[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7DM69EC5DX659EV06Q075GZ":3,"skill-same:01M7DM69EC5DX659EV06Q075GZ":56},{"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":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M7DM69EC5DX659EV06Q075GZ","reference-mining",null,"Extracts a licence-checked port map from one pinned reference submodule under references\u002F (autogpt, openhands, dify, crewai, deepseek_harness, revfactory_harness, openharness, meta_harness, openrig, harness_100, hive, cowagent, repository_harness, paperclip): which path:line holds a runtime pattern (agent loop, compaction, token meter, path containment, sandbox policy, DAG\u002Fcheckpoint, delegation, SSE streaming, MCP client, eval runner) or a harness-factory pattern (agent\u002Fskill conventions, coordinator prompts, verification rules, permission checks, judge panels, hooks, memory ledgers), and whether it may be ported. Use for: mine references, build a port map, re-mine dify, where does deepseek do compaction, what can we port from crewai, what does openharness do for verification, update the port map. Not for reading arbitrary third-party repos outside references\u002F or for general library docs questions.","danielnguyenfinhub\u002Ffinhub-harness","danielnguyenfinhub","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F264356854?v=4","01M7DM55XETDR9HF92FWC28SBN","Daniel Nguyen","01M7DM55YGJYBWW488JZSQYEJ4",1,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Freference-mining\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: reference-mining\ndescription: \"Extracts a licence-checked port map from one pinned reference submodule under references\u002F (autogpt, openhands, dify, crewai, deepseek_harness, revfactory_harness, openharness, meta_harness, openrig, harness_100, hive, cowagent, repository_harness, paperclip): which path:line holds a runtime pattern (agent loop, compaction, token meter, path containment, sandbox policy, DAG\u002Fcheckpoint, delegation, SSE streaming, MCP client, eval runner) or a harness-factory pattern (agent\u002Fskill conventions, coordinator prompts, verification rules, permission checks, judge panels, hooks, memory ledgers), and whether it may be ported. Use for: mine references, build a port map, re-mine dify, where does deepseek do compaction, what can we port from crewai, what does openharness do for verification, update the port map. Not for reading arbitrary third-party repos outside references\u002F or for general library docs questions.\"\n---\n\n# Reference Mining\n\nProduce one port map per submodule so the scout can rank, the architect can cite exact lines and the judge can re-open them. The port map is the only bridge between fourteen large repos and a small design — a wrong line number here becomes a REJECTED claim two phases later, so precision beats coverage.\n\n## Steps\n\n1. **Confirm the submodule is present.** List `references\u002F\u003Csubmodule>\u002F`. If empty or only `.gitkeep`, stop and return a port map with an empty Findings table and a Gaps line naming the path. Do not guess contents from memory of the upstream project.\n2. **Read the licence rules** in `references\u002Flicence-rules.md` and note the constraint row for this submodule before you look at code. Licence decides what the \"Port as\" column may say.\n3. **Locate targets.** Search for the patterns the slice order needs (see table below). Use Grep\u002FGlob to find candidates, then open the specific region. Read the actual lines; record the first line of the function\u002Fclass.\n4. **Record findings** in the schema from `references\u002Fport-map-schema.md`. One row per pattern unit (a function or class), not per file.\n5. **Record exclusions and gaps.** Anything licence-blocked goes under **Do not port**. Any target pattern that this submodule does not contain goes under **Gaps** with what you searched for.\n6. **Return the map** as your final message (you run as Explore — read-only; the orchestrator saves it).