[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7889EMFDCDWK53DP1XDX26J":3,"skill-same:01M7889EMFDCDWK53DP1XDX26J":73},{"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":33,"commitSha":34,"skillDirectory":35,"forksCount":14,"repositoryHtmlUrl":36,"repositoryHomepageUrl":6,"ownerHtmlUrl":37,"ownerBlogUrl":38,"ownerDescription":39,"updatedAt":40,"install":41,"resources":54},"01M7889EMFDCDWK53DP1XDX26J","tao-launch-workflow",null,"Shared launch intake for any TAO workflow or action. Use when the user wants to run TAO AutoML, train, evaluate, infer, export, generate TensorRT engines, or launch DEFT\u002Fworkflow jobs on an execution platform.","Solizardking\u002Ftrain2earn","Solizardking","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F108096810?v=4","01M786KDB10AYV042N0F9BD5B5","Clawd","01M786KDC5C3NPE4W44C5641EZ",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: tao-launch-workflow\ndescription: >-\n  Shared launch intake for any TAO workflow or action. Use when the user wants\n  to run TAO AutoML, train, evaluate, infer, export, generate TensorRT engines,\n  or launch DEFT\u002Fworkflow jobs on an execution platform.\nlicense: Apache-2.0\ncompatibility: Requires the packaged TAO skill bank helper scripts.\nmetadata:\n  author: NVIDIA Corporation\n  version: \"0.1.0\"\nallowed-tools: Read Bash\ntags:\n- tao\n- workflow\n- launch\n---\n\n# TAO Workflow Launch Intake\n\nUse this skill before launching any TAO workflow or model action.\n\n## Quick Start\n\nRun the platform helper, ask for platform and monitoring preferences, then run\nthe selected platform detail helper before asking for credentials.\n\n## Non-Negotiable Launch Gate\n\nThis gate is model-agnostic. Apply it to every TAO model, data action, and\napplication workflow before launching side-effecting work.\n\nDo **not** create runner scripts, launch scripts, compatibility shims,\nworkspace folders, state files, logs, or dependency-install side effects until\nthe launch preflight passes.\n\nPreflight passes only after all of these are true:\n\n1. The execution platform is selected from the packaged platform helper.\n2. Platform credentials and required credential groups are satisfied.\n3. Model-specific credentials are satisfied.\n4. The default container image is resolved from packaged model\u002Faction metadata,\n   shown to the user, and either confirmed or replaced by an explicit\n   `image=\u003Coverride>`.\n5. The platform access check succeeds from the launch host.\n6. Dataset inputs are mapped to concrete spec keys and verified from the\n   selected platform's point of view.\n7. Required compute shape fields from the model\u002Fworkflow skill are known.\n8. Required local tools for the selected data\u002Fplatform path are present, or the\n   user approved installing the smallest missing dependency and preflight was\n   rerun.\n9. A launch review with image, platform, datasets, compute shape, expected\n   runtime, and any generated\u002Fdefault configuration changes has been shown and\n   confirmed by the user. For AutoML, the launch review must explicitly state\n   recommendation count\u002Fbudget, max concurrency, algorithm, metric, direction,\n   and searched parameters\u002Franges even when defaults are used.\n\nIf any item is missing, ask for the missing input and stop before generating\nartifacts. This applies to AutoML, normal train\u002Feval\u002Finfer\u002Fexport\u002FTRT, and\nDEFT\u002Fapplication workflows.\n\nWhen preflight work clears a blocker, keep track of the original user request.\nAfter the fix, rerun the relevant preflight and continue toward that request;\ndo not stop at \"blocker fixed\" unless the user explicitly asked only for the\nrepair.\n\n## Initial Questions\n\nAfter the user confirms what they want to do, ask for the execution platform\nusing the packaged helper. Do not scan platform docs, skill folders, or config\nfolders to build the choices.\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Flist_tao_platforms.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} --format text\n```\n\nThen ask:\n\n- Which supported platform should run this workflow?\n- Should I monitor the run in this chat? Monitoring means I keep polling the\n  backend\u002Fjob logs after launch and report progress until the job finishes,\n  fails, or you ask me to stop, even if the job stays queued for hours or days.\n  If disabled, I launch the job, give you the job id\u002Flog path, and stop\n  polling. Default: monitor in chat.\n- How often should I post status? Default: every 5 minutes. Use 1-2 minutes for\n  smoke tests, 5 minutes for normal training, or 10-15 minutes for long runs.\n\nUse `long_running_enabled=true` and `status_interval_minutes=5` when the user\naccepts the defaults.\n\nWhen monitoring is enabled, do not send a final summary just because several\npolls have elapsed or the job is still `PENDING`. Keep the turn attached and\nemit status every `status_interval_minutes` until a terminal state or explicit\nuser stop\u002Fdetach request. If the runtime environment cannot keep the chat turn\nopen, say that clearly and leave a durable watcher\u002Flog path; do not imply that\nchat updates will continue after the turn ends.\n\nFinal-answer rule: a `final` response ends chat-side monitoring. While\n`long_running_enabled=true` and any launched job is non-terminal, status\nmessages must be sent as in-progress updates and the agent must continue\npolling. Only send a final response when the workflow reaches terminal state,\nthe user explicitly asks to detach\u002Fstop monitoring, or the runtime genuinely\ncannot keep the turn open; in that last case, say it is a runtime limitation\nand provide the exact durable status command\u002Flog path.