[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AB779SDHSV2W27SDWTCGMZ":3,"skill-same:01M7AB779SDHSV2W27SDWTCGMZ":68},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M7AB779SDHSV2W27SDWTCGMZ","auto-research",null,"Run a general Solution Autoresearch System (SAS) with an adaptive GPT-only agent graph, deterministic evaluation, a locally integrity-checked evidence log, and Open Knowledge Format (OKF) memory. Use for hard problem solving, invention loops, research campaigns, repeated optimization, hypothesis search, or any task where agents should learn across measured cycles without turning guesses into memory.","Babinci\u002Fcopernicus","Babinci","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F56366501?v=4","01M7AB6P4BSH34FQCJ8CVSF9J4","Wojciech Bieniek","01M7AB6P50AYTQF8F8JBS3QFQS",0,"MIT","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: auto-research\ndescription: Run a general Solution Autoresearch System (SAS) with an adaptive GPT-only agent graph, deterministic evaluation, a locally integrity-checked evidence log, and Open Knowledge Format (OKF) memory. Use for hard problem solving, invention loops, research campaigns, repeated optimization, hypothesis search, or any task where agents should learn across measured cycles without turning guesses into memory.\n---\n\n# Copernicus Auto-Research\n\nSAS is the invention loop. OKF is its memory layer. Build the smallest graph the actual problem needs; do not import a standing agent society.\n\nResolve `\u003Cauto-research-skill-dir>` to the directory containing this `SKILL.md`.\nResolve all bundled scripts and references from that directory, never from the\nuser's current working directory.\n\nRead [guide.md](references\u002Fguide.md) when explaining the method, designing a\nfirst research cycle, selecting an evaluator, or changing this skill. It defines\nthe concepts, purpose, outputs, examples, self-contained boundary, and honest\nlimitations.\n\n## Start\n\n1. Write the user's problem and acceptance test into a plain Markdown file.\n2. Create a run bundle:\n\n```bash\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py init RUN_DIR --problem PROBLEM.md --title \"Short title\"\n```\n\n3. Read [contracts.md](references\u002Fcontracts.md). Fill `graph.json`, set its status to `ready`, and run:\n\n```bash\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py verify RUN_DIR\n```\n\n## Invent the graph\n\nFreeze the problem regime before choosing agents: objective, inputs, evaluator, constraints, budget, freshness boundary, and stop conditions. Then create only roles that change a named artifact or decision.\n\nEvery node must declare:\n\n- bounded input and typed output;\n- lane: `deterministic`, `sol`, `terra`, `luna`, or `human`;\n- dependencies;\n- authority ceiling;\n- independent verifier;\n- falsifier;\n- terminal status.\n\nPrefer deterministic controller\u002Fevaluator nodes. Use models as bounded mappers, proposers, critics, implementers, or curators. A role need not be a separate agent. Delete any role that does not change a named artifact or decision.\n\n## Run one cycle\n\n1. Snapshot the regime and append a `regime` receipt.\n2. Select applicable OKF cards deterministically and record their IDs plus file hashes. Baselines record an explicit empty-card receipt.\n3. Execute ready graph nodes through the sibling `$fleet` skill. Run independent nodes in parallel and dependency edges in order. If Fleet is unavailable, execute the same nodes sequentially; the evidence contract matters more than parallelism.\n4. Make each candidate content-addressed. Bind it to the regime, parent candidate, changed hypothesis, prompt\u002Fstrategy identity, selected card hashes, and artifact hash.\n5. Run the cheapest trusted filter first, then the authoritative evaluator, then an independent semantic or normative review when correctness is broader than a test result.\n6. Retry only when the failure signature and hypothesis changed. Never retry identical candidate bytes or prompts.\n7. Join candidate, evaluation, review, and cost evidence into one `cycle_join` receipt.\n8. Only at the cycle boundary, promote, reject, quarantine, or retire an OKF card. A hypothesis never promotes itself.\n\nAppend each receipt from a JSON object:\n\n```bash\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py append RUN_DIR\u002Fevidence.jsonl receipt.json\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py verify RUN_DIR\n```\n\n## OKF memory\n\nKeep one concept per Markdown file with YAML frontmatter. Start with only three card families:\n\n- `selection`: what is worth attempting;\n- `strategy`: what change is likely to help;\n- `risk`: what can invalidate or mislead the evaluation.