\n\n## Target patterns by slice\n\n| Slice | Pattern | Likely source |\n|---|---|---|\n| 1 | agent loop: preStep → turn → step; stop when no tool calls | `deepseek_harness\u002Fpackages\u002Fcore\u002Fagent-loop\u002F` |\n| 2 | token estimate (len\u002F\u002F4), threshold compaction, tool-result pruning | `deepseek_harness\u002Fpackages\u002Fllm\u002Ftoken-meter\u002F`, `packages\u002Fcompaction\u002Fcompaction-basic\u002F` |\n| 3 | path containment (`isPathUnder`), sandbox policy modes, upload path checks | `deepseek_harness\u002Fpackages\u002Ffs\u002Ffs-sandbox\u002F`, `packages\u002Fsandbox\u002Fsandbox-policy\u002F`, `openhands\u002Fsrc\u002Fapi\u002Fworkspace-upload-path.ts` |\n| 4 | provider\u002Fmodel routing by profile | `deepseek_harness\u002Fpackages\u002Fllm\u002F`, `crewai` `llm.py` |\n| 5–6 | checkpoint per step, limits\u002Fobservability layers around an executor | `dify\u002Fapi\u002Fcore\u002Fworkflow\u002F` (pattern only) |\n| 7 | delegation \u002F ask-question tools, task context passing | `crewai` `tools\u002Fagent_tools\u002F` |\n| 8 | container ingress\u002Fentrypoint pattern | `openhands\u002Fdocker\u002F` |\n| 9 | MCP client shape | `dify\u002Fapi\u002Fcore\u002Fmcp\u002Fmcp_client.py` (pattern only) |\n| 10 | SSE ping\u002Fidle streaming | `dify\u002Fapi\u002Fcore\u002Fapp\u002Fapps\u002Fstreaming_utils.py` (pattern only) |\n| 11 | benchmark harness\u002Frunner\u002Fevaluator | `autogpt\u002Fclassic\u002Fdirect_benchmark\u002F` (MIT only) |\n| meta | harness\u002Fskill\u002Fagent conventions | `revfactory_harness\u002F` |\n\n## Target patterns for the harness factory\n\nMined when the orchestrator's scope line says `factory` or `both` (the default). These feed the plugin under `skills\u002F` and the team under `.claude\u002F`, not the runtime.\n\n| Area | Pattern | Likely source |\n|---|---|---|\n| coordinator | orchestrator\u002Fcoordinator prompts: role split, hand-off, standing reminders | `openharness\u002Fsrc\u002Fopenharness\u002Fcoordinator\u002F`, `revfactory_harness\u002F` |\n| verification | QA and reviewer rules: probe before pass, rationalisation lists, before-fail guards | `openharness\u002Fsrc\u002Fopenharness\u002Fcoordinator\u002Fagent_definitions.py` |\n| permissions | permission modes, denylists, path and command rules evaluated before allow | `openharness\u002Fsrc\u002Fopenharness\u002Fpermissions\u002F` |\n| memory | compaction state ledgers, session memory, what survives a context reset | `openharness\u002F`, `deepseek_harness\u002Fpackages\u002Fcompaction\u002F` |\n| hooks | pre\u002Fpost tool hooks, plugin loaders | `openharness\u002Fsrc\u002Fopenharness\u002Fhooks\u002F`, `openharness\u002Fplugins\u002F` |\n| evaluation | judge panels, multi-reviewer consensus, benchmark scoring | `autogpt\u002Fclassic\u002Fdirect_benchmark\u002F`, `crewai` |\n| retrospective | learning loops that feed run results back into prompts or skills | `revfactory_harness\u002F` (evolve), `openharness\u002F` |\n\nPaths are starting points; open each before citing it. Record a factory-pattern row with the same schema as a runtime row; the `Slice` column takes the area name instead of a number.\n\nPaths above are starting points from the plan's exploration, not facts — verify each one exists before citing it.\n\n## Rules that prevent downstream rejections\n\n- Cite `references\u002F\u003Csubmodule>\u002F\u003Cpath>:\u003Cline>` exactly as it exists on disk, forward slashes, line = first line of the unit.\n- Never list `references\u002Fautogpt\u002Fautogpt_platform\u002F**` as a source. It is Polyform Shield.\n- For `dify`, the \"Port as\" value is always `pattern` — modified Apache-2.0 forbids treating its files as copy sources for a SaaS-capable runtime.\n- Describe what the code does in your own words in ≤20 words; do not paste code blocks longer than 5 lines.\n- Mark language: the runtime is Python; TypeScript sources are ports of the idea, not the syntax.\n\n## On re-run\n\nIf the orchestrator passes a prior port map, read it first, re-verify only the rows or gaps it flags, and return the full map with only those parts changed. Re-verifying everything wastes the budget and can introduce churn the architect then has to chase.\n","# Reference Mining\n\nProduce one port map per submodule so the scout can rank, the architect can cite exact lines and the judge can re-open them. The port map is the only bridge between fourteen large repos and a small design — a wrong line number here becomes a REJECTED claim two phases later, so precision beats coverage.