[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74MD8F9PKZC4XAXSVEA2B75":3,"skill-same:01M74MD8F9PKZC4XAXSVEA2B75":47},{"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":29,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M74MD8F9PKZC4XAXSVEA2B75","managed-model-endpoints",null,"Register a model service in the managed family — a local model server container the daemon starts\u002Fstops on demand, or a remote upstream model API (https). Read the runbook, allocate a port (local only), compose idempotent start\u002Fstop scripts (local only), register once. Load when the user wants a model service available for inference, or when list_compute shows managed endpoints.","pengqianhan\u002FAI-Human-Research-OS","pengqianhan","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F4286531?v=4","01M74M3DT9HNZ6HPAK8TX69EX9","PengqianHan","01M74M3DTVSN2GXW0JH2W396FK",3,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Fmanaged-model-endpoints\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: managed-model-endpoints\ndescription: Register a model service in the managed family — a local model server container the daemon starts\u002Fstops on demand, or a remote upstream model API (https). Read the runbook, allocate a port (local only), compose idempotent start\u002Fstop scripts (local only), register once. Load when the user wants a model service available for inference, or when list_compute shows managed endpoints.\nlicense: Apache-2.0\n---\n\n# Managed model endpoints\n\nA **managed model endpoint** is a model service the **daemon** owns: you\nregister it **once**, then every `compute_provider` cell against it just\nworks — the daemon swaps the resident model off the device (one model at a\ntime, via the resident's own approved stop), runs your approved start\nscript, waits for the readiness route, then runs your cell, streaming its\nlifecycle progress into the cell as it goes. You never run the container\nruntime yourself, never poll readiness in cells, and never see the\ncredential value. Two verbs: `register()` (asks the user once) and ordinary\ninference cells.\n\nContainer specifics — image, registry login, internal port, cache mount\ntarget, readiness route — come from the **model's own runbook skill**; this\nskill is the translation contract.\n\n## Calling a registered endpoint — inference cells\n\nCalling a registered endpoint — use the `using-model-endpoint` skill\n(this skill is the REGISTRATION contract; that one documents the call\nside in full).\n\nThe ONLY dispatch form is the `compute_provider` tool with the endpoint's\nregistered name (`list_compute` shows them):\n\n```\ncompute_provider(provider=\"boltz2-service\", code=\"\"\"\nimport requests\nr = requests.post(BASE_URL + \"\u002Fv1\u002Finfer\", json=payload)\n\"\"\")\n```\n\nThe daemon brings the model up on demand (a first cold start downloads\nimage + weights — minutes; let it run) and preloads `BASE_URL` into the\ncell — both as a Python variable (use it directly, as above) and as\n`os.environ[\"BASE_URL\"]` (plus `INFER_API_KEY` for remote endpoints). Endpoints are\n**not kernel environments**: `environment=\"boltz2-service\"` on a plain\npython cell fails — plain cells get no `BASE_URL`.\n\n## Enablement — once per machine\n\nThe user connects the family under **Customize → Compute → Model\nendpoints** (the setup flow saves the family credential first —\nconnect-without-key is not a state) and picks ONE mode: **Local**\n(container registrations) or **URL\u002Fremote** (https against the configured\nhost). Until connected, `free_port()`\u002F`register()` raise a precise error —\nrelay it; in the wrong mode they refuse with a teaching error naming the\nsetting (existing endpoints of the unarmed leg keep dispatching — only NEW\nregistrations refuse). **Disconnecting is a full teardown**: every active\nlocal service is stopped via its approved stop script and every\nregistration (local AND hosted) is removed; caches stay on disk; a failing\nstop keeps that one row, FAILED. The \"Local machine GPU\" toggle never\ngates registration — it governs cell GPU access only; the approval card is\nthe per-registration gate.