\n\nCards require evidence, regime scope, a falsifier, and a disposition. New domains may rename these concepts, but do not add a fourth family until a measured workflow cannot fit one of the three.\n\nUse relative Markdown links and keep `knowledge\u002Findex.md` as progressive navigation. Update `knowledge\u002Flog.md` only for material curation changes.\n\n## Stop rules\n\nStop and report rather than guessing when the evaluator is missing, the regime changed, evidence cannot be joined, a provider\u002Fmodel is unavailable, the same failure repeats, or the budget is exhausted. Stop for user authority before purchases, publishing, deployment, messaging, signing, secrets, or destructive\u002Fexternal mutations.\n\nDo not add a graph database, vector store, agent genome, role league, custom queue, dashboard, daemon, or second state store until a measured run proves the flat graph, files, and JSONL ledger insufficient.\n\nAuto-Research is self-contained: it uses Markdown, JSON, the bundled Python\nstandard-library receipt tool, and optional sibling Fleet execution. It does not\ndepend on a private repository, hosted database, external memory service, or\nundisclosed evaluator. The user must supply the problem and the real evaluator.\n","# Copernicus Auto-Research\n\nSAS is the invention loop. OKF is its memory layer. Build the smallest graph the actual problem needs; do not import a standing agent society.\n\nResolve `\u003Cauto-research-skill-dir>` to the directory containing this `SKILL.md`.\nResolve all bundled scripts and references from that directory, never from the\nuser's current working directory.\n\nRead [guide.md](references\u002Fguide.md) when explaining the method, designing a\nfirst research cycle, selecting an evaluator, or changing this skill. It defines\nthe concepts, purpose, outputs, examples, self-contained boundary, and honest\nlimitations.\n\n## Start\n\n1. Write the user's problem and acceptance test into a plain Markdown file.\n2. Create a run bundle:\n\n```bash\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py init RUN_DIR --problem PROBLEM.md --title \"Short title\"\n```\n\n3. Read [contracts.md](references\u002Fcontracts.md). Fill `graph.json`, set its status to `ready`, and run:\n\n```bash\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py verify RUN_DIR\n```\n\n## Invent the graph\n\nFreeze the problem regime before choosing agents: objective, inputs, evaluator, constraints, budget, freshness boundary, and stop conditions. Then create only roles that change a named artifact or decision.\n\nEvery node must declare:\n\n- bounded input and typed output;\n- lane: `deterministic`, `sol`, `terra`, `luna`, or `human`;\n- dependencies;\n- authority ceiling;\n- independent verifier;\n- falsifier;\n- terminal status.\n\nPrefer deterministic controller\u002Fevaluator nodes. Use models as bounded mappers, proposers, critics, implementers, or curators. A role need not be a separate agent. Delete any role that does not change a named artifact or decision.\n\n## Run one cycle\n\n1. Snapshot the regime and append a `regime` receipt.\n2. Select applicable OKF cards deterministically and record their IDs plus file hashes. Baselines record an explicit empty-card receipt.\n3. Execute ready graph nodes through the sibling `$fleet` skill. Run independent nodes in parallel and dependency edges in order. If Fleet is unavailable, execute the same nodes sequentially; the evidence contract matters more than parallelism.\n4. Make each candidate content-addressed. Bind it to the regime, parent candidate, changed hypothesis, prompt\u002Fstrategy identity, selected card hashes, and artifact hash.\n5. Run the cheapest trusted filter first, then the authoritative evaluator, then an independent semantic or normative review when correctness is broader than a test result.\n6. Retry only when the failure signature and hypothesis changed. Never retry identical candidate bytes or prompts.\n7. Join candidate, evaluation, review, and cost evidence into one `cycle_join` receipt.\n8. Only at the cycle boundary, promote, reject, quarantine, or retire an OKF card. A hypothesis never promotes itself.\n\nAppend each receipt from a JSON object:\n\n```bash\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py append RUN_DIR\u002Fevidence.jsonl receipt.json\npython3 \u003Cauto-research-skill-dir>\u002Fscripts\u002Freceipts.py verify RUN_DIR\n```\n\n## OKF memory\n\nKeep one concept per Markdown file with YAML frontmatter. Start with only three card families:\n\n- `selection`: what is worth attempting;\n- `strategy`: what change is likely to help;\n- `risk`: what can invalidate or mislead the evaluation.