\n\n## Missing-Input Prompt Shape\n\nWhen asking for launch inputs, include concrete examples and both dataset input\nmodes. Do not ask only for \"dataset root\".\n\nUse this structure and adapt spec keys to the selected model\u002Faction:\n\n```text\nI need these launch inputs before I can create specs or runner files:\n\n1. Execution platform: brev, slurm, local-docker, or kubernetes.\n\n2. Dataset inputs. You can provide either mode:\n   A) Root mode: give train\u002Feval roots and I map required files automatically.\n      Example Cosmos-RL:\n      train_root=\u002Flustre\u002Ffsw\u002F...\u002Fcosmos\u002Ftrain\n      -> custom.train_dataset.annotation_path=train_root\u002Fannotations.json\n      -> custom.train_dataset.media_path=train_root\n   B) Direct spec mode: give the exact config\u002Fspec parameters yourself.\n      Example:\n      custom.train_dataset.annotation_path=\u002Flustre\u002Ffsw\u002F...\u002Ftrain_annotations.json\n      custom.train_dataset.media_path=\u002Flustre\u002Ffsw\u002F...\u002Fvideos_train.tar.gz\n      custom.val_dataset.annotation_path=\u002Flustre\u002Ffsw\u002F...\u002Feval_annotations.json\n      custom.val_dataset.media_path=\u002Flustre\u002Ffsw\u002F...\u002Feval_videos\u002F\n\n   Platform examples:\n   - SLURM\u002FLustre: \u002Flustre\u002Ffsw\u002F...\u002Fdata\u002Ftrain or lustre:\u002F\u002F\u002Flustre\u002Ffsw\u002F...\u002Fdata\u002Ftrain\n   - Brev\u002FKubernetes: s3:\u002F\u002Fbucket\u002Fpath\u002Ftrain and s3:\u002F\u002Fbucket\u002Fpath\u002Feval\n   - local-docker: \u002Fdata\u002Ftao\u002F\u003Cmodel>\u002Ftrain or file:\u002F\u002F\u002Fdata\u002Ftao\u002F\u003Cmodel>\u002Feval\n\n3. Container image. I will resolve the default from packaged model metadata and\n   show it before launch, for example:\n   default image for \u003Cmodel>\u002F\u003Caction>: \u003Cresolved container image>\n   Use this image, or provide image=\u003Coverride> to pin a different TAO build.\n\n4. Compute shape required by the model, for example GPUs\u002Fnodes.\n\n5. Required credentials from platform\u002Fmodel docs, for example HF_TOKEN for\n   gated Hugging Face models.\n\n6. Monitoring preference. By default I monitor in this chat and post progress\n   every 5 minutes; choose 1-2 minutes for smoke tests or 10-15 minutes for\n   long training.\n```\n\n## Container Image Confirmation\n\nBefore creating specs, runner scripts, workspaces, logs, state files, or\nsubmitting a job, resolve the image for the selected model\u002Faction:\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Fresolve_tao_image.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} \\\n  --model \u003Cnetwork> --action \u003Caction> --format text\n```\n\nIf the helper is unavailable, read `skills\u002Fmodels\u002F\u003Cnetwork>\u002Fconfig.json` through\n`SkillBank().get_model_config(network_arch)`. Resolve image fields in this\norder:\n\n1. `actions.\u003Caction>.container_image`\n2. `actions.\u003Caction>.image`\n3. top-level `container_image`\n4. top-level `image`\n\nShow the exact image and ask:\n\n```text\nContainer image for \u003Cnetwork>\u002F\u003Caction>:\ndefault=\u003Cresolved image>\n\nUse this image, or provide image=\u003Coverride>?\n```\n\nIf the user accepts, pass the resolved image as the job `image`. If the user\noverrides, require a non-empty image reference and pass that value instead.\nDo not silently launch on the default image. This confirmation applies to\ntraining, AutoML recommendations, evaluation, inference, export, TensorRT\nengine generation, and application workflows that submit TAO containers.\n\n## Credential Filtering\n\nAfter the user chooses a platform, get the credential list for only that\nplatform:\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Flist_tao_platforms.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} \\\n  --platform \u003Cplatform> --format text\n```\n\nAsk only for credentials returned by that command, plus model-specific\ncredentials from the selected model skill. Do not ask for Brev credentials on\nSLURM, Kubernetes, or local Docker. Do not ask for SLURM credentials on Brev,\nKubernetes, or local Docker. Ask S3 credentials only when the selected\nplatform and the dataset\u002Fresult URIs require `s3:\u002F\u002F` access.\nCredentials may already be present in the process environment or in a\nuser-approved secret env file such as `~\u002F.tao\u002Fsecrets.env` or\n`~\u002F.config\u002Ftao\u002F.env`; source such files only when needed and never print,\ngrep, cat, paste, or log their contents. Verify only variable presence.\n\nFor initial launch intake, ask for required credentials and required credential\ngroups only. Treat the helper's optional credentials\u002Fsettings section as\nreference material; do not request those values unless their `only_when`\ncondition applies, the selected workflow cannot proceed without them, or the\nuser asks to customize that setting.\n\nWhen the helper output includes a \"Required credential groups\" section, satisfy\none credential from each group before proceeding. Explain each requested value\nusing the helper's description and \"How to get it\" text.\n\nFor SLURM, user-facing prompts should ask for `SSH_KEY_PATH` first. Mention\n`SSH_AUTH_SOCK` only if the user says they already use an SSH agent.\n\n## Dependency Remediation\n\nIf a required CLI\u002Flibrary is missing, say exactly what is missing and why it is\nneeded, then ask before installing. Examples:\n\n- S3 dataset or results path -> require an S3-capable client such as `aws`.\n- SDK-backed platform launch -> require the platform-specific\n  `nvidia-tao-sdk[...]` extra.\n- Local Docker SDK path -> require the Docker Python client and the configured\n  Docker network.\n\nAfter user approval and installation, rerun the same preflight. Do not create\nrunner files or launch jobs between the failed check and the rerun.