\n\n**Credential contract (platform rule):** every registration passes\n`credential=\"NVIDIA_API_KEY\"` — the daemon rejects any other name. Locally\nthe value feeds the start script's registry login and never enters your\nkernel env; for remote endpoints it authenticates the upstream and is\ndelivered only into the inference cell's env (as `INFER_API_KEY`), never\nthe repl kernel.\n\n## Register (repl kernel)\n\n```python\nport = host.model_endpoints.free_port()        # local only; random 20000-29999\n\nhost.model_endpoints.register(\n    name=\"boltz2-service\",            # \u003Cmodel>-service -- descriptive, never the\n                                      # bare model name (collides\u002Fambiguous)\n    url=f\"http:\u002F\u002F127.0.0.1:{port}\",   # LITERAL 127.0.0.1 -- `localhost` rejected\n    credential=\"NVIDIA_API_KEY\",      # the family credential NAME, never a value\n    skill=\"\u003Cmodel-runbook-skill>\",\n    start=START_SCRIPT,               # composed below\n    stop=\"docker stop boltz2-service\",# exit 0 ONLY once actually stopped\n    live=\"\u002Fv1\u002Fhealth\u002Fready\",          # readiness ROUTE (200 = model answers;\n                                      # \"up but loading\" must read not-ready)\n)\n```\n\nName endpoints `\u003Cmodel>-service` (e.g. `diffdock-service`) — unambiguous in\nprovider lists; never just the bare model name. Name the CONTAINER after\nthe endpoint too (the template above does): the UI then follows the\nservice's own logs live while it starts.\n\n`register()` **always cards the user** (scripts verbatim, port, service dir,\ncredential name). One exception: a **byte-identical** re-registration is\nsilent — same bytes are approved forever; any byte change re-cards. The\nregistration stays inspectable under Customize → Compute. Re-registering to\nfix scripts: **reuse the existing url** — never call `free_port()` again\n(the port is the endpoint's stable mutex).\n\n## Remote endpoints — upstream APIs (no lifecycle)\n\nPass `url=\"https:\u002F\u002F\u003Cupstream>\"` and **omit `start`\u002F`stop`\u002F`live`** — no\nport, no scripts, no readiness. Requires URL\u002Fremote mode (the setup\nradio; in Local mode https registrations refuse). The url's HOST must\nequal the configured upstream host exactly — you pick the path leaf,\nnever the authority. After approval,\ncells are plain HTTP clients of `BASE_URL` authenticating with\n`$INFER_API_KEY`. `list_compute` labels every row\n`location: \"local\" | \"remote\"`.\n\n## Composing the start script\n\nThe daemon hands scripts three things in their **process environment**\n(never argv, never sudo): `HOST_PORT` (the registered port), `SERVICE_DIR`\n(this endpoint's persistent directory — put the model cache here), and the\ncredential value under its own name. Nothing else is inherited — ambient\ntokens are not visible; the ONLY secret a script sees is its registered\ncredential.\n\nThe start script must be **idempotent** (cold create \u002F warm start \u002F crash\nre-entry), with the **port-mismatch guard** — the runtime freezes port\nmappings at container creation, so a container created under an OLD port\nmust be recreated or readiness can never pass:\n\n```bash\nmkdir -p \"$SERVICE_DIR\u002Fcache\"\n# docker login persists auth in $DOCKER_CONFIG\u002Fconfig.json; scope it to the\n# service dir so the credential dies with the service (never ~\u002F.docker).\nexport DOCKER_CONFIG=\"$SERVICE_DIR\u002F.docker\"\ncreate_service() {\n  docker run -d --name boltz2-service \\\n    --restart unless-stopped \\\n    -p 127.0.0.1:${HOST_PORT}:8000 --gpus all \\\n    -e NVIDIA_API_KEY \\\n    -v \"$SERVICE_DIR\u002Fcache:\u003Ccache target from the runbook>\" \\\n    \u003Cimage from the runbook>\n}\nif docker inspect boltz2-service >\u002Fdev\u002Fnull 2>&1 && \\\n   [ \"$(docker inspect -f '{{(index (index .HostConfig.PortBindings \"8000\u002Ftcp\") 0).HostPort}}' boltz2-service)\" != \"$HOST_PORT\" ]; then\n  docker rm -f boltz2-service          # stale port mapping -- recreate below\nfi\nif docker inspect boltz2-service >\u002Fdev\u002Fnull 2>&1; then\n  docker start boltz2-service          # warm wake -- no credential, no chown needed\nelse\n  echo \"$NVIDIA_API_KEY\" | docker login \u003Cregistry> --username '\u003Cuser>' --password-stdin\n  docker pull \u003Cimage from the runbook>\n  # Cache must be writable by the CONTAINER's user, whose uid the image\n  # defines (container uid != host uid). chown needs root the script doesn't\n  # have; a throwaway root container does it -- and the chmod, which the\n  # host user can no longer do once the dir is chowned away -- without sudo.