[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EWWEJKFWX5M9PJS7EEJ5VP":3,"skill-same:01M7EWWEJKFWX5M9PJS7EEJ5VP":133},{"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":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":6,"updatedAt":32,"install":33,"resources":46},"01M7EWWEJKFWX5M9PJS7EEJ5VP","headroom",null,"Token-reduction skill for coding agents. Compresses JSON tool output, build logs, grep\u002Fripgrep search results, source code, and conversation history the same way the Headroom proxy does — but in-context, with no binary install. Use this skill whenever an agent is about to paste a large tool result (a 1000-row JSON array, a 5000-line pytest log, a 200-result grep, a 1000-line source file) into its next message, or when the agent's conversation is approaching the context limit, or when the user complains about token cost or latency. Also use it to shape the agent's own output: skip ceremony, don't restate code already on screen, drop reasoning on routine turns. Triggers: \"compress this output\", \"this is too long\", \"trim\", \"summarize the log\", \"reduce tokens\", \"context is full\", \"running out of context\", \"too many tokens\", \"make it shorter\", \"be terse\".","TanNhatCMS\u002Fjarvis","TanNhatCMS","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F113796420?v=4","01M753MV17KG5CD3152G3WQEQY","KietNT","01M7EKEDYVYQ6PWVJJMZ911VMX",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: headroom\ndescription: >\n  Token-reduction skill for coding agents. Compresses JSON tool output, build\n  logs, grep\u002Fripgrep search results, source code, and conversation history the\n  same way the Headroom proxy does — but in-context, with no binary install.\n  Use this skill whenever an agent is about to paste a large tool result (a\n  1000-row JSON array, a 5000-line pytest log, a 200-result grep, a 1000-line\n  source file) into its next message, or when the agent's conversation is\n  approaching the context limit, or when the user complains about token cost or\n  latency. Also use it to shape the agent's own output: skip ceremony, don't\n  restate code already on screen, drop reasoning on routine turns. Triggers:\n  \"compress this output\", \"this is too long\", \"trim\", \"summarize the log\",\n  \"reduce tokens\", \"context is full\", \"running out of context\", \"too many\n  tokens\", \"make it shorter\", \"be terse\".\n---\n\n# Headroom — portable token-reduction skill\n\nThis skill teaches any coding agent to apply the Headroom compression\ntechniques in-context. It does **not** require the Headroom binary.\n\n## When to apply which transform\n\nPick the transform by content type. If a tool result is shorter than ~200\ntokens, skip compression — the overhead isn't worth it.\n\n| Content the agent is about to use \u002F paste         | Apply                  | Helper script                          |\n|---------------------------------------------------|------------------------|----------------------------------------|\n| JSON array of dicts \u002F objects (tool results, API responses) | SmartCrusher           | `scripts\u002Fsmart_crusher.py`             |\n| Build \u002F test logs (pytest, npm, cargo, make)      | LogCompressor          | `scripts\u002Flog_compressor.py`            |\n| grep \u002F ripgrep \u002F ag output (`file:line:content`)  | SearchCompressor       | `scripts\u002Fsearch_compressor.py`         |\n| Source code (Python, JS, TS, Go, Rust, Java, C\u002FC++) | CodeCompressor       | `scripts\u002Fcode_compressor.py`           |\n| Conversation history over the context budget      | RollingWindow          | `scripts\u002Fcontext_manager.py`           |\n| Anything the model is about to **write back**     | Output Shaper (L2)     | `prompts\u002Fverbosity_steering.md`        |\n\n## The seven rules\n\nThese are the core rules the agent should internalise. Full detail in\n`AGENTS.md`; this is the short version.\n\n1. **Compress before paste.** Before pasting any tool result over ~200 tokens\n   into the next message, run the matching helper script (or apply the matching\n   heuristic by hand if no script is available). Cite the script in your\n   narration: \"Compressed 1,432 → 187 tokens with SmartCrusher (kept first 3,\n   last 2, all errors, top 5 by BM25 to query).\"\n2. **Never drop errors.** Errors, stack traces, `is_error: true` results, and\n   `FAILED` lines are sacred. Always preserve them 100%, even when compressing\n   hard.\n3. **Keep first N and last N.** For arrays, logs, and search results, always\n   keep the first 3 and last 2 items. They give pagination context and recency.\n4. **Stabilise the prefix.