\n\n## Dataset Intake\n\nAccept dataset inputs in either mode:\n\n- **Dataset root mode:** the user gives train\u002Feval\u002Fcalibration roots, and the\n  model skill maps required files by convention. Example for Cosmos-RL train:\n  `custom.train_dataset.annotation_path=\u003Croot>\u002Fannotations.json` and\n  `custom.train_dataset.media_path=\u003Croot>`.\n- **Direct spec mode:** the user gives exact spec-key paths when annotations,\n  media archives, videos, or image folders live in different places. Preserve\n  those keys directly, for example\n  `custom.train_dataset.annotation_path=\u002Flustre\u002F...\u002Ftrain_annotations.json`\n  and `custom.train_dataset.media_path=\u002Flustre\u002F...\u002Fvideos.tar.gz`.\n\nAsk for dataset examples that match the selected platform:\n\n- SLURM: shared cluster paths such as\n  `\u002Flustre\u002Ffsw\u002Fportfolios\u002F\u003Cteam>\u002F\u003Cyour-dir>\u002Fdata\u002F\u003Cmodel>\u002Ftrain` (where\n  `\u003Cyour-dir>` is your per-user directory on the cluster), or direct\n  spec paths under `\u002Flustre\u002F...`.\n- Brev, Kubernetes: usually `s3:\u002F\u002Fbucket\u002Fpath\u002Ftrain` and\n  `s3:\u002F\u002Fbucket\u002Fpath\u002Feval` unless the platform profile mounts shared storage.\n- Local Docker: local paths visible to the Docker host, such as\n  `\u002Fdata\u002Ftao\u002F\u003Cmodel>\u002Ftrain`, or direct spec paths visible inside the planned\n  container mount.\n- Remote Docker: absolute paths visible on the remote Docker host named by\n  `DOCKER_HOST`, not paths on the local agent machine.\n\nDo not assume \"dataset root\" is the only acceptable input. When direct spec\npaths are supplied, validate the exact spec paths rather than appending default\nfilenames.\n\n## Platform Preflight\n\nRun the selected platform's preflight checks before any launch artifact is\ncreated.\n\nPrefer the packaged preflight helper when the needed inputs are available:\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Fcheck_tao_launch_preflight.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} \\\n  --platform \u003Cplatform> \\\n  --container-image \u003Cselected-image> \\\n  --path train_annotation=\u003Cpath> \\\n  --path train_media=\u003Cpath>\n```\n\nPass exact direct spec paths when the user supplied them. For root-mode inputs,\nexpand model-required files first, then pass those concrete annotation\u002Fmedia\npaths to the helper.\n\nIf the helper reports a missing client tool such as `aws` for `s3:\u002F\u002F` path\nverification, install the smallest needed package after user approval, then\nrerun the same command with `--install-missing-tools` and do not proceed until\nthe rerun verifies the paths.\n\nWhen the selected model skill warns that large S3 media should be staged, copy\nor extract the data once to platform-visible storage before creating launch\nartifacts, then validate those staged paths with the same preflight helper.\nRecord the source URI and staged path in the run workspace so AutoML summaries\ncan distinguish data staging time from training\u002Fevaluation time.\n\nFor `local-docker` and `remote-docker`, always pass the selected image with\n`--container-image` after resolving `container_image` from\n`skill_info.yaml`\u002F`versions.yaml`. The helper verifies Docker reachability,\nNVIDIA Container Toolkit registration, GPU memory, selected-image architecture\ncompatibility when known, and a GPU-visible smoke container before launch. For\n`remote-docker`, pass `--docker-host` or export `DOCKER_HOST`; the helper queries\nGPUs and validates bind-mounted paths through the remote daemon instead of using\nlocal host state. If the selected or smoke image is not present on the target\nDocker host, ask before pulling it or rerun with `--pull-smoke-image` after\napproval.\n\nWhen a model skill lists annotation-level required fields, pass them with\n`--json-required-field \u003Cpath-label>=\u003Cfield>[,\u003Cfield>...]` so schema\u002Fdata\ncontent issues fail during preflight rather than inside the first training\ncontainer. Do not add required annotation fields from old failure history; only\nenforce fields documented as required by the current model skill.\nFor local JSON\u002FJSONL annotation paths, the helper prints `records=\u003CN>`; use the\ntrain annotation count as `automl_settings[\"train_sample_count\"]` for\nsample-count-sensitive AutoML runs before recommendations are generated.\nIf the model skill documents a run-local patch strategy for a missing required\nfield, create the patched copy in the current run workspace, update the spec\npaths to that copy, and rerun the content check before launch. Do not ask the\nuser to mutate source datasets unless the model skill says patching is\nimpossible.\n\nDo not use `--skip-platform-access` for a real launch. That flag is only for\ndry environment checks or for cases where the user has already provided explicit\nmanual proof of platform and storage access. If the helper cannot verify remote\nAPI, CLI, cluster, or object-store access, treat preflight as failed and do not\ngenerate launch artifacts.\n\nFor SLURM:\n\n1. Require `SLURM_USER`, `SLURM_HOSTNAME`, a partition intent, and one of\n   `SSH_KEY_PATH` or `SSH_AUTH_SOCK`. If the user says to use the cluster\n   default partition, pass an empty partition\u002Fomit the partition directive; do\n   not substitute a site-specific value such as `batch`.\n   Use the selected platform helper's `Resource defaults` for runtime values.\n   For the packaged SLURM defaults, generate launchers with\n   `SLURM_TIME_HOURS=4` and `SLURM_TIMEOUT_HOURS=3.8`; never invent a\n   12-hour default for the 4-hour partition list.