\n  CUID=\"$(docker inspect --format '{{.Config.User}}' \u003Cimage from the runbook> 2>\u002Fdev\u002Fnull | cut -d: -f1)\"\n  case \"$CUID\" in ''|root) CUID=0;; *[!0-9]*) CUID=1000;; esac   # named user -> default 1000; runbook may override\n  if [ \"$CUID\" != \"0\" ]; then\n    docker run --rm -v \"$SERVICE_DIR\u002Fcache:\u002Fc\" alpine sh -c \"chown -R $CUID:$CUID \u002Fc && chmod 700 \u002Fc\"\n  else\n    chmod 700 \"$SERVICE_DIR\u002Fcache\" 2>\u002Fdev\u002Fnull || true\n  fi\n  create_service\nfi\n# RUNTIME-binding guard (one retry): after a port-conflict crash the engine\n# can start the container yet silently skip port programming -- the CONFIG\n# still matches $HOST_PORT (so the guard above cannot catch it) but\n# `docker port` prints nothing and the model serves to nobody. Recreate.\nif [ -z \"$(docker port boltz2-service 2>\u002Fdev\u002Fnull)\" ]; then\n  docker rm -f boltz2-service\n  create_service\nfi\n```\n\nTranslation rules:\n\n- **Keep scripts ASCII** — non-ASCII (em dashes, arrows, curly quotes)\n  triggers the approval card's spoofing warning; use `--` and `->` in\n  comments.\n- **`export DOCKER_CONFIG=\"$SERVICE_DIR\u002F.docker\"`** before any `docker\n  login` — login persists the credential in `config.json`, and scoping it\n  to the service dir means Remove honestly reclaims it (never `~\u002F.docker`,\n  which outlives stop\u002FRemove\u002FDisable).\n- **`-p 127.0.0.1:${HOST_PORT}:\u003Cinternal>`** — loopback-only publish; the\n  internal port comes from the runbook.\n- If the image reads a different env name, bridge env→env at the top:\n  `export OTHER_NAME=\"$NVIDIA_API_KEY\"` (never argv, never a file).\n- **`-e NAME` bare** (argv is world-readable); the key rides the login\n  stdin pipe only.\n- **`-d`, no `--rm`** — managed containers are **stopped, never removed**:\n  stop parks them with weights loaded; `--rm` throws the cache away.\n- Cache under `$SERVICE_DIR`, owned by the **container's** uid: the\n  runbook states it when it matters; otherwise derive it post-pull with\n  `docker inspect --format '{{.Config.User}}' \u003Cimage>` (empty or `root`\n  ⇒ runs as root, no chown needed; a NAMED user can't be resolved\n  without running the image — default `1000`). Getting it wrong is the\n  cache-empty symptom: the container can't write the mount, weights leak\n  into the writable layer and die on recreate (or the image crash-loops\n  on Permission denied). Never `777` — world-writable cache on a\n  multi-user host. The mount TARGET comes from the runbook.\n- Cells need **no auth header** against local endpoints — the credential is\n  a pull key that never enters your kernel.\n\n## Failures\n\nA failed start\u002Fstop flips the endpoint **FAILED** (transcript on the\nendpoint panel — never echoed into cell errors; ask the user to read it\nthere) and your cell errors with the daemon's one-line cause. FAILED is\n**sticky**: further cells fail fast until the user presses Stop or you\nre-register (byte-identical re-register also clears it). If a stop is stuck\n(exit 0 but the port never frees), removal is refused while the port is\nbound — recover out-of-band; the daemon absorbs the freed port on its next\nprobe. A first-ever cold start downloads image + weights — minutes, once;\nthe cell streams the phase lines live and the endpoint detail view streams\nthe full script output, so let it run.\n","# Managed model endpoints\n\nA **managed model endpoint** is a model service the **daemon** owns: you\nregister it **once**, then every `compute_provider` cell against it just\nworks — the daemon swaps the resident model off the device (one model at a\ntime, via the resident's own approved stop), runs your approved start\nscript, waits for the readiness route, then runs your cell, streaming its\nlifecycle progress into the cell as it goes. You never run the container\nruntime yourself, never poll readiness in cells, and never see the\ncredential value. Two verbs: `register()` (asks the user once) and ordinary\ninference cells.