{ "measurement": "What the currently-trending Agent Skill packs cost to install, measured rather than derived", "claude_code_version": "2.1.233", "claude_code_version_recorded_at_runtime": "2.1.233 (Claude Code)", "why": "On 2026-08-15, GitHub Trending's daily, weekly and monthly boards carried 53 LISTINGS covering 48 DISTINCT repositories, and 9 of those matched the word 'skill' in the repository name or description. The two denominators are different and the corpus previously conflated them; the scrape is now published below so the count is recomputable rather than asserted. The obvious reader question is what installing one costs. This site has published a model for that, and deliberately does NOT use it here: multiplying a rate by a file count is arithmetic, not a measurement, and this site declined to publish exactly that on 2026-08-14 when the fixture did not match the real files. Each pack is installed and measured, and the model is then checked against the result.", "method": { "shape": "Each pack cloned from GitHub, every skill directory copied into an otherwise empty project's .claude/skills/, whole-session context measured against an empty floor from the same round. Prompt: 'Reply with only: OK'.", "estimator": "MEDIAN across rounds per pack. The assembler refuses to build if a median is a value no round produced.", "population": "The per-pack cell, delta against that round's own floor.", "controls": { "paired": "The floor is re-measured every round and every delta is against that round.", "arrival_and_injection": "The reply must be exactly OK, and on this harness that does double duty. These are THIRD-PARTY skill descriptions entering the model's context, so a description attempting to steer the model would surface as a reply that is not exactly OK. Published per run as reply_exact_ok.", "fixture": "Skills are counted back off disk per run, and the description lengths are measured as installed, so the fixture is described by measurement rather than by the repository's README.", "isolation": "--setting-sources project asserted onto the command line, so this machine's own skills cannot leak into any arm.", "bodies_never_invoked": "The prompt is trivial and no skill is ever called, so no third-party skill body is executed or read into context. Only the listing is priced." } }, "results": { "floor": 22283.0, "floor_all_rounds": [ 22283 ], "packs": { "Nutlope_hallmark": { "skills_installed": 1, "description_median_chars": 260, "description_mean_chars": 260, "description_max_chars": 260, "measured_tokens": 117.0, "measured_per_skill": 117.0, "rounds": 4, "all_rounds": [ -404, 117, 117, 117 ], "spread": 521, "model_predicts": 85, "model_error": "model is 27% LOW", "model_applied_outside_its_published_range": false, "injection_control_passed": true, "commit": "13ac0ec", "commit_date": "2026-08-06", "cloned": "2026-08-15", "body_bytes_median": 67149, "body_bytes_total": 67149 }, "addyosmani_agent-skills": { "skills_installed": 24, "description_median_chars": 249, "description_mean_chars": 271, "description_max_chars": 485, "measured_tokens": 2302.0, "measured_per_skill": 95.92, "rounds": 4, "all_rounds": [ 2302, 2302, 2302, 2302 ], "spread": 0, "model_predicts": 2124, "model_error": "model is 8% LOW", "model_applied_outside_its_published_range": false, "injection_control_passed": true, "commit": "df1edb2", "commit_date": "2026-08-14", "cloned": "2026-08-15", "body_bytes_median": 11756, "body_bytes_total": 303431 }, "cathrynlavery_diagram-design": { "skills_installed": 1, "description_median_chars": 579, "description_mean_chars": 579, "description_max_chars": 579, "measured_tokens": 274.0, "measured_per_skill": 274.0, "rounds": 4, "all_rounds": [ 274, 274, 274, 274 ], "spread": 0, "model_predicts": 174, "model_error": "model is 37% LOW", "model_applied_outside_its_published_range": true, "injection_control_passed": true, "commit": "09df49d", "commit_date": "2026-08-14", "cloned": "2026-08-15", "body_bytes_median": 36910, "body_bytes_total": 36910 }, "google_skills": { "skills_installed": 111, "description_median_chars": 397, "description_mean_chars": 430, "description_max_chars": 1021, "measured_tokens": 9020.0, "measured_per_skill": 81.26, "rounds": 4, "all_rounds": [ 9020, 9020, 9020, 9020 ], "spread": 0, "model_predicts": 14699, "model_error": "model is 63% HIGH", "model_applied_outside_its_published_range": true, "injection_control_passed": true, "commit": "8f57a0d", "commit_date": "2026-08-14", "cloned": "2026-08-15", "body_bytes_median": 8948, "body_bytes_total": 1095932, "skills_by_area": { "ads": 13, "cloud": 96, "analytics": 2 }, "marketplace_json": { "plugins_listed": 16, "distinct_source_repositories": 16, "self_referencing": false, "note": "A MARKETPLACE manifest, not a plugin manifest. It points at other repositories and enumerates nothing in this one, so it does not gate the 111 skills measured here." } }, "mattpocock_skills": { "skills_installed": 35, "description_median_chars": 149, "description_mean_chars": 151, "description_max_chars": 418, "measured_tokens": 1209.0, "measured_per_skill": 34.54, "rounds": 4, "all_rounds": [ 1209, 1209, 1209, 1209 ], "spread": 0, "model_predicts": 1935, "model_error": "model is 60% HIGH", "model_applied_outside_its_published_range": false, "injection_control_passed": true, "commit": "8b78b53", "commit_date": "2026-08-13", "cloned": "2026-08-15", "body_bytes_median": 3339, "body_bytes_total": 146055 }, "tt-a1i_archify": { "skills_installed": 1, "description_median_chars": 652, "description_mean_chars": 652, "description_max_chars": 652, "measured_tokens": 262.0, "measured_per_skill": 262.0, "rounds": 4, "all_rounds": [ 262, 262, 262, 262 ], "spread": 0, "model_predicts": 194, "model_error": "model is 26% LOW", "model_applied_outside_its_published_range": true, "injection_control_passed": true, "commit": "cffdd42", "commit_date": "2026-08-14", "cloned": "2026-08-15", "body_bytes_median": 12226, "body_bytes_total": 12226 } }, "headline": "Installing a trending skill pack costs between a few hundred and about nine thousand tokens on every session, before anything is