\n   Launching the orchestrator with `nohup` or in the background is allowed for\n   durability, but it does not satisfy chat monitoring by itself. After launch,\n   keep a foreground chat-side polling loop attached until terminal state or\n   explicit detach.\n2. Split comma-separated `SLURM_HOSTNAME`, resolve hosts where possible, and\n   require passwordless `ssh -o BatchMode=yes` to at least one host.\n3. If SSH fails, do not offer several equivalent choices. Ask for\n   `SSH_KEY_PATH=\u002Fpath\u002Fto\u002Fprivate_key` and show the passwordless setup steps:\n   create a key if needed with\n   `ssh-keygen -t ed25519 -N \"\" -f ~\u002F.ssh\u002Fid_ed25519`; install it with\n   `ssh-copy-id -i ~\u002F.ssh\u002Fid_ed25519.pub \u003CSLURM_USER>@\u003Clogin-host>`; trust the\n   host with `ssh-keyscan -H \u003Clogin-host> >> ~\u002F.ssh\u002Fknown_hosts`; set\n   `chmod 600 ~\u002F.ssh\u002Fid_ed25519`; verify with\n   `ssh -o BatchMode=yes -i ~\u002F.ssh\u002Fid_ed25519 \u003CSLURM_USER>@\u003Clogin-host> 'hostname'`;\n   then rerun with `SSH_KEY_PATH=~\u002F.ssh\u002Fid_ed25519`.\n4. After SSH passes, validate dataset annotation\u002Fmedia paths on the remote login\n   host with `test -e` or an equivalent read-only command.\n5. Only then create runner scripts, specs, workspaces, or submit jobs.\n6. For multi-GPU Slurm jobs, rely on the SDK Slurm backend to request\n   `--gpus-per-node=\u003CN>`. Do not generate manual `--gpus=\u003CN>` sbatch snippets;\n   that can spread GPUs across nodes and leave allocated GPUs idle.\n7. For full-matrix or multi-node launches, submit one smoke job first. Launch\n   the full matrix only after the smoke reaches training, emits the requested\n   metric\u002Fstatus record, and shows expected GPU utilization.\n\nFor AutoML status, prefer structured controller\u002Fbrain state and job metadata\n(`active_jobs.json`, `.automl\u002Fcontroller\u002F*.json`, result JSON, and\n`results_dir\u002Ftrain\u002Fstatus.json`) before scanning raw logs. Parse logs only as a\nfallback or when the user specifically asks for log-level investigation.\n\nFor local Docker, validate Docker\u002FGPU access and local dataset paths before\nwriting launch artifacts. For Brev and Kubernetes, validate API or\ncluster access plus object-storage credentials and `aws s3 ls` readability for\n`s3:\u002F\u002F` inputs before writing launch artifacts. For mounted shared-storage or\nPVC paths on those remote platforms, require manual proof that the path is\nmounted into the job environment; the helper fails closed rather than accepting\nunverified remote mount paths.\n\n## Runtime And Configuration Review\n\nBefore any side-effecting launch, show a concise review:\n\n- selected platform and exact container image\n- GPU ids\u002Fcount and nodes, including any GPUs avoided because they are already\n  occupied\n- dataset roots or direct spec paths, with sample counts when available\n- important model\u002Fworkflow overrides that differ from template defaults\n- estimated runtime and the assumptions behind it\n- monitoring interval and whether chat-side monitoring will stay attached\n\nFor AutoML, also show the algorithm, metric\u002Fdirection, recommendation budget,\nsearch parameters, ranges, and generated\u002Fdefault recommendation details as\ndescribed in `skills\u002Fapplications\u002Ftao-run-automl\u002FSKILL.md`. Ask for confirmation after\nthis review. If the user supplied a time limit, flag any plan that exceeds it\nand offer concrete reductions before launch.\n","# TAO Workflow Launch Intake\n\nUse this skill before launching any TAO workflow or model action.\n\n## Quick Start\n\nRun the platform helper, ask for platform and monitoring preferences, then run\nthe selected platform detail helper before asking for credentials.\n\n## Non-Negotiable Launch Gate\n\nThis gate is model-agnostic. Apply it to every TAO model, data action, and\napplication workflow before launching side-effecting work.\n\nDo **not** create runner scripts, launch scripts, compatibility shims,\nworkspace folders, state files, logs, or dependency-install side effects until\nthe launch preflight passes.\n\nPreflight passes only after all of these are true:\n\n1. The execution platform is selected from the packaged platform helper.\n2. Platform credentials and required credential groups are satisfied.\n3. Model-specific credentials are satisfied.\n4. The default container image is resolved from packaged model\u002Faction metadata,\n   shown to the user, and either confirmed or replaced by an explicit\n   `image=\u003Coverride>`.\n5. The platform access check succeeds from the launch host.\n6. Dataset inputs are mapped to concrete spec keys and verified from the\n   selected platform's point of view.\n7. Required compute shape fields from the model\u002Fworkflow skill are known.\n8. Required local tools for the selected data\u002Fplatform path are present, or the\n   user approved installing the smallest missing dependency and preflight was\n   rerun.\n9. A launch review with image, platform, datasets, compute shape, expected\n   runtime, and any generated\u002Fdefault configuration changes has been shown and\n   confirmed by the user. For AutoML, the launch review must explicitly state\n   recommendation count\u002Fbudget, max concurrency, algorithm, metric, direction,\n   and searched parameters\u002Franges even when defaults are used.\n\nIf any item is missing, ask for the missing input and stop before generating\nartifacts. This applies to AutoML, normal train\u002Feval\u002Finfer\u002Fexport\u002FTRT, and\nDEFT\u002Fapplication workflows.\n\nWhen preflight work clears a blocker, keep track of the original user request.\nAfter the fix, rerun the relevant preflight and continue toward that request;\ndo not stop at \"blocker fixed\" unless the user explicitly asked only for the\nrepair.