\n\nContainer specifics — image, registry login, internal port, cache mount\ntarget, readiness route — come from the **model's own runbook skill**; this\nskill is the translation contract.\n\n## Calling a registered endpoint — inference cells\n\nCalling a registered endpoint — use the `using-model-endpoint` skill\n(this skill is the REGISTRATION contract; that one documents the call\nside in full).\n\nThe ONLY dispatch form is the `compute_provider` tool with the endpoint's\nregistered name (`list_compute` shows them):\n\n```\ncompute_provider(provider=\"boltz2-service\", code=\"\"\"\nimport requests\nr = requests.post(BASE_URL + \"\u002Fv1\u002Finfer\", json=payload)\n\"\"\")\n```\n\nThe daemon brings the model up on demand (a first cold start downloads\nimage + weights — minutes; let it run) and preloads `BASE_URL` into the\ncell — both as a Python variable (use it directly, as above) and as\n`os.environ[\"BASE_URL\"]` (plus `INFER_API_KEY` for remote endpoints). Endpoints are\n**not kernel environments**: `environment=\"boltz2-service\"` on a plain\npython cell fails — plain cells get no `BASE_URL`.\n\n## Enablement — once per machine\n\nThe user connects the family under **Customize → Compute → Model\nendpoints** (the setup flow saves the family credential first —\nconnect-without-key is not a state) and picks ONE mode: **Local**\n(container registrations) or **URL\u002Fremote** (https against the configured\nhost). Until connected, `free_port()`\u002F`register()` raise a precise error —\nrelay it; in the wrong mode they refuse with a teaching error naming the\nsetting (existing endpoints of the unarmed leg keep dispatching — only NEW\nregistrations refuse). **Disconnecting is a full teardown**: every active\nlocal service is stopped via its approved stop script and every\nregistration (local AND hosted) is removed; caches stay on disk; a failing\nstop keeps that one row, FAILED. The \"Local machine GPU\" toggle never\ngates registration — it governs cell GPU access only; the approval card is\nthe per-registration gate.\n\n**Credential contract (platform rule):** every registration passes\n`credential=\"NVIDIA_API_KEY\"` — the daemon rejects any other name. Locally\nthe value feeds the start script's registry login and never enters your\nkernel env; for remote endpoints it authenticates the upstream and is\ndelivered only into the inference cell's env (as `INFER_API_KEY`), never\nthe repl kernel.\n\n## Register (repl kernel)\n\n```python\nport = host.model_endpoints.free_port()        # local only; random 20000-29999\n\nhost.model_endpoints.register(\n    name=\"boltz2-service\",            # \u003Cmodel>-service -- descriptive, never the\n                                      # bare model name (collides\u002Fambiguous)\n    url=f\"http:\u002F\u002F127.0.0.1:{port}\",   # LITERAL 127.0.0.1 -- `localhost` rejected\n    credential=\"NVIDIA_API_KEY\",      # the family credential NAME, never a value\n    skill=\"\u003Cmodel-runbook-skill>\",\n    start=START_SCRIPT,               # composed below\n    stop=\"docker stop boltz2-service\",# exit 0 ONLY once actually stopped\n    live=\"\u002Fv1\u002Fhealth\u002Fready\",          # readiness ROUTE (200 = model answers;\n                                      # \"up but loading\" must read not-ready)\n)\n```\n\nName endpoints `\u003Cmodel>-service` (e.g. `diffdock-service`) — unambiguous in\nprovider lists; never just the bare model name. Name the CONTAINER after\nthe endpoint too (the template above does): the UI then follows the\nservice's own logs live while it starts.\n\n`register()` **always cards the user** (scripts verbatim, port, service dir,\ncredential name). One exception: a **byte-identical** re-registration is\nsilent — same bytes are approved forever; any byte change re-cards. The\nregistration stays inspectable under Customize → Compute. Re-registering to\nfix scripts: **reuse the existing url** — never call `free_port()` again\n(the port is the endpoint's stable mutex).