[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M733960ZH7Z29KF1GES9QCSR":3,"skill-same:01M733960ZH7Z29KF1GES9QCSR":44},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M733960ZH7Z29KF1GES9QCSR","hive-capability-search",null,"Discover, rank, and return a bounded capability map across HivemindOS shared skills, packaged skills, runtime skills, tool schemas, API routes, connected apps\u002Fendpoints, runtime capabilities, workspace context, existing specialty agents, and safe credential-presence signals. Use whenever an agent must choose which available tools, skills, apps, agents, or delivery channels can satisfy a task, especially before hive skill fusion, hive workflow fusion, hive AEON fusion, or any adaptive multi-step run.","LiamVisionary\u002Fhivemindos","LiamVisionary","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F102138294?v=4","01M73379NZZ99TT4ZS17XR3Q9D","01M73379PX6AKQNWN05VNSYRXW",13,"MIT","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fpackaged-skills\u002Fauto-install\u002Fhive-capability-search\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: hive-capability-search\ndescription: Discover, rank, and return a bounded capability map across HivemindOS shared skills, packaged skills, runtime skills, tool schemas, API routes, connected apps\u002Fendpoints, runtime capabilities, workspace context, existing specialty agents, and safe credential-presence signals. Use whenever an agent must choose which available tools, skills, apps, agents, or delivery channels can satisfy a task, especially before hive skill fusion, hive workflow fusion, hive AEON fusion, or any adaptive multi-step run.\n---\n\n# Hive Capability Search\n\nUse this skill to answer: \"What can this hive currently do for this task, and which parts should the agent use?\"\n\nReturn a capability map. Do not execute side effects unless a calling skill explicitly asks for a status check or no-side-effect rehearsal.\n\n## Search Boundary\n\nSearch only surfaces available to the current agent\u002Fruntime:\n\n- Shared brain skill index: read `Skills\u002FREADME.md` at most once, then load only directly relevant `Skills\u002F\u003Cslug>\u002FSKILL.md` files.\n- Packaged auto-install skills, installed optional skills, and optional packaged catalog metadata exposed through the context index as installable workflow playbooks. Optional catalog hits are not active runtime skills until the user installs them into the shared brain.\n- Runtime\u002Fprovider skills, tool schemas, slash commands, app schemas, API routes, and local CLI capabilities exposed to the agent.\n- Connected apps\u002Fendpoints through the app discovery\u002Fcatalog surfaces and app request proxy. Do not hard-code Tailnet endpoints, private IPs, local machine names, or transient URLs.\n- Existing specialty agents and agent subclasses when the runtime exposes them.\n- Workspace docs\u002Ffiles only when task-relevant.\n- Shared hive env credential presence by key name only. Use approved presence\u002Fstatus checks such as `hive-env-check KEY`; never read, print, copy, summarize, or persist secret values.\n\n## Discovery Surfaces\n\nUse the best surface available in the current runtime; do not assume every agent has shell access.\n\n1. If the runtime already injected \"Hive capability search\" hits or context-index evidence, consume that first and avoid rerunning discovery.\n2. If an authenticated dashboard\u002Fapp\u002Fphone bridge, MCP tool, or route tool exposes `\u002Fapi\u002Fcontext-index`, call it with the task query. This is the preferred no-shell path, especially for phone-hosted or app-routed agents.\n3. If shell access exists, run `hive-capability-search \"\u003Ctask>\"`. Use `--json` for machine-readable output, and `--live` only when fresh connected-app\u002Fendpoints matter and dashboard auth is available.\n4. If none of those surfaces are available, build the map from visible shared skills, runtime tool schemas, app context, and current workspace evidence, then mark `gaps` with the missing capability-search surface instead of claiming the route was verified.\n\nThe CLI is a delivery channel for this skill, not the skill itself. A missing CLI is a gap only for shell-based agents; it is not a failure when another verified context-index or injected-context surface is available.\n\n## Bounded Retrieval\n\n- Start with one broad query using the user's full goal.\n- Add at most one targeted query per capability intent, such as research, writing, image generation, code execution, messaging, scheduling, approvals, wallet\u002Fpayment, data source, app action, or deployment.