\n\n## Initial Questions\n\nAfter the user confirms what they want to do, ask for the execution platform\nusing the packaged helper. Do not scan platform docs, skill folders, or config\nfolders to build the choices.\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Flist_tao_platforms.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} --format text\n```\n\nThen ask:\n\n- Which supported platform should run this workflow?\n- Should I monitor the run in this chat? Monitoring means I keep polling the\n  backend\u002Fjob logs after launch and report progress until the job finishes,\n  fails, or you ask me to stop, even if the job stays queued for hours or days.\n  If disabled, I launch the job, give you the job id\u002Flog path, and stop\n  polling. Default: monitor in chat.\n- How often should I post status? Default: every 5 minutes. Use 1-2 minutes for\n  smoke tests, 5 minutes for normal training, or 10-15 minutes for long runs.\n\nUse `long_running_enabled=true` and `status_interval_minutes=5` when the user\naccepts the defaults.\n\nWhen monitoring is enabled, do not send a final summary just because several\npolls have elapsed or the job is still `PENDING`. Keep the turn attached and\nemit status every `status_interval_minutes` until a terminal state or explicit\nuser stop\u002Fdetach request. If the runtime environment cannot keep the chat turn\nopen, say that clearly and leave a durable watcher\u002Flog path; do not imply that\nchat updates will continue after the turn ends.\n\nFinal-answer rule: a `final` response ends chat-side monitoring. While\n`long_running_enabled=true` and any launched job is non-terminal, status\nmessages must be sent as in-progress updates and the agent must continue\npolling. Only send a final response when the workflow reaches terminal state,\nthe user explicitly asks to detach\u002Fstop monitoring, or the runtime genuinely\ncannot keep the turn open; in that last case, say it is a runtime limitation\nand provide the exact durable status command\u002Flog path.\n\n## Missing-Input Prompt Shape\n\nWhen asking for launch inputs, include concrete examples and both dataset input\nmodes. Do not ask only for \"dataset root\".\n\nUse this structure and adapt spec keys to the selected model\u002Faction:\n\n```text\nI need these launch inputs before I can create specs or runner files:\n\n1. Execution platform: brev, slurm, local-docker, or kubernetes.\n\n2. Dataset inputs. You can provide either mode:\n   A) Root mode: give train\u002Feval roots and I map required files automatically.\n      Example Cosmos-RL:\n      train_root=\u002Flustre\u002Ffsw\u002F...\u002Fcosmos\u002Ftrain\n      -> custom.train_dataset.annotation_path=train_root\u002Fannotations.json\n      -> custom.train_dataset.media_path=train_root\n   B) Direct spec mode: give the exact config\u002Fspec parameters yourself.\n      Example:\n      custom.train_dataset.annotation_path=\u002Flustre\u002Ffsw\u002F...\u002Ftrain_annotations.json\n      custom.train_dataset.media_path=\u002Flustre\u002Ffsw\u002F...\u002Fvideos_train.tar.gz\n      custom.val_dataset.annotation_path=\u002Flustre\u002Ffsw\u002F...\u002Feval_annotations.json\n      custom.val_dataset.media_path=\u002Flustre\u002Ffsw\u002F...\u002Feval_videos\u002F\n\n   Platform examples:\n   - SLURM\u002FLustre: \u002Flustre\u002Ffsw\u002F...\u002Fdata\u002Ftrain or lustre:\u002F\u002F\u002Flustre\u002Ffsw\u002F...\u002Fdata\u002Ftrain\n   - Brev\u002FKubernetes: s3:\u002F\u002Fbucket\u002Fpath\u002Ftrain and s3:\u002F\u002Fbucket\u002Fpath\u002Feval\n   - local-docker: \u002Fdata\u002Ftao\u002F\u003Cmodel>\u002Ftrain or file:\u002F\u002F\u002Fdata\u002Ftao\u002F\u003Cmodel>\u002Feval\n\n3. Container image. I will resolve the default from packaged model metadata and\n   show it before launch, for example:\n   default image for \u003Cmodel>\u002F\u003Caction>: \u003Cresolved container image>\n   Use this image, or provide image=\u003Coverride> to pin a different TAO build.\n\n4. Compute shape required by the model, for example GPUs\u002Fnodes.\n\n5. Required credentials from platform\u002Fmodel docs, for example HF_TOKEN for\n   gated Hugging Face models.\n\n6. Monitoring preference. By default I monitor in this chat and post progress\n   every 5 minutes; choose 1-2 minutes for smoke tests or 10-15 minutes for\n   long training.\n```\n\n## Container Image Confirmation\n\nBefore creating specs, runner scripts, workspaces, logs, state files, or\nsubmitting a job, resolve the image for the selected model\u002Faction:\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Fresolve_tao_image.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} \\\n  --model \u003Cnetwork> --action \u003Caction> --format text\n```\n\nIf the helper is unavailable, read `skills\u002Fmodels\u002F\u003Cnetwork>\u002Fconfig.json` through\n`SkillBank().get_model_config(network_arch)`. Resolve image fields in this\norder:\n\n1. `actions.\u003Caction>.container_image`\n2. `actions.\u003Caction>.image`\n3. top-level `container_image`\n4. top-level `image`\n\nShow the exact image and ask:\n\n```text\nContainer image for \u003Cnetwork>\u002F\u003Caction>:\ndefault=\u003Cresolved image>\n\nUse this image, or provide image=\u003Coverride>?\n```\n\nIf the user accepts, pass the resolved image as the job `image`. If the user\noverrides, require a non-empty image reference and pass that value instead.\nDo not silently launch on the default image. This confirmation applies to\ntraining, AutoML recommendations, evaluation, inference, export, TensorRT\nengine generation, and application workflows that submit TAO containers.\n\n## Credential Filtering\n\nAfter the user chooses a platform, get the credential list for only that\nplatform:\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Flist_tao_platforms.