\n\n## Remote endpoints — upstream APIs (no lifecycle)\n\nPass `url=\"https:\u002F\u002F\u003Cupstream>\"` and **omit `start`\u002F`stop`\u002F`live`** — no\nport, no scripts, no readiness. Requires URL\u002Fremote mode (the setup\nradio; in Local mode https registrations refuse). The url's HOST must\nequal the configured upstream host exactly — you pick the path leaf,\nnever the authority. After approval,\ncells are plain HTTP clients of `BASE_URL` authenticating with\n`$INFER_API_KEY`. `list_compute` labels every row\n`location: \"local\" | \"remote\"`.\n\n## Composing the start script\n\nThe daemon hands scripts three things in their **process environment**\n(never argv, never sudo): `HOST_PORT` (the registered port), `SERVICE_DIR`\n(this endpoint's persistent directory — put the model cache here), and the\ncredential value under its own name. Nothing else is inherited — ambient\ntokens are not visible; the ONLY secret a script sees is its registered\ncredential.\n\nThe start script must be **idempotent** (cold create \u002F warm start \u002F crash\nre-entry), with the **port-mismatch guard** — the runtime freezes port\nmappings at container creation, so a container created under an OLD port\nmust be recreated or readiness can never pass:\n\n```bash\nmkdir -p \"$SERVICE_DIR\u002Fcache\"\n# docker login persists auth in $DOCKER_CONFIG\u002Fconfig.json; scope it to the\n# service dir so the credential dies with the service (never ~\u002F.docker).\nexport DOCKER_CONFIG=\"$SERVICE_DIR\u002F.docker\"\ncreate_service() {\n  docker run -d --name boltz2-service \\\n    --restart unless-stopped \\\n    -p 127.0.0.1:${HOST_PORT}:8000 --gpus all \\\n    -e NVIDIA_API_KEY \\\n    -v \"$SERVICE_DIR\u002Fcache:\u003Ccache target from the runbook>\" \\\n    \u003Cimage from the runbook>\n}\nif docker inspect boltz2-service >\u002Fdev\u002Fnull 2>&1 && \\\n   [ \"$(docker inspect -f '{{(index (index .HostConfig.PortBindings \"8000\u002Ftcp\") 0).HostPort}}' boltz2-service)\" != \"$HOST_PORT\" ]; then\n  docker rm -f boltz2-service          # stale port mapping -- recreate below\nfi\nif docker inspect boltz2-service >\u002Fdev\u002Fnull 2>&1; then\n  docker start boltz2-service          # warm wake -- no credential, no chown needed\nelse\n  echo \"$NVIDIA_API_KEY\" | docker login \u003Cregistry> --username '\u003Cuser>' --password-stdin\n  docker pull \u003Cimage from the runbook>\n  # Cache must be writable by the CONTAINER's user, whose uid the image\n  # defines (container uid != host uid). chown needs root the script doesn't\n  # have; a throwaway root container does it -- and the chmod, which the\n  # host user can no longer do once the dir is chowned away -- without sudo.\n  CUID=\"$(docker inspect --format '{{.Config.User}}' \u003Cimage from the runbook> 2>\u002Fdev\u002Fnull | cut -d: -f1)\"\n  case \"$CUID\" in ''|root) CUID=0;; *[!0-9]*) CUID=1000;; esac   # named user -> default 1000; runbook may override\n  if [ \"$CUID\" != \"0\" ]; then\n    docker run --rm -v \"$SERVICE_DIR\u002Fcache:\u002Fc\" alpine sh -c \"chown -R $CUID:$CUID \u002Fc && chmod 700 \u002Fc\"\n  else\n    chmod 700 \"$SERVICE_DIR\u002Fcache\" 2>\u002Fdev\u002Fnull || true\n  fi\n  create_service\nfi\n# RUNTIME-binding guard (one retry): after a port-conflict crash the engine\n# can start the container yet silently skip port programming -- the CONFIG\n# still matches $HOST_PORT (so the guard above cannot catch it) but\n# `docker port` prints nothing and the model serves to nobody. Recreate.\nif [ -z \"$(docker port boltz2-service 2>\u002Fdev\u002Fnull)\" ]; then\n  docker rm -f boltz2-service\n  create_service\nfi\n```\n\nTranslation rules:\n\n- **Keep scripts ASCII** — non-ASCII (em dashes, arrows, curly quotes)\n  triggers the approval card's spoofing warning; use `--` and `->` in\n  comments.\n- **`export DOCKER_CONFIG=\"$SERVICE_DIR\u002F.docker\"`** before any `docker\n  login` — login persists the credential in `config.json`, and scoping it\n  to the service dir means Remove honestly reclaims it (never `~\u002F.docker`,\n  which outlives stop\u002FRemove\u002FDisable).\n- **`-p 127.0.0.1:${HOST_PORT}:\u003Cinternal>`** — loopback-only publish; the\n  internal port comes from the runbook.