\n- Use at most six targeted `\u002Fapi\u002Fcontext-index` queries by default. If the task has more intents, combine adjacent intents or state that deeper search is needed.\n- Stop if the same search\u002Fread step would repeat. Move to synthesis with the evidence already collected.\n- If retrieval is unavailable, stale, or inconsistent, state the gap and produce the best map from observed evidence.\n\n## Ranking Rules\n\nRank candidate parts by:\n\n1. Task fit and exact capability match.\n2. Confirmed availability in the current runtime or shared brain.\n3. Safety and side-effect controls.\n4. User preference or configured default. Connected apps can carry user routing preferences: a priority pin, prose usage notes such as \"use this app for anime images\", and preferred models per task type. When a usage note matches the requested style or task, that app wins over name-pattern matches, and its preferred model for the task type should be passed in the generation request.\n5. Cost policy, including whether paid fallback is configured or allowed.\n6. Quality and freshness for current-information tasks.\n7. Worker-class fit: hits marked class-preferred or agent task preference match the active agent's specialization; prefer them on ties. The class is a prior, not a gate — any listed capability remains usable, and strongly mismatched work should be routed back through Queen Bee instead of ground through with weak priors.\n8. Simplicity: prefer one capable specialist when it truly covers the job; otherwise compose specialists.\n\nAsk a short clarification only when multiple viable choices materially affect output quality, identity, cost, or external side effects and no configured default makes the choice obvious. If exactly one viable channel\u002Ftool exists, infer it and continue.\n\n## Output Format\n\nReturn a compact capability map:\n\n- `task`: the user's goal in one sentence\n- `intents`: atomic jobs such as gather, transform, create, verify, deliver, schedule, monitor\n- `selected`: one selected component per intent, with proof of availability\n- `alternates`: viable fallback components and when to use them\n- `agents`: specialty agents or subclasses that should own work, if any\n- `apps`: discovered app\u002Fendpoints selected through catalog logic, not hard-coded endpoints\n- `delivery`: slash commands, API routes, MCP tools, CLIs, or dashboard surfaces that should carry the selected capability when relevant\n- `credentials`: required env keys or status checks by name only, never values\n- `side_effects`: publish\u002Fsend\u002Fpay\u002Fdeploy\u002Fmutate actions and required approval gates\n- `gaps`: missing, ambiguous, or unavailable components\n- `questions`: only the clarifications that are actually blocking or materially consequential\n\nIf the selected component is a remote connected app, remote collector, Hivemind Link endpoint, or fleet machine capability, include enough handoff detail for `hive-remote-capability-use`: app or service name, machine identity, fresh catalog URL fields when available, required route\u002Fschema endpoints, transfer needs, verification expectations, and side-effect gates.\n\n## Build-Task Approval Handoff\n\nFor a user-requested build or implementation task, classify every selected component as either `available` or `setup-required` and include the strongest alternatives. Capability approval governs selection, installation, and ordinary setup only; it never replaces spend, secret, deploy, destructive-action, payment, or external-send approvals.\n\n- In the HivemindOS Chat route, return the structured capability map to the chat approval surface and pause. Setup-required items default to approve; available items expose Browse so the user can select an alternative, supply a GitHub repository, or attach an instruction. The user may remove an entire capability intent; when they do, redesign the task so that whole step and its output are absent while the remaining task stays coherent. Continue only after the submitted plan returns to the agent.\n- In Codex, Claude Code, terminals, and other chat surfaces without dynamic controls, present the same information as a compact natural-language list. Mark each item `available` or `setup required`, state the proposed default, include material alternatives, then end with one final question: \"I've drafted the capability list. Is this okay, and may I continue?