** When building prompts, keep the system prompt and\n   the first user turn byte-identical across requests. Move dynamic content\n   (timestamps, \"today is\", session IDs) to the *tail* of the prompt so the\n   provider's KV cache stays warm.\n5. **Drop oldest tool outputs first.** When the conversation exceeds the token\n   budget, drop complete `tool_call` + `tool_result` pairs from the oldest end.\n   Never split a pair. Never drop the system prompt. Never drop the last 2\n   user turns.\n6. **Shape your own output.** Default verbosity is L2: skip preambles, don't\n   restate code already on screen, reference by `path:line`. Drop to L3\n   (conclusions only) when the user asks for brevity. Never narrate tool\n   results the user can already see.\n7. **Be reversible.** When you compress a tool result, write the original to\n   `.headroom-cache\u002F\u003Csha256>.txt` and include the SHA in your narration: \"CCR\n   key: `a3f9…`\". If a later turn needs the original, retrieve it from there.\n\n## What \"compression\" means here\n\nCompression is **selection**, not zipping. We don't run gzip. We pick which\nitems \u002F lines \u002F nodes to keep, and emit a shorter representation that\npreserves the information the LLM needs to answer the user's question. The\noriginal is always retrievable from the CCR cache.\n\nThe selection heuristics, in priority order:\n\n1. **Errors** — always kept.\n2. **First N** — pagination \u002F setup context.\n3. **Last N** — recency.\n4. **Anomalies** — values >2σ from the mean (for numeric arrays) or items\n   whose fields differ sharply from their neighbours.\n5. **Relevant** — top-K by BM25 similarity to the user's current question.\n6. **Change points** — items where a field value transitions (status flips,\n   counters reset).\n7. **Sample** — a statistically representative sample of the remainder, so the\n   LLM still has a sense of the distribution.\n\n## Output-token shaping (the \"L2 default\")\n\nWhen generating your own reply, follow `prompts\u002Fverbosity_steering.md` at\nlevel 2 unless the user asks for more or less. The five levels:\n\n- **L0** — off\n- **L1** — skip ceremony (\"Great, let me…\")\n- **L2** — L1 + don't restate code\u002Foutput already on screen; reference by\n  `path:line` *(default)*\n- **L3** — L2 + conclusions only, skip reasoning unless asked\n- **L4** — fragments, minimum tokens\n\nThe \"effort routing\" companion rule: if the previous turn was a tool result\nwith `is_error: false` and no new user ask, this is a *mechanical\ncontinuation*. Reply at L3 and skip deep reasoning. If the previous turn was\nan error, or a new user question, keep full effort. See\n`prompts\u002Feffort_routing.md`.\n\n## Files in this skill\n\n- `AGENTS.md` — universal agent rules (works with any agent framework). Read\n  this first.\n- `CLAUDE.md` — Claude Code \u002F Cursor specific copy of the same rules.\n- `docs\u002Ftechniques.md` — deep dive into each transform, with worked examples.\n- `docs\u002Fintegrations.md` — per-agent setup (Hermes, Claw, Codex, Cursor,\n  Aider, Goose, Continue, OpenHands, vibe, Copilot CLI).\n- `docs\u002Freversible-ccr.md` — the on-disk CCR cache convention.\n- `scripts\u002F` — pure-Python stdlib helpers. No `pip install` required.\n- `prompts\u002F` — system-prompt fragments for output shaping.\n- `examples\u002F` — before\u002Fafter comparisons for each transform.\n\n## Important safety rules (do not violate)\n\nThese mirror the safety rules in upstream Headroom. Each prevents a concrete\nfailure mode.\n\n1. **Never drop user or assistant text.** Tool results can be compressed;\n   human-authored messages cannot.\n2. **Never split a tool_call \u002F tool_result pair.** Dropping one half orphans\n   the other and the model will hallucinate a result.\n3. **Parse failures are no-ops.** If the input doesn't parse as the expected\n   type, pass it through unchanged. A bad compression is worse than no\n   compression.\n4. **Always preserve recency.** The last N turns (default 2 user turns) are\n   never dropped, even if they have low importance scores.\n5. **Always preserve the system prompt.** Never.\n6. **100% error preservation.** No exceptions.\n7. **Idempotent.** Running the compressor on already-compressed output should\n   be a no-op (or close to it).\n\n## Credits\n\nThis skill is a portable re-implementation of the techniques in\n[chopratejas\u002Fheadroom](https:\u002F\u002Fgithub.com\u002Fchopratejas\u002Fheadroom) by Tejas Chopra\nand contributors, licensed Apache 2.0. See `NOTICE` and the \"Credits\" section\nof `README.md` for full attribution. The upstream project is faster, more\naccurate, and ML-backed; install it if you can.