\n- If the image reads a different env name, bridge env→env at the top:\n  `export OTHER_NAME=\"$NVIDIA_API_KEY\"` (never argv, never a file).\n- **`-e NAME` bare** (argv is world-readable); the key rides the login\n  stdin pipe only.\n- **`-d`, no `--rm`** — managed containers are **stopped, never removed**:\n  stop parks them with weights loaded; `--rm` throws the cache away.\n- Cache under `$SERVICE_DIR`, owned by the **container's** uid: the\n  runbook states it when it matters; otherwise derive it post-pull with\n  `docker inspect --format '{{.Config.User}}' \u003Cimage>` (empty or `root`\n  ⇒ runs as root, no chown needed; a NAMED user can't be resolved\n  without running the image — default `1000`). Getting it wrong is the\n  cache-empty symptom: the container can't write the mount, weights leak\n  into the writable layer and die on recreate (or the image crash-loops\n  on Permission denied). Never `777` — world-writable cache on a\n  multi-user host. The mount TARGET comes from the runbook.\n- Cells need **no auth header** against local endpoints — the credential is\n  a pull key that never enters your kernel.\n\n## Failures\n\nA failed start\u002Fstop flips the endpoint **FAILED** (transcript on the\nendpoint panel — never echoed into cell errors; ask the user to read it\nthere) and your cell errors with the daemon's one-line cause. FAILED is\n**sticky**: further cells fail fast until the user presses Stop or you\nre-register (byte-identical re-register also clears it). If a stop is stuck\n(exit 0 but the port never frees), removal is refused while the port is\nbound — recover out-of-band; the daemon absorbs the freed port on its next\nprobe. A first-ever cold start downloads image + weights — minutes, once;\nthe cell streams the phase lines live and the endpoint detail view streams\nthe full script output, so let it run.\n",{"name":5,"license":15,"description":7},"main","e71245365df78e06cd166602744bf2491f79910e","research-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Fmanaged-model-endpoints",0,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS","https:\u002F\u002Fpengqianhan.github.io\u002FAI-Human-Research-OS\u002F","https:\u002F\u002Fgithub.com\u002Fpengqianhan","Ph.D. candidate of The University of Auckland, New Zealand","2026-11-14 18:08:16",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add pengqianhan\u002FAI-Human-Research-OS --skill managed-model-endpoints",{"tool":39,"command":40},"bunx","bunx skills add pengqianhan\u002FAI-Human-Research-OS --skill managed-model-endpoints",{"tool":42,"command":43},"pnpm","pnpm dlx skills add pengqianhan\u002FAI-Human-Research-OS --skill managed-model-endpoints","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Fmanaged-model-endpoints\u002FSKILL.md",true,[],[48,56,63,70,77,84],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":19},"01M74MWYTHT498SVMN823896T7","workflow-skill-creator","Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say \"make this a skill\", \"create a skill from what we just did\", \"package this workflow\" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).","MIT","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fworkflow_skill_creator\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M74MWWVCFZ8HQS8840F1V1FT","uv","Checks whether the uv Python package manager is installed and installs it if missing. Ensures uv is on PATH. Use when another skill requires uv as a prerequisite.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fuv\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M74MWRH5BTSWZ2C5PCTD5FV9","uniprot-database","Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funiprot_database\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M74MWN1D3ER9F1FZPWH8HQC1","unibind-database","Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED\u002FFASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funibind_database\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M74MWHHB5QZY3NR2FYA1EK4N","ucsc-conservation-and-tfbs","Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fucsc_conservation_and_tfbs\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M74MWERYHSY87P21FDMVE6BA","string-database","Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fstring_database\u002FSKILL.md",[],[]]