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} \\\n  --platform \u003Cplatform> --format text\n```\n\nAsk only for credentials returned by that command, plus model-specific\ncredentials from the selected model skill. Do not ask for Brev credentials on\nSLURM, Kubernetes, or local Docker. Do not ask for SLURM credentials on Brev,\nKubernetes, or local Docker. Ask S3 credentials only when the selected\nplatform and the dataset\u002Fresult URIs require `s3:\u002F\u002F` access.\nCredentials may already be present in the process environment or in a\nuser-approved secret env file such as `~\u002F.tao\u002Fsecrets.env` or\n`~\u002F.config\u002Ftao\u002F.env`; source such files only when needed and never print,\ngrep, cat, paste, or log their contents. Verify only variable presence.\n\nFor initial launch intake, ask for required credentials and required credential\ngroups only. Treat the helper's optional credentials\u002Fsettings section as\nreference material; do not request those values unless their `only_when`\ncondition applies, the selected workflow cannot proceed without them, or the\nuser asks to customize that setting.\n\nWhen the helper output includes a \"Required credential groups\" section, satisfy\none credential from each group before proceeding. Explain each requested value\nusing the helper's description and \"How to get it\" text.\n\nFor SLURM, user-facing prompts should ask for `SSH_KEY_PATH` first. Mention\n`SSH_AUTH_SOCK` only if the user says they already use an SSH agent.\n\n## Dependency Remediation\n\nIf a required CLI\u002Flibrary is missing, say exactly what is missing and why it is\nneeded, then ask before installing. Examples:\n\n- S3 dataset or results path -> require an S3-capable client such as `aws`.\n- SDK-backed platform launch -> require the platform-specific\n  `nvidia-tao-sdk[...]` extra.\n- Local Docker SDK path -> require the Docker Python client and the configured\n  Docker network.\n\nAfter user approval and installation, rerun the same preflight. Do not create\nrunner files or launch jobs between the failed check and the rerun.\n\n## Dataset Intake\n\nAccept dataset inputs in either mode:\n\n- **Dataset root mode:** the user gives train\u002Feval\u002Fcalibration roots, and the\n  model skill maps required files by convention. Example for Cosmos-RL train:\n  `custom.train_dataset.annotation_path=\u003Croot>\u002Fannotations.json` and\n  `custom.train_dataset.media_path=\u003Croot>`.\n- **Direct spec mode:** the user gives exact spec-key paths when annotations,\n  media archives, videos, or image folders live in different places. Preserve\n  those keys directly, for example\n  `custom.train_dataset.annotation_path=\u002Flustre\u002F...\u002Ftrain_annotations.json`\n  and `custom.train_dataset.media_path=\u002Flustre\u002F...\u002Fvideos.tar.gz`.\n\nAsk for dataset examples that match the selected platform:\n\n- SLURM: shared cluster paths such as\n  `\u002Flustre\u002Ffsw\u002Fportfolios\u002F\u003Cteam>\u002F\u003Cyour-dir>\u002Fdata\u002F\u003Cmodel>\u002Ftrain` (where\n  `\u003Cyour-dir>` is your per-user directory on the cluster), or direct\n  spec paths under `\u002Flustre\u002F...`.\n- Brev, Kubernetes: usually `s3:\u002F\u002Fbucket\u002Fpath\u002Ftrain` and\n  `s3:\u002F\u002Fbucket\u002Fpath\u002Feval` unless the platform profile mounts shared storage.\n- Local Docker: local paths visible to the Docker host, such as\n  `\u002Fdata\u002Ftao\u002F\u003Cmodel>\u002Ftrain`, or direct spec paths visible inside the planned\n  container mount.\n- Remote Docker: absolute paths visible on the remote Docker host named by\n  `DOCKER_HOST`, not paths on the local agent machine.\n\nDo not assume \"dataset root\" is the only acceptable input. When direct spec\npaths are supplied, validate the exact spec paths rather than appending default\nfilenames.\n\n## Platform Preflight\n\nRun the selected platform's preflight checks before any launch artifact is\ncreated.\n\nPrefer the packaged preflight helper when the needed inputs are available:\n\n```bash\n${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external}\u002Fscripts\u002Fcheck_tao_launch_preflight.py \\\n  --skill-bank ${TAO_SKILL_BANK_PATH:-~\u002Ftao-skills-external} \\\n  --platform \u003Cplatform> \\\n  --container-image \u003Cselected-image> \\\n  --path train_annotation=\u003Cpath> \\\n  --path train_media=\u003Cpath>\n```\n\nPass exact direct spec paths when the user supplied them. For root-mode inputs,\nexpand model-required files first, then pass those concrete annotation\u002Fmedia\npaths to the helper.\n\nIf the helper reports a missing client tool such as `aws` for `s3:\u002F\u002F` path\nverification, install the smallest needed package after user approval, then\nrerun the same command with `--install-missing-tools` and do not proceed until\nthe rerun verifies the paths.\n\nWhen the selected model skill warns that large S3 media should be staged, copy\nor extract the data once to platform-visible storage before creating launch\nartifacts, then validate those staged paths with the same preflight helper.\nRecord the source URI and staged path in the run workspace so AutoML summaries\ncan distinguish data staging time from training\u002Fevaluation time.\n\nFor `local-docker` and `remote-docker`, always pass the selected image with\n`--container-image` after resolving `container_image` from\n`skill_info.yaml`\u002F`versions.yaml`. The helper verifies Docker reachability,\nNVIDIA Container Toolkit registration, GPU memory, selected-image architecture\ncompatibility when known, and a GPU-visible smoke container before launch. For\n`remote-docker`, pass `--docker-host` or export `DOCKER_HOST`; the helper queries\nGPUs and validates bind-mounted paths through the remote daemon instead of using\nlocal host state. If the selected or smoke image is not present on the target\nDocker host, ask before pulling it or rerun with `--pull-smoke-image` after\napproval.