\n","# Headroom — portable token-reduction skill\n\nThis skill teaches any coding agent to apply the Headroom compression\ntechniques in-context. It does **not** require the Headroom binary.\n\n## When to apply which transform\n\nPick the transform by content type. If a tool result is shorter than ~200\ntokens, skip compression — the overhead isn't worth it.\n\n| Content the agent is about to use \u002F paste         | Apply                  | Helper script                          |\n|---------------------------------------------------|------------------------|----------------------------------------|\n| JSON array of dicts \u002F objects (tool results, API responses) | SmartCrusher           | `scripts\u002Fsmart_crusher.py`             |\n| Build \u002F test logs (pytest, npm, cargo, make)      | LogCompressor          | `scripts\u002Flog_compressor.py`            |\n| grep \u002F ripgrep \u002F ag output (`file:line:content`)  | SearchCompressor       | `scripts\u002Fsearch_compressor.py`         |\n| Source code (Python, JS, TS, Go, Rust, Java, C\u002FC++) | CodeCompressor       | `scripts\u002Fcode_compressor.py`           |\n| Conversation history over the context budget      | RollingWindow          | `scripts\u002Fcontext_manager.py`           |\n| Anything the model is about to **write back**     | Output Shaper (L2)     | `prompts\u002Fverbosity_steering.md`        |\n\n## The seven rules\n\nThese are the core rules the agent should internalise. Full detail in\n`AGENTS.md`; this is the short version.\n\n1. **Compress before paste.** Before pasting any tool result over ~200 tokens\n   into the next message, run the matching helper script (or apply the matching\n   heuristic by hand if no script is available). Cite the script in your\n   narration: \"Compressed 1,432 → 187 tokens with SmartCrusher (kept first 3,\n   last 2, all errors, top 5 by BM25 to query).\"\n2. **Never drop errors.** Errors, stack traces, `is_error: true` results, and\n   `FAILED` lines are sacred. Always preserve them 100%, even when compressing\n   hard.\n3. **Keep first N and last N.** For arrays, logs, and search results, always\n   keep the first 3 and last 2 items. They give pagination context and recency.\n4. **Stabilise the prefix.** When building prompts, keep the system prompt and\n   the first user turn byte-identical across requests. Move dynamic content\n   (timestamps, \"today is\", session IDs) to the *tail* of the prompt so the\n   provider's KV cache stays warm.\n5. **Drop oldest tool outputs first.** When the conversation exceeds the token\n   budget, drop complete `tool_call` + `tool_result` pairs from the oldest end.\n   Never split a pair. Never drop the system prompt. Never drop the last 2\n   user turns.\n6. **Shape your own output.** Default verbosity is L2: skip preambles, don't\n   restate code already on screen, reference by `path:line`. Drop to L3\n   (conclusions only) when the user asks for brevity. Never narrate tool\n   results the user can already see.\n7. **Be reversible.** When you compress a tool result, write the original to\n   `.headroom-cache\u002F\u003Csha256>.txt` and include the SHA in your narration: \"CCR\n   key: `a3f9…`\". If a later turn needs the original, retrieve it from there.\n\n## What \"compression\" means here\n\nCompression is **selection**, not zipping. We don't run gzip. We pick which\nitems \u002F lines \u002F nodes to keep, and emit a shorter representation that\npreserves the information the LLM needs to answer the user's question. The\noriginal is always retrievable from the CCR cache.\n\nThe selection heuristics, in priority order:\n\n1. **Errors** — always kept.\n2. **First N** — pagination \u002F setup context.\n3. **Last N** — recency.\n4. **Anomalies** — values >2σ from the mean (for numeric arrays) or items\n   whose fields differ sharply from their neighbours.\n5. **Relevant** — top-K by BM25 similarity to the user's current question.\n6. **Change points** — items where a field value transitions (status flips,\n   counters reset).\n7. **Sample** — a statistically representative sample of the remainder, so the\n   LLM still has a sense of the distribution.\n\n## Output-token shaping (the \"L2 default\")\n\nWhen generating your own reply, follow `prompts\u002Fverbosity_steering.md` at\nlevel 2 unless the user asks for more or less. The five levels:\n\n- **L0** — off\n- **L1** — skip ceremony (\"Great, let me…\")\n- **L2** — L1 + don't restate code\u002Foutput already on screen; reference by\n  `path:line` *(default)*\n- **L3** — L2 + conclusions only, skip reasoning unless asked\n- **L4** — fragments, minimum tokens\n\nThe \"effort routing\" companion rule: if the previous turn was a tool result\nwith `is_error: false` and no new user ask, this is a *mechanical\ncontinuation*. Reply at L3 and skip deep reasoning. If the previous turn was\nan error, or a new user question, keep full effort. See\n`prompts\u002Feffort_routing.md`.