\" Wait for the answer before capability setup or task execution.\n- In autonomous Work Board and Zero Human Company runs, select and set up capabilities automatically by default so the work does not pause. If the task's or company's capability policy is `ask`, use its normal Needs You \u002F `ACTION NEEDED:` path, present the capability list in natural language, and wait. Removing a capability still removes that task step; rejecting setup means use an available route or omit only the unsupported output when no coherent available route exists.\n\n## Verification\n\nFor fusion callers, include enough proof that the selected parts are reachable:\n\n- Retrieval hit, skill path, slash-command metadata, app catalog entry, route\u002Ftool schema, runtime capability, agent profile, or status endpoint.\n- For media generation capabilities, require later artifact verification before claiming success.\n- For external delivery capabilities, require later provider\u002Ftool receipts such as `success: true`, `message_id`, URL, transaction id, post id, or equivalent.\n- For remote connected apps or fleet capabilities, load `hive-remote-capability-use` before execution so the agent refreshes discovery, avoids stale Tailnet URLs, transfers sensitive local inputs privately, and verifies receipts.\n","# Hive Capability Search\n\nUse this skill to answer: \"What can this hive currently do for this task, and which parts should the agent use?\"\n\nReturn a capability map. Do not execute side effects unless a calling skill explicitly asks for a status check or no-side-effect rehearsal.\n\n## Search Boundary\n\nSearch only surfaces available to the current agent\u002Fruntime:\n\n- Shared brain skill index: read `Skills\u002FREADME.md` at most once, then load only directly relevant `Skills\u002F\u003Cslug>\u002FSKILL.md` files.\n- Packaged auto-install skills, installed optional skills, and optional packaged catalog metadata exposed through the context index as installable workflow playbooks. Optional catalog hits are not active runtime skills until the user installs them into the shared brain.\n- Runtime\u002Fprovider skills, tool schemas, slash commands, app schemas, API routes, and local CLI capabilities exposed to the agent.\n- Connected apps\u002Fendpoints through the app discovery\u002Fcatalog surfaces and app request proxy. Do not hard-code Tailnet endpoints, private IPs, local machine names, or transient URLs.\n- Existing specialty agents and agent subclasses when the runtime exposes them.\n- Workspace docs\u002Ffiles only when task-relevant.\n- Shared hive env credential presence by key name only. Use approved presence\u002Fstatus checks such as `hive-env-check KEY`; never read, print, copy, summarize, or persist secret values.\n\n## Discovery Surfaces\n\nUse the best surface available in the current runtime; do not assume every agent has shell access.\n\n1. If the runtime already injected \"Hive capability search\" hits or context-index evidence, consume that first and avoid rerunning discovery.\n2. If an authenticated dashboard\u002Fapp\u002Fphone bridge, MCP tool, or route tool exposes `\u002Fapi\u002Fcontext-index`, call it with the task query. This is the preferred no-shell path, especially for phone-hosted or app-routed agents.\n3. If shell access exists, run `hive-capability-search \"\u003Ctask>\"`. Use `--json` for machine-readable output, and `--live` only when fresh connected-app\u002Fendpoints matter and dashboard auth is available.\n4. If none of those surfaces are available, build the map from visible shared skills, runtime tool schemas, app context, and current workspace evidence, then mark `gaps` with the missing capability-search surface instead of claiming the route was verified.\n\nThe CLI is a delivery channel for this skill, not the skill itself. A missing CLI is a gap only for shell-based agents; it is not a failure when another verified context-index or injected-context surface is available.\n\n## Bounded Retrieval\n\n- Start with one broad query using the user's full goal.\n- Add at most one targeted query per capability intent, such as research, writing, image generation, code execution, messaging, scheduling, approvals, wallet\u002Fpayment, data source, app action, or deployment.\n- Use at most six targeted `\u002Fapi\u002Fcontext-index` queries by default. If the task has more intents, combine adjacent intents or state that deeper search is needed.\n- Stop if the same search\u002Fread step would repeat. Move to synthesis with the evidence already collected.