[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-1043":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":28,"readmeContent":29,"aiSummary":30,"trendingCount":15,"starSnapshotCount":15,"syncStatus":31,"lastSyncTime":32,"discoverSource":33},1043,"codex-autoresearch","TheGreenCedar\u002Fcodex-autoresearch","TheGreenCedar","A codex plugin for running optimization loops inside a codebase. It is useful when you have a measurable target and many possible changes to try: test runtime, build speed, bundle size, model loss, Lighthouse scores, memory use, query latency, or any other metric you can print from a script.","",null,"TypeScript",615,50,1,0,35,69,116,105,9.12,"Apache License 2.0",false,"main",true,[26,27],"autoresearcher","codex","2026-06-12 02:00:22","\u003Cdiv align=\"center\">\n\n# Codex Autoresearch\n\n### Measured improvement loops for Codex\n\n**[Try it](#try-it)** - **[Install](#install)** - **[How it works](#how-it-works)** - **[Dashboard](#dashboard)** - **[Docs](#docs)** - **[Changelog](#changelog)**\n\u003C\u002Fdiv>\n\nCodex Autoresearch helps Codex turn \"make this better\" into a measured loop.\n\nGive Codex a goal, a benchmark contract, and a safe edit scope. Codex can run small experiment packets, keep or discard changes with evidence, preserve ASI and metrics across context loss, and package useful work for review.\n\n![Codex Autoresearch live dashboard showing a demo runtime improvement](plugins\u002Fcodex-autoresearch\u002Fassets\u002Fshowcase\u002Fdashboard-demo.png)\n\nInspired by the AI-focused [karpathy\u002Fautoresearch](https:\u002F\u002Fgithub.com\u002Fkarpathy\u002Fautoresearch) and [pi-autoresearch](https:\u002F\u002Fgithub.com\u002Fdavebcn87\u002Fpi-autoresearch). Codex Autoresearch adapts the measured-loop idea for Codex plugin workflows, repo-local benchmarks, durable session files, an evidence trail, live dashboards, and reviewable finalization.\n\n## Try it\n\nAsk Codex to use Codex Autoresearch.\n\nBroad prompts work:\n\n```text\nUse $Codex Autoresearch to improve the speed of my indexer's pipeline, while keeping it memory efficient.\n````\n\n```text\nUse $Codex Autoresearch to keep reducing bugs in the codebase, starting with\nthe most obvious low hanging fruits. Keep doing this 100 times.\n```\n\nYou can also hand it a sharper investigation:\n\n```text\nUse $Codex Autoresearch to figure out why my graphql service's p99 latency is so much higher\nthan its p90 latency at 1 minute metric resolution. I suspect: DNS lookup, event loop throttling,\nmemory spike, CPU spike. For each, run the 4-5 appropriate experiments @experiments.md and if the\nresults are promising keep iterating, otherwise stop and report back.\n```\n\nOr be exact about the benchmark and scope:\n\n```text\nUse $Codex Autoresearch to optimize my unit tests' speed. different libraries are allowed, but try to avoid it.\nBenchmark: npm test -- --runInBand\nMetric: seconds, lower is better\nChecks: npm test\nScope: test runner config and test helpers only\n```\n\nCodex should start by checking Git state, identifying the target package, creating or resuming the session, verifying the benchmark, starting the dashboard, running one packet, and logging the result with experiment details.\n\n## Install\n\nThis repository is a Codex plugin marketplace. Add the marketplace:\n\n```bash\ncodex plugin marketplace add TheGreenCedar\u002Fcodex-autoresearch\n```\n\nThen open Codex in the repo you want to improve:\n\n```text\n\u002Fplugins\n```\n\nChoose:\n\n```text\nTheGreenCedar Autoresearch -> codex-autoresearch -> Install plugin\n```\n\nStart a new Codex thread after installation.\n\n## How it works\n\nA normal session follows this shape:\n\n```text\nTarget -> Onboard -> Setup -> Doctor -> Dashboard -> Packet -> Log -> Continue or Finalize\n```\n\nCodex Autoresearch helps Codex:\n\n1. identify the target repo or child package\n2. check for an existing session\n3. return a shared next-step contract with stage, reason, CLI command, safety, and missing essentials\n4. verify the benchmark contract\n5. run a measured packet with command identity, output tails, metrics, artifacts, checks, and a freshness fingerprint\n6. log the result as `keep`, `discard`, `crash`, or `checks_failed`\n7. preserve ASI, packet fingerprints, promotion labels, and metrics in durable files\n8. continue safely or preview finalization into reviewable branches\n\nA packet is one measured experiment cycle: make a scoped change, run the benchmark, inspect the metric, and log the decision.\n\nASI means Accumulated Structured Intelligence. It is the structured memory attached to each packet decision: hypothesis, evidence, rollback reason, next action hint, and optional lane, family, or risk metadata. It tells the next Codex session what happened, what was learned, and which path deserves the next attempt.