\n\n## Files in this skill\n\n- `AGENTS.md` — universal agent rules (works with any agent framework). Read\n  this first.\n- `CLAUDE.md` — Claude Code \u002F Cursor specific copy of the same rules.\n- `docs\u002Ftechniques.md` — deep dive into each transform, with worked examples.\n- `docs\u002Fintegrations.md` — per-agent setup (Hermes, Claw, Codex, Cursor,\n  Aider, Goose, Continue, OpenHands, vibe, Copilot CLI).\n- `docs\u002Freversible-ccr.md` — the on-disk CCR cache convention.\n- `scripts\u002F` — pure-Python stdlib helpers. No `pip install` required.\n- `prompts\u002F` — system-prompt fragments for output shaping.\n- `examples\u002F` — before\u002Fafter comparisons for each transform.\n\n## Important safety rules (do not violate)\n\nThese mirror the safety rules in upstream Headroom. Each prevents a concrete\nfailure mode.\n\n1. **Never drop user or assistant text.** Tool results can be compressed;\n   human-authored messages cannot.\n2. **Never split a tool_call \u002F tool_result pair.** Dropping one half orphans\n   the other and the model will hallucinate a result.\n3. **Parse failures are no-ops.** If the input doesn't parse as the expected\n   type, pass it through unchanged. A bad compression is worse than no\n   compression.\n4. **Always preserve recency.** The last N turns (default 2 user turns) are\n   never dropped, even if they have low importance scores.\n5. **Always preserve the system prompt.** Never.\n6. **100% error preservation.** No exceptions.\n7. **Idempotent.** Running the compressor on already-compressed output should\n   be a no-op (or close to it).\n\n## Credits\n\nThis skill is a portable re-implementation of the techniques in\n[chopratejas\u002Fheadroom](https:\u002F\u002Fgithub.com\u002Fchopratejas\u002Fheadroom) by Tejas Chopra\nand contributors, licensed Apache 2.0. See `NOTICE` and the \"Credits\" section\nof `README.md` for full attribution. The upstream project is faster, more\naccurate, and ML-backed; install it if you can.\n",{"name":5,"description":7},"main","af388a77c2fe5e0b4fd879cca4300a0ffa9a6579","skills\u002Fheadroom",1,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis","https:\u002F\u002Fgithub.com\u002FTanNhatCMS","https:\u002F\u002Fmrkiet.name.vn","2026-11-07 14:23:08",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add TanNhatCMS\u002Fjarvis --skill headroom",{"tool":39,"command":40},"bunx","bunx skills add TanNhatCMS\u002Fjarvis --skill headroom",{"tool":42,"command":43},"pnpm","pnpm dlx skills add TanNhatCMS\u002Fjarvis --skill headroom","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FSKILL.md",true,[47,53,59,63,67,71,75,79,83,87,91,95,99,103,107,113,117,121,125,129],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},".gitignore","OTHER","其他",244,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002F.gitignore",{"path":54,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":57,"sourceHtmlUrl":58},"AGENTS.md","REFERENCE","参考文档",13014,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FAGENTS.md",{"path":60,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":61,"sourceHtmlUrl":62},"CLAUDE.md",13009,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FCLAUDE.md",{"path":64,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":65,"sourceHtmlUrl":66},"CONTRIBUTING.md",2711,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FCONTRIBUTING.md",{"path":68,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":69,"sourceHtmlUrl":70},"docs\u002Fintegrations.md",7569,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fdocs\u002Fintegrations.md",{"path":72,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":73,"sourceHtmlUrl":74},"docs\u002Freversible-ccr.md",4261,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fdocs\u002Freversible-ccr.md",{"path":76,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":77,"sourceHtmlUrl":78},"docs\u002Ftechniques.md",16145,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fdocs\u002Ftechniques.md",{"path":80,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":81,"sourceHtmlUrl":82},"examples\u002Fbefore_after.md",10371,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fexamples\u002Fbefore_after.md",{"path":84,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":86,"sourceHtmlUrl":16},"LICENSE","许可证",10770,{"path":88,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":89,"sourceHtmlUrl":90},"NOTICE",857,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FNOTICE",{"path":92,