\n- If retrieval is unavailable, stale, or inconsistent, state the gap and produce the best map from observed evidence.\n\n## Ranking Rules\n\nRank candidate parts by:\n\n1. Task fit and exact capability match.\n2. Confirmed availability in the current runtime or shared brain.\n3. Safety and side-effect controls.\n4. User preference or configured default. Connected apps can carry user routing preferences: a priority pin, prose usage notes such as \"use this app for anime images\", and preferred models per task type. When a usage note matches the requested style or task, that app wins over name-pattern matches, and its preferred model for the task type should be passed in the generation request.\n5. Cost policy, including whether paid fallback is configured or allowed.\n6. Quality and freshness for current-information tasks.\n7. Worker-class fit: hits marked class-preferred or agent task preference match the active agent's specialization; prefer them on ties. The class is a prior, not a gate — any listed capability remains usable, and strongly mismatched work should be routed back through Queen Bee instead of ground through with weak priors.\n8. Simplicity: prefer one capable specialist when it truly covers the job; otherwise compose specialists.\n\nAsk a short clarification only when multiple viable choices materially affect output quality, identity, cost, or external side effects and no configured default makes the choice obvious. If exactly one viable channel\u002Ftool exists, infer it and continue.\n\n## Output Format\n\nReturn a compact capability map:\n\n- `task`: the user's goal in one sentence\n- `intents`: atomic jobs such as gather, transform, create, verify, deliver, schedule, monitor\n- `selected`: one selected component per intent, with proof of availability\n- `alternates`: viable fallback components and when to use them\n- `agents`: specialty agents or subclasses that should own work, if any\n- `apps`: discovered app\u002Fendpoints selected through catalog logic, not hard-coded endpoints\n- `delivery`: slash commands, API routes, MCP tools, CLIs, or dashboard surfaces that should carry the selected capability when relevant\n- `credentials`: required env keys or status checks by name only, never values\n- `side_effects`: publish\u002Fsend\u002Fpay\u002Fdeploy\u002Fmutate actions and required approval gates\n- `gaps`: missing, ambiguous, or unavailable components\n- `questions`: only the clarifications that are actually blocking or materially consequential\n\nIf the selected component is a remote connected app, remote collector, Hivemind Link endpoint, or fleet machine capability, include enough handoff detail for `hive-remote-capability-use`: app or service name, machine identity, fresh catalog URL fields when available, required route\u002Fschema endpoints, transfer needs, verification expectations, and side-effect gates.\n\n## Build-Task Approval Handoff\n\nFor a user-requested build or implementation task, classify every selected component as either `available` or `setup-required` and include the strongest alternatives. Capability approval governs selection, installation, and ordinary setup only; it never replaces spend, secret, deploy, destructive-action, payment, or external-send approvals.\n\n- In the HivemindOS Chat route, return the structured capability map to the chat approval surface and pause. Setup-required items default to approve; available items expose Browse so the user can select an alternative, supply a GitHub repository, or attach an instruction. The user may remove an entire capability intent; when they do, redesign the task so that whole step and its output are absent while the remaining task stays coherent. Continue only after the submitted plan returns to the agent.\n- In Codex, Claude Code, terminals, and other chat surfaces without dynamic controls, present the same information as a compact natural-language list. Mark each item `available` or `setup required`, state the proposed default, include material alternatives, then end with one final question: \"I've drafted the capability list. Is this okay, and may I continue?\" Wait for the answer before capability setup or task execution.\n- In autonomous Work Board and Zero Human Company runs, select and set up capabilities automatically by default so the work does not pause. If the task's or company's capability policy is `ask`, use its normal Needs You \u002F `ACTION NEEDED:` path, present the capability list in natural language, and wait. Removing a capability still removes that task step; rejecting setup means use an available route or omit only the unsupported output when no coherent available route exists.