\n\nCards require evidence, regime scope, a falsifier, and a disposition. New domains may rename these concepts, but do not add a fourth family until a measured workflow cannot fit one of the three.\n\nUse relative Markdown links and keep `knowledge\u002Findex.md` as progressive navigation. Update `knowledge\u002Flog.md` only for material curation changes.\n\n## Stop rules\n\nStop and report rather than guessing when the evaluator is missing, the regime changed, evidence cannot be joined, a provider\u002Fmodel is unavailable, the same failure repeats, or the budget is exhausted. Stop for user authority before purchases, publishing, deployment, messaging, signing, secrets, or destructive\u002Fexternal mutations.\n\nDo not add a graph database, vector store, agent genome, role league, custom queue, dashboard, daemon, or second state store until a measured run proves the flat graph, files, and JSONL ledger insufficient.\n\nAuto-Research is self-contained: it uses Markdown, JSON, the bundled Python\nstandard-library receipt tool, and optional sibling Fleet execution. It does not\ndepend on a private repository, hosted database, external memory service, or\nundisclosed evaluator. The user must supply the problem and the real evaluator.\n",{"name":5,"description":7},"main","caa1cc6207632cbd529bd320d2e3818f3078c276","plugins\u002Fcopernicus\u002Fskills\u002Fauto-research","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus","https:\u002F\u002Fgithub.com\u002FBabinci","2026-09-30 00:49:12",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add Babinci\u002Fcopernicus --skill auto-research",{"tool":36,"command":37},"bunx","bunx skills add Babinci\u002Fcopernicus --skill auto-research",{"tool":39,"command":40},"pnpm","pnpm dlx skills add Babinci\u002Fcopernicus --skill auto-research","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002FSKILL.md",true,[44,50,54,58,64],{"path":45,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":48,"sourceHtmlUrl":49},"agents\u002Fopenai.yaml","REFERENCE","参考文档",242,"https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002Fagents\u002Fopenai.yaml",{"path":51,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Fcontracts.md",2648,"https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002Freferences\u002Fcontracts.md",{"path":55,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":56,"sourceHtmlUrl":57},"references\u002Fguide.md",6648,"https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002Freferences\u002Fguide.md",{"path":59,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":62,"sourceHtmlUrl":63},"scripts\u002Freceipts.py","SCRIPT","脚本",14832,"https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002Fscripts\u002Freceipts.py",{"path":65,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":66,"sourceHtmlUrl":67},"scripts\u002Ftest_receipts.py",4618,"https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fplugins\u002Fcopernicus\u002Fskills\u002Fauto-research\u002Fscripts\u002Ftest_receipts.py",[69,76,83,90,97,104],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M7AB9C5QZW0WQKQGT36KMRT7","upstream-dmg-watchdog","Run the versioned upstream DMG probe, repair worker, guarded PR, and sequential Nix refresh flow.","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Fscripts\u002Fautomation\u002Fupstream-dmg-watchdog\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M7AB9AW84M3TKGRCM7EX4ZN3","Unsafe Symlink","Use when testing symlink safety.","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Frecord-replay-linux\u002Ftests\u002Ffixtures\u002Fskills\u002Funsafe-symlink\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M7AB99PC174BDQGPWKW0F6DZ","Mac Only","Use when automating Finder with AppleScript.","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Frecord-replay-linux\u002Ftests\u002Ffixtures\u002Fskills\u002Fplatform-macos\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M7AB985WEE2Q5ZA0MG21V96T","Instruction Only","Use when answering with reusable written instructions.","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Frecord-replay-linux\u002Ftests\u002Ffixtures\u002Fskills\u002Finstruction-only\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":19},"01M7AB95F9KS3TQFG183BB95PB","Desktop Act","Use when clicking and typing in a Linux desktop app.","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Frecord-replay-linux\u002Ftests\u002Ffixtures\u002Fskills\u002Fdesktop-act\u002FSKILL.md",[],[],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":19},"01M7AB93YB9J8BGZ9M28KB9CPN","Collision Tripwire","Use when testing script handling.","https:\u002F\u002Fgithub.com\u002FBabinci\u002Fcopernicus\u002Fblob\u002Fmain\u002Frecord-replay-linux\u002Ftests\u002Ffixtures\u002Fskills\u002Fcollision-tripwire\u002FSKILL.md",[],[]]