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":93,"sourceHtmlUrl":94},"prompts\u002Fcache_alignment.md",5591,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fprompts\u002Fcache_alignment.md",{"path":96,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":97,"sourceHtmlUrl":98},"prompts\u002Feffort_routing.md",6012,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fprompts\u002Feffort_routing.md",{"path":100,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":101,"sourceHtmlUrl":102},"prompts\u002Fverbosity_steering.md",4984,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fprompts\u002Fverbosity_steering.md",{"path":104,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":105,"sourceHtmlUrl":106},"README.md",12019,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002FREADME.md",{"path":108,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":111,"sourceHtmlUrl":112},"scripts\u002F_common.py","SCRIPT","脚本",7888,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fscripts\u002F_common.py",{"path":114,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":115,"sourceHtmlUrl":116},"scripts\u002Fcode_compressor.py",12204,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fscripts\u002Fcode_compressor.py",{"path":118,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":119,"sourceHtmlUrl":120},"scripts\u002Fcontext_manager.py",11902,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fscripts\u002Fcontext_manager.py",{"path":122,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":123,"sourceHtmlUrl":124},"scripts\u002Flog_compressor.py",6206,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fscripts\u002Flog_compressor.py",{"path":126,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":127,"sourceHtmlUrl":128},"scripts\u002Fsearch_compressor.py",7094,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fscripts\u002Fsearch_compressor.py",{"path":130,"resourceType":109,"resourceTypeDesc":110,"sizeBytes":131,"sourceHtmlUrl":132},"scripts\u002Fsmart_crusher.py",10289,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fheadroom\u002Fscripts\u002Fsmart_crusher.py",[134,142,150,157,164,171],{"skillNo":135,"name":136,"nameZh":6,"description":137,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":138,"licenseFileUrl":6,"sourceHtmlUrl":139,"categories":140,"keywords":141,"publishedAt":20},"01M7EWYJR0J8DARY7E59Z6Y9MV","web-scrape","Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open\u002Ffetch fails on JS-heavy pages, or when you need clean markdown\u002Fstructured extraction from a site. Requires FIRECRAWL_API_KEY.","MIT","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fweb-scrape\u002FSKILL.md",[],[],{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":138,"licenseFileUrl":146,"sourceHtmlUrl":147,"categories":148,"keywords":149,"publishedAt":20},"01M7EWYBN5P6F2AJG3FR24T9BP","design-taste-frontend","tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML\u002FCSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, \"AI-made\", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Ftaste-design\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Ftaste-design\u002FSKILL.md",[],[],{"skillNo":151,"name":152,"nameZh":6,"description":153,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":138,"licenseFileUrl":6,"sourceHtmlUrl":154,"categories":155,"keywords":156,"publishedAt":20},"01M7EWY8FQFNQ2GB7NZZWGSSDX","test-driven-development","Use when implementing any feature or bugfix, before writing implementation code","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-tdd\u002FSKILL.md",[],[],{"skillNo":158,"name":159,"nameZh":6,"description":160,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":138,"licenseFileUrl":6,"sourceHtmlUrl":161,"categories":162,"keywords":163,"publishedAt":20},"01M7EWY5SHBFRH1HDHVWM9K7TV","systematic-debugging","Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-debugging\u002FSKILL.md",[],[],{"skillNo":165,"name":166,"nameZh":6,"description":167,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":138,"licenseFileUrl":6,"sourceHtmlUrl":168,"categories":169,"keywords":170,"publishedAt":20},"01M7EWY3NM0Z0JAKS7HDZ24K66","requesting-code-review","Use when completing tasks, implementing major features, or before merging to verify work meets requirements","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-code-review\u002FSKILL.md",[],[],{"skillNo":172,"name":173,"nameZh":6,"description":174,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":138,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":177,"publishedAt":20},"01M7EWY0WRMWX4M9ZFY0WKC593","brainstorming","You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-brainstorming\u002FSKILL.md",[],[]]