\n\n## When to use it\n\nUse Codex Autoresearch when:\n\n* the goal can be measured\n* the benchmark is repeatable\n* correctness checks exist or can be added\n* the editable scope is small enough to review\n* kept work should become reviewable commits or branches\n\nUse a regular Codex task when:\n\n* the work needs one careful edit\n* the goal is mainly taste or judgment\n* the benchmark is flaky or very expensive\n* the metric can improve by weakening the benchmark\n* secrets, deployment paths, or unrelated dirty files are in scope\n\n## Dashboard\n\nAsk codex to boot up the dashboard if it hasn't already.\n\nThe dashboard shows:\n\n* baseline, latest, best, confidence, and weighted metric formulas\n* Codex brief and session memory\n* next safe action, evidence label, proof gaps, and why the action is safe\n* ledger entries, ASI, and handoff context\n* best kept change and recent failures\n* strategy lanes, scaffold health, research integrity, runtime drift, and finalization readiness\n* copyable status reports and agent handoff packets\n\nUse the dashboard to inspect state. Talk to Codex for everything else.\n\n## Quality-gap loops\n\nFor product, docs, UX, or broad research, ask for a quality-gap loop:\n\n```text\nUse Codex Autoresearch to study this project and improve the dashboard.\nTurn accepted findings into a quality-gap loop, implement them, and keep the live dashboard open.\n```\n\n`quality_gap=0` means the accepted checklist for that round is closed. It does not mean discovery is complete. Start another round if the question is still alive.\n\n## Finalization\n\nAsk the plugin to finalize once a loop has useful kept work mixed with exploratory history.\n\nFinalization should:\n\n1. select kept evidence\n2. exclude session artifacts from review branches unless requested\n3. block later-discarded, invalidated, or reverted keeps\n4. show dirty-tree, overlap, semantic-safety, and final-tree coverage warnings\n5. prepare clean review branches or a current-final-tree plan\n6. preserve metric evidence and verification commands\n7. leave cleanup until review branches are verified\n\n## Docs\n\n* [Docs index](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Findex.md)\n* [Concepts glossary](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Fconcepts.md)\n* [Start](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Fstart.md)\n* [Workflow diagrams](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Fworkflows.md)\n* [Architecture diagrams](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Farchitecture.md)\n* [Operate](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Foperate.md)\n* [Trust](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Ftrust.md)\n* [Finish](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Ffinish.md)\n* [Recipes](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Frecipes.md)\n* [Troubleshooting](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Ftroubleshooting.md)\n* [Hooks](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Fhooks.md)\n* [Maintainers](plugins\u002Fcodex-autoresearch\u002Fdocs\u002Fmaintainers.md)\n\nThe active package lives under:\n\n```text\nplugins\u002Fcodex-autoresearch\n```\n\nThe plugin skill lives at:\n\n```text\nplugins\u002Fcodex-autoresearch\u002Fskills\u002Fcodex-autoresearch\u002FSKILL.md\n```\n\n## Development\n\nThe plugin and dashboard source are written in TypeScript.\n\nThe package uses `tsdown` for Node builds, `tsgo` for typechecking, `oxlint` for linting, `oxfmt` for formatting, Vite for the dashboard, and `npm-run-all2` for combined gates.\n\nFrom `plugins\u002Fcodex-autoresearch`:\n\n```bash\nnpm install\nnpm run check\nnpm test\nnode scripts\u002Fautoresearch.mjs --help\n```\n\nTargeted checks:\n\n```bash\nnpm run typecheck\nnpm run lint\nnpm run format:check\nnode scripts\u002Fautoresearch.mjs doctor --cwd . --check-benchmark --explain\ngit diff --check\n```\n\n## Update or remove\n\nRefresh the marketplace:\n\n```bash\ncodex plugin marketplace upgrade thegreencedar-autoresearch\n```\n\nRemove the marketplace:\n\n```bash\ncodex plugin marketplace remove thegreencedar-autoresearch\n```\n\nTo uninstall the plugin, open Codex:\n\n```text\n\u002Fplugins\n```\n\nThen choose:\n\n```text\ncodex-autoresearch -> Uninstall plugin\n```\n\n## Changelog\n\nUser-facing changes are tracked in [CHANGELOG.md](CHANGELOG.md).\n\n## License\n\nThis project is licensed under the terms of the [Apache License 2.0](.\u002FLICENSE). Copyright (c) 2026 Albert Najjar.\n","Codex Autoresearch 是一个用于在代码库中运行优化循环的 Codex 插件。它允许用户设定可量化的目标，并尝试多种可能的更改，如测试运行时间、构建速度、包大小、模型损失、Lighthouse 评分、内存使用、查询延迟等指标。该项目通过提供自动化的实验包执行、基于证据的变化保留或丢弃机制以及跨上下文丢失时的ASI和度量保存功能，帮助开发者实现持续改进。此外，它还支持实时仪表盘监控和最终审查前的工作打包。适用于需要对软件性能进行精细化调优的各种开发场景。",2,"2026-06-11 02:41:18","CREATED_QUERY"]