\n\n## Steps\n\n1. **Confirm the submodule is present.** List `references\u002F\u003Csubmodule>\u002F`. If empty or only `.gitkeep`, stop and return a port map with an empty Findings table and a Gaps line naming the path. Do not guess contents from memory of the upstream project.\n2. **Read the licence rules** in `references\u002Flicence-rules.md` and note the constraint row for this submodule before you look at code. Licence decides what the \"Port as\" column may say.\n3. **Locate targets.** Search for the patterns the slice order needs (see table below). Use Grep\u002FGlob to find candidates, then open the specific region. Read the actual lines; record the first line of the function\u002Fclass.\n4. **Record findings** in the schema from `references\u002Fport-map-schema.md`. One row per pattern unit (a function or class), not per file.\n5. **Record exclusions and gaps.** Anything licence-blocked goes under **Do not port**. Any target pattern that this submodule does not contain goes under **Gaps** with what you searched for.\n6. **Return the map** as your final message (you run as Explore — read-only; the orchestrator saves it).\n\n## Target patterns by slice\n\n| Slice | Pattern | Likely source |\n|---|---|---|\n| 1 | agent loop: preStep → turn → step; stop when no tool calls | `deepseek_harness\u002Fpackages\u002Fcore\u002Fagent-loop\u002F` |\n| 2 | token estimate (len\u002F\u002F4), threshold compaction, tool-result pruning | `deepseek_harness\u002Fpackages\u002Fllm\u002Ftoken-meter\u002F`, `packages\u002Fcompaction\u002Fcompaction-basic\u002F` |\n| 3 | path containment (`isPathUnder`), sandbox policy modes, upload path checks | `deepseek_harness\u002Fpackages\u002Ffs\u002Ffs-sandbox\u002F`, `packages\u002Fsandbox\u002Fsandbox-policy\u002F`, `openhands\u002Fsrc\u002Fapi\u002Fworkspace-upload-path.ts` |\n| 4 | provider\u002Fmodel routing by profile | `deepseek_harness\u002Fpackages\u002Fllm\u002F`, `crewai` `llm.py` |\n| 5–6 | checkpoint per step, limits\u002Fobservability layers around an executor | `dify\u002Fapi\u002Fcore\u002Fworkflow\u002F` (pattern only) |\n| 7 | delegation \u002F ask-question tools, task context passing | `crewai` `tools\u002Fagent_tools\u002F` |\n| 8 | container ingress\u002Fentrypoint pattern | `openhands\u002Fdocker\u002F` |\n| 9 | MCP client shape | `dify\u002Fapi\u002Fcore\u002Fmcp\u002Fmcp_client.py` (pattern only) |\n| 10 | SSE ping\u002Fidle streaming | `dify\u002Fapi\u002Fcore\u002Fapp\u002Fapps\u002Fstreaming_utils.py` (pattern only) |\n| 11 | benchmark harness\u002Frunner\u002Fevaluator | `autogpt\u002Fclassic\u002Fdirect_benchmark\u002F` (MIT only) |\n| meta | harness\u002Fskill\u002Fagent conventions | `revfactory_harness\u002F` |\n\n## Target patterns for the harness factory\n\nMined when the orchestrator's scope line says `factory` or `both` (the default). These feed the plugin under `skills\u002F` and the team under `.claude\u002F`, not the runtime.\n\n| Area | Pattern | Likely source |\n|---|---|---|\n| coordinator | orchestrator\u002Fcoordinator prompts: role split, hand-off, standing reminders | `openharness\u002Fsrc\u002Fopenharness\u002Fcoordinator\u002F`, `revfactory_harness\u002F` |\n| verification | QA and reviewer rules: probe before pass, rationalisation lists, before-fail guards | `openharness\u002Fsrc\u002Fopenharness\u002Fcoordinator\u002Fagent_definitions.py` |\n| permissions | permission modes, denylists, path and command rules evaluated before allow | `openharness\u002Fsrc\u002Fopenharness\u002Fpermissions\u002F` |\n| memory | compaction state ledgers, session memory, what survives a context reset | `openharness\u002F`, `deepseek_harness\u002Fpackages\u002Fcompaction\u002F` |\n| hooks | pre\u002Fpost tool hooks, plugin loaders | `openharness\u002Fsrc\u002Fopenharness\u002Fhooks\u002F`, `openharness\u002Fplugins\u002F` |\n| evaluation | judge panels, multi-reviewer consensus, benchmark scoring | `autogpt\u002Fclassic\u002Fdirect_benchmark\u002F`, `crewai` |\n| retrospective | learning loops that feed run results back into prompts or skills | `revfactory_harness\u002F` (evolve), `openharness\u002F` |\n\nPaths are starting points; open each before citing it. Record a factory-pattern row with the same schema as a runtime row; the `Slice` column takes the area name instead of a number.