\n\nWhen a model skill lists annotation-level required fields, pass them with\n`--json-required-field \u003Cpath-label>=\u003Cfield>[,\u003Cfield>...]` so schema\u002Fdata\ncontent issues fail during preflight rather than inside the first training\ncontainer. Do not add required annotation fields from old failure history; only\nenforce fields documented as required by the current model skill.\nFor local JSON\u002FJSONL annotation paths, the helper prints `records=\u003CN>`; use the\ntrain annotation count as `automl_settings[\"train_sample_count\"]` for\nsample-count-sensitive AutoML runs before recommendations are generated.\nIf the model skill documents a run-local patch strategy for a missing required\nfield, create the patched copy in the current run workspace, update the spec\npaths to that copy, and rerun the content check before launch. Do not ask the\nuser to mutate source datasets unless the model skill says patching is\nimpossible.\n\nDo not use `--skip-platform-access` for a real launch. That flag is only for\ndry environment checks or for cases where the user has already provided explicit\nmanual proof of platform and storage access. If the helper cannot verify remote\nAPI, CLI, cluster, or object-store access, treat preflight as failed and do not\ngenerate launch artifacts.\n\nFor SLURM:\n\n1. Require `SLURM_USER`, `SLURM_HOSTNAME`, a partition intent, and one of\n   `SSH_KEY_PATH` or `SSH_AUTH_SOCK`. If the user says to use the cluster\n   default partition, pass an empty partition\u002Fomit the partition directive; do\n   not substitute a site-specific value such as `batch`.\n   Use the selected platform helper's `Resource defaults` for runtime values.\n   For the packaged SLURM defaults, generate launchers with\n   `SLURM_TIME_HOURS=4` and `SLURM_TIMEOUT_HOURS=3.8`; never invent a\n   12-hour default for the 4-hour partition list.\n   Launching the orchestrator with `nohup` or in the background is allowed for\n   durability, but it does not satisfy chat monitoring by itself. After launch,\n   keep a foreground chat-side polling loop attached until terminal state or\n   explicit detach.\n2. Split comma-separated `SLURM_HOSTNAME`, resolve hosts where possible, and\n   require passwordless `ssh -o BatchMode=yes` to at least one host.\n3. If SSH fails, do not offer several equivalent choices. Ask for\n   `SSH_KEY_PATH=\u002Fpath\u002Fto\u002Fprivate_key` and show the passwordless setup steps:\n   create a key if needed with\n   `ssh-keygen -t ed25519 -N \"\" -f ~\u002F.ssh\u002Fid_ed25519`; install it with\n   `ssh-copy-id -i ~\u002F.ssh\u002Fid_ed25519.pub \u003CSLURM_USER>@\u003Clogin-host>`; trust the\n   host with `ssh-keyscan -H \u003Clogin-host> >> ~\u002F.ssh\u002Fknown_hosts`; set\n   `chmod 600 ~\u002F.ssh\u002Fid_ed25519`; verify with\n   `ssh -o BatchMode=yes -i ~\u002F.ssh\u002Fid_ed25519 \u003CSLURM_USER>@\u003Clogin-host> 'hostname'`;\n   then rerun with `SSH_KEY_PATH=~\u002F.ssh\u002Fid_ed25519`.\n4. After SSH passes, validate dataset annotation\u002Fmedia paths on the remote login\n   host with `test -e` or an equivalent read-only command.\n5. Only then create runner scripts, specs, workspaces, or submit jobs.\n6. For multi-GPU Slurm jobs, rely on the SDK Slurm backend to request\n   `--gpus-per-node=\u003CN>`. Do not generate manual `--gpus=\u003CN>` sbatch snippets;\n   that can spread GPUs across nodes and leave allocated GPUs idle.\n7. For full-matrix or multi-node launches, submit one smoke job first. Launch\n   the full matrix only after the smoke reaches training, emits the requested\n   metric\u002Fstatus record, and shows expected GPU utilization.\n\nFor AutoML status, prefer structured controller\u002Fbrain state and job metadata\n(`active_jobs.json`, `.automl\u002Fcontroller\u002F*.json`, result JSON, and\n`results_dir\u002Ftrain\u002Fstatus.json`) before scanning raw logs. Parse logs only as a\nfallback or when the user specifically asks for log-level investigation.\n\nFor local Docker, validate Docker\u002FGPU access and local dataset paths before\nwriting launch artifacts. For Brev and Kubernetes, validate API or\ncluster access plus object-storage credentials and `aws s3 ls` readability for\n`s3:\u002F\u002F` inputs before writing launch artifacts. For mounted shared-storage or\nPVC paths on those remote platforms, require manual proof that the path is\nmounted into the job environment; the helper fails closed rather than accepting\nunverified remote mount paths.\n\n## Runtime And Configuration Review\n\nBefore any side-effecting launch, show a concise review:\n\n- selected platform and exact container image\n- GPU ids\u002Fcount and nodes, including any GPUs avoided because they are already\n  occupied\n- dataset roots or direct spec paths, with sample counts when available\n- important model\u002Fworkflow overrides that differ from template defaults\n- estimated runtime and the assumptions behind it\n- monitoring interval and whether chat-side monitoring will stay attached\n\nFor AutoML, also show the algorithm, metric\u002Fdirection, recommendation budget,\nsearch parameters, ranges, and generated\u002Fdefault recommendation details as\ndescribed in `skills\u002Fapplications\u002Ftao-run-automl\u002FSKILL.md`. Ask for confirmation after\nthis review. If the user supplied a time limit, flag any plan that exceeds it\nand offer concrete reductions before launch.