\n\n## Verification\n\nFor fusion callers, include enough proof that the selected parts are reachable:\n\n- Retrieval hit, skill path, slash-command metadata, app catalog entry, route\u002Ftool schema, runtime capability, agent profile, or status endpoint.\n- For media generation capabilities, require later artifact verification before claiming success.\n- For external delivery capabilities, require later provider\u002Ftool receipts such as `success: true`, `message_id`, URL, transaction id, post id, or equivalent.\n- For remote connected apps or fleet capabilities, load `hive-remote-capability-use` before execution so the agent refreshes discovery, avoids stale Tailnet URLs, transfers sensitive local inputs privately, and verifies receipts.\n",{"name":5,"description":7},"main","f270c4ae4f9d09c4e5da9cfa8adab7e89ae1f53e","packaged-skills\u002Fauto-install\u002Fhive-capability-search",4,"https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos","https:\u002F\u002Fgithub.com\u002FLiamVisionary","2026-11-10 21:48:41",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add LiamVisionary\u002Fhivemindos --skill hive-capability-search",{"tool":36,"command":37},"bunx","bunx skills add LiamVisionary\u002Fhivemindos --skill hive-capability-search",{"tool":39,"command":40},"pnpm","pnpm dlx skills add LiamVisionary\u002Fhivemindos --skill hive-capability-search","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fpackaged-skills\u002Fauto-install\u002Fhive-capability-search\u002FSKILL.md",true,[],[45,52,59,66,73,80],{"skillNo":46,"name":47,"nameZh":6,"description":48,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":49,"categories":50,"keywords":51,"publishedAt":18},"01M734AA85KQEXDQXQJ3T7769M","tailscale-generation-skill-authoring","This skill should be used when the user asks to \"create a Tailscale generation skill\", \"add a local generation skill\", \"document a generation endpoint\", \"add an image generation provider\", \"add a TTS provider\", \"add video generation\", \"add 3D generation\", or needs guidance on where HivemindOS Tailscale or local generation providers belong in the shared skills shelf.","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fskills\u002Ftailscale-generation-skill-authoring\u002FSKILL.md",[],[],{"skillNo":53,"name":54,"nameZh":6,"description":55,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"publishedAt":18},"01M734A7VD69PRA0C0EV5EFY9T","localtts","This skill should be used when the user asks to \"generate local TTS\", \"use local text to speech\", \"use remote TTS over Tailscale\", \"make speech audio\", \"add a TTS provider\", \"use Qwen3-TTS\", or needs TTS endpoint setup through local or HivemindOS Tailscale routes.","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fskills\u002Flocaltts\u002FSKILL.md",[],[],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":18},"01M734A4K2Z0F74QZP7V7WBFRY","karpathy-guidelines","Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fskills\u002Fkarpathy-guidelines\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":18},"01M734A1HMVYVJX3YJGNQ6KF6C","comfyui-image-generation","This skill should be used when the user asks to \"generate images\", \"generate images with ComfyUI\", \"use local ComfyUI\", \"use remote ComfyUI\", \"use Z-Image\", \"add an image generation provider\", or needs image-generation endpoint setup through local or HivemindOS Tailscale routes.","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fskills\u002Fcomfyui-image-generation\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":18},"01M7349Y684Q676DNNHZPSC8W1","wrangler","Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fskills\u002Fcloudflare\u002Fwrangler\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M7349V3WCBF835QP74SA9Q8C","workers-best-practices","Reviews and authors Cloudflare Workers code against production best practices. Load when writing new Workers, reviewing Worker code, configuring wrangler.jsonc, or checking for common Workers anti-patterns (streaming, floating promises, global state, secrets, bindings, observability). Biases towards retrieval from Cloudflare docs over pre-trained knowledge.","https:\u002F\u002Fgithub.com\u002FLiamVisionary\u002Fhivemindos\u002Fblob\u002Fmain\u002Fskills\u002Fcloudflare\u002Fworkers-best-practices\u002FSKILL.md",[],[]]