\n\nPaths above are starting points from the plan's exploration, not facts — verify each one exists before citing it.\n\n## Rules that prevent downstream rejections\n\n- Cite `references\u002F\u003Csubmodule>\u002F\u003Cpath>:\u003Cline>` exactly as it exists on disk, forward slashes, line = first line of the unit.\n- Never list `references\u002Fautogpt\u002Fautogpt_platform\u002F**` as a source. It is Polyform Shield.\n- For `dify`, the \"Port as\" value is always `pattern` — modified Apache-2.0 forbids treating its files as copy sources for a SaaS-capable runtime.\n- Describe what the code does in your own words in ≤20 words; do not paste code blocks longer than 5 lines.\n- Mark language: the runtime is Python; TypeScript sources are ports of the idea, not the syntax.\n\n## On re-run\n\nIf the orchestrator passes a prior port map, read it first, re-verify only the rows or gaps it flags, and return the full map with only those parts changed. Re-verifying everything wastes the budget and can introduce churn the architect then has to chase.\n",{"name":5,"description":7},"main","dda161ff9daf0d9bccbdba26e34bbee0438a18b0",".claude\u002Fskills\u002Freference-mining",0,"https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub","www.finhub.net.au","2026-11-12 17:07:32",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add danielnguyenfinhub\u002Ffinhub-harness --skill reference-mining",{"tool":38,"command":39},"bunx","bunx skills add danielnguyenfinhub\u002Ffinhub-harness --skill reference-mining",{"tool":41,"command":42},"pnpm","pnpm dlx skills add danielnguyenfinhub\u002Ffinhub-harness --skill reference-mining","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Freference-mining\u002FSKILL.md",true,[46,52],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Flicence-rules.md","REFERENCE","参考文档",5636,"https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Freference-mining\u002Freferences\u002Flicence-rules.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fport-map-schema.md",2472,"https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Freference-mining\u002Freferences\u002Fport-map-schema.md",[57,64,71,77,83,90],{"skillNo":58,"name":59,"nameZh":6,"description":60,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":61,"categories":62,"keywords":63,"publishedAt":19},"01M7DMBP1MQ9X39BGJ9VF62M9P","linky","Fixture whose links and bundled-file mentions all resolve. Use to prove no false errors.","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fharness_links_ok\u002Fskills\u002Flinky\u002FSKILL.md",[],[],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":19},"01M7DMBK70KXF6A3V1VDRHYGVT","dup-one","Second owner of the name. Use for the duplicate fixture.","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fharness_links_bad\u002Fskills\u002Fdup-two\u002FSKILL.md",[],[],{"skillNo":72,"name":66,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7DMBFWK8HD4AWNK89KGE89P","First owner of the name. Use for the duplicate fixture.","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fharness_links_bad\u002Fskills\u002Fdup-one\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":6,"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},"01M7DMBETX7172TFNBAA3BVREQ","nested","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fharness_links_bad\u002Fskills\u002Fbroken\u002Fnested\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},"01M7DMBDCH5G61QVAB1QFMQA55","broken","Fixture with one defect per rule. Use to prove each rule fires.","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fharness_links_bad\u002Fskills\u002Fbroken\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},"01M7DMBAQHY0Z2PTYGCPM8H7MT","notes-style","House style for client notes. Use when writing or editing a client note.","https:\u002F\u002Fgithub.com\u002Fdanielnguyenfinhub\u002Ffinhub-harness\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fharness_good\u002Fskills\u002Fnotes-style\u002FSKILL.md",[],[]]