\n",{"name":5,"tags":24,"license":15,"metadata":28,"description":7,"allowed-tools":31,"compatibility":32},[25,26,27],"tao","workflow","launch",{"author":29,"version":30},"NVIDIA Corporation","0.1.0","Read Bash","Requires the packaged TAO skill bank helper scripts.","main","86cc27ffd16af6cb85948f1155578e6f3f61cc4c","skillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn","https:\u002F\u002Fgithub.com\u002FSolizardking","https:\u002F\u002Fmusebook.trade","Got a mile high and never came down","2026-10-08 11:52:53",{"source":8,"skillName":5,"commands":42,"prompt":52,"zipAvailable":53},[43,46,49],{"tool":44,"command":45},"npx","npx skills add Solizardking\u002Ftrain2earn --skill tao-launch-workflow",{"tool":47,"command":48},"bunx","bunx skills add Solizardking\u002Ftrain2earn --skill tao-launch-workflow",{"tool":50,"command":51},"pnpm","pnpm dlx skills add Solizardking\u002Ftrain2earn --skill tao-launch-workflow","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow\u002FSKILL.md",true,[55,61,65,69],{"path":56,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":59,"sourceHtmlUrl":60},"BENCHMARK.md","REFERENCE","参考文档",3952,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow\u002FBENCHMARK.md",{"path":62,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":63,"sourceHtmlUrl":64},"metadata.json",416,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow\u002Fmetadata.json",{"path":66,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":67,"sourceHtmlUrl":68},"skill-card.md",3807,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow\u002Fskill-card.md",{"path":70,"resourceType":57,"resourceTypeDesc":58,"sizeBytes":71,"sourceHtmlUrl":72},"verification.json",1404,"https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-launch-workflow\u002Fverification.json",[74,82,90,98,105,113],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M78B94RSCAKMH1QS79VE6DG9","testing","For testing with Light Protocol programs and clients on localnet, devnet, and mainnet validation.","MIT","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Ftesting\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":86,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M78B81BKY8RX12HETPG8YKWK","qedgen","Formally verify programs by writing Lean 4 proofs. Trigger this skill whenever the user wants to formally verify code, generate Lean 4 proofs, prove properties about algorithms or smart contracts, verify invariants, convert program logic into formal specifications, or anything involving Lean 4 and formal verification. Also trigger when the user mentions \"qedgen\", \"lean proof\", \"formal proof\", \"verify my code\", \"prove correctness\", \"formal verification\", or wants mathematical guarantees about their implementation.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fsolana-formal-verification\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fsolana-formal-verification\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":78,"licenseFileUrl":94,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M78B7Q7C5K54E01DN83A60KG","solana-dev","Use when user asks to \"build a Solana dapp\", \"write an Anchor program\", \"create a token\", \"debug Solana errors\", \"set up wallet connection\", \"test my Solana program\", \"deploy to devnet\", or \"explain Solana concepts\" (rent, accounts, PDAs, CPIs, etc.). Also use for quick on-chain lookups via public RPC + curl — \"what's the balance of \u003Cwallet>\", \"look up transaction \u003Csig>\", \"token balance for \u003Caccount>\", \"check this address on mainnet\u002Fdevnet\". End-to-end Solana development playbook covering wallet connection, Anchor\u002FPinocchio programs, Codama client generation, LiteSVM\u002FMollusk\u002FSurfpool testing, security checklists, and JSON-RPC curl lookups against public clusters. Integrates with the Solana MCP server for live documentation search. Prefers framework-kit (@solana\u002Fclient + @solana\u002Freact-hooks) for UI, wallet-standard-first connection (incl. ConnectorKit), @solana\u002Fkit for client\u002FRPC code, and @solana\u002Fweb3-compat for legacy boundaries.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fsolana-dev\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fsolana-dev\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M78B7HXD6N6GWKSVRQ8F6Q31","solana-common-errors","Diagnose and fix common Solana development errors — GLIBC mismatches, Anchor version conflicts, cargo build-sbf \u002F platform-tools failures, LiteSVM issues, edition2024 crate pins, RPC airdrops, and Anchor 0.29→0.32 migrations. Use when builds fail, Anchor CLI won't install, or CI breaks on Debian\u002FUbuntu.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fsolana-common-errors\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":109,"sourceHtmlUrl":110,"categories":111,"keywords":112,"publishedAt":19},"01M78B6T7YSYWQ0N4DP40QDXEZ","skill-creator","Create or update AgentSkills. Use when designing, structuring, or packaging skills with scripts, references, and assets.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fskill-creator\u002Flicense.txt","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fskill-creator\u002FSKILL.md",[],[],{"skillNo":114,"name":115,"nameZh":6,"description":116,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":78,"licenseFileUrl":6,"sourceHtmlUrl":117,"categories":118,"keywords":119,"publishedAt":19},"01M78B1VFT9X5AND7AE1HZPZGF","phantom-wallet-mcp","Execute wallet operations through the Phantom MCP server. Use when the user wants to interact with their Phantom wallet directly — get addresses, transfer SOL or SPL tokens, buy\u002Fswap tokens, sign transactions, and sign messages across Solana, Ethereum, Bitcoin, and Sui. Requires the @phantom\u002Fmcp-server to be configured as an MCP server.","https:\u002F\u002Fgithub.com\u002FSolizardking\u002Ftrain2earn\u002Fblob\u002Fmain\u002Fskillhub\u002Fskills\u002Fphantom-wallet-mcp\u002FSKILL.md",[],[]]