing maintained source or tests, or changing Evidence selection.
alchemy-domain
Build backends with @alchemy.run/domain (domain-driven design on Effect) - turn requirements into aggregates, commands, events, rejections, views, policies, sagas, projections, streams, queries, external systems and Apis; lay them out in a monorepo (domain / server / web packages, kind-suffixed files); test them with stories; host them on Durable Objects (Cloudflare, Rivet or Effect Cluster) with alchemy; and drive them from a foldkit frontend. Use when designing or changing an @alchemy.run/domain model, adding a building block, wiring hosting or a frontend, or reviewing such code.
alchemy-provider
暂无简介
write-docs
暂无简介