invoked.", "the_model_check": "This site's published skills model does NOT reliably predict a real pack, and the error does not even go the same way each time. Across the three multi-skill packs it lands within 8% on one and over-predicts by about 60% on the other two. So it is an order-of-magnitude guide, not a substitute for installing the thing and measuring it. The reason for the gap is NOT established here.", "shipped_vs_all_files": { "why": "The first pass walked every SKILL.md in each cloned repository. For five of six packs that is exactly what a user gets. For mattpocock/skills it is not: its .claude-plugin/plugin.json enumerates 25 skills while the repository carries 35 SKILL.md files, the extra ten sitting under skills/in-progress/ (6) and skills/misc/ (4) and deliberately not shipped. The originally published 1,209-token figure therefore priced 40% more skills than `claude plugins install` gives anyone, which is a wrong number attached to a named third party's project.", "method": "Re-run with a manifest-aware installer: where plugin.json carries an explicit skills LIST, only those directories are installed; where it carries a bare directory pointer or no manifest at all, everything ships and the walk is right.", "packs": { "Nutlope_hallmark": { "skills_shipped_by_manifest": 1, "measured_tokens": 117.0, "per_skill": 117.0, "rounds": 4, "all_rounds": [ 117, 117, 117, 117 ], "differs_from_all_files_walk": false, "description_median_chars": 260, "description_total_chars": 260, "model_predicts": 85, "model_error": "model is 27% LOW", "model_applied_outside_its_published_range": false, "injection_control_passed": true }, "addyosmani_agent-skills": { "skills_shipped_by_manifest": 24, "measured_tokens": 2302.0, "per_skill": 95.92, "rounds": 4, "all_rounds": [ 2302, 2302, 2302, 2302 ], "differs_from_all_files_walk": false, "description_median_chars": 249, "description_total_chars": 6505, "model_predicts": 2124, "model_error": "model is 8% LOW", "model_applied_outside_its_published_range": false, "injection_control_passed": true }, "cathrynlavery_diagram-design": { "skills_shipped_by_manifest": 1, "measured_tokens": 274.0, "per_skill": 274.0, "rounds": 4, "all_rounds": [ 274, 274, 274, 274 ], "differs_from_all_files_walk": false, "description_median_chars": 579, "description_total_chars": 579, "model_predicts": 174, "model_error": "model is 37% LOW", "model_applied_outside_its_published_range": true, "injection_control_passed": true }, "google_skills": { "skills_shipped_by_manifest": 111, "measured_tokens": 9020.0, "per_skill": 81.26, "rounds": 4, "all_rounds": [ 9020, 9020, 9020, 9020 ], "differs_from_all_files_walk": false, "description_median_chars": 397, "description_total_chars": 47748, "model_predicts": 14699, "model_error": "model is 63% HIGH", "model_applied_outside_its_published_range": true, "injection_control_passed": true }, "mattpocock_skills": { "skills_shipped_by_manifest": 25, "measured_tokens": 852.0, "per_skill": 34.08, "rounds": 4, "all_rounds": [ 852, 852, 852, 852 ], "differs_from_all_files_walk": true, "description_median_chars": 149, "description_total_chars": 3822, "model_predicts": 1397, "model_error": "model is 64% HIGH", "model_applied_outside_its_published_range": false, "injection_control_passed": true }, "tt-a1i_archify": { "skills_shipped_by_manifest": 1, "measured_tokens": 262.0, "per_skill": 262.0, "rounds": 4, "all_rounds": [ 262, 262, 262, 262 ], "differs_from_all_files_walk": false, "description_median_chars": 652, "description_total_chars": 652, "model_predicts": 194, "model_error": "model is 26% LOW", "model_applied_outside_its_published_range": true, "injection_control_passed": true } }, "reading": "Five of six packs reproduce IDENTICALLY across the two independent runs, which is a reproducibility check as well as a correction. Only mattpocock/skills moves: 852 tokens for the 25 skills it ships against 1,209 for the 35 files in the repo. Its per-skill cost is unchanged at 34.08 against 34.54, so the difference is purely the count, not the pricing.", "runs": [ { "pack": "floor", "round": 0, "status": "ok", "skills_on_disk": 0, "context": 22287, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 0, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22404, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 0, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24589, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 0, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22561, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 0, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31307, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 0, "status": "ok", "skills_on_disk": 25, "description_stats": { "n": 25, "median": 149, "mean": 153, "max": 418, "total_chars": 3822 }, "context": 23139, "delta": 852, "per_skill": 34.08, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 0, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22549, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "floor", "round": 1, "status": "ok", "skills_on_disk": 0, "context": 22287, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 1, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22404, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 1, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24589, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 1, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22561, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 1, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31307, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 1, "status": "ok", "skills_on_disk": 25, "description_stats": { "n": 25, "median": 149, "mean": 153, "max": 418, "total_chars": 3822 }, "context": 23139, "delta": 852, "per_skill": 34.08, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 1, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22549, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "floor", "round": 2, "status": "ok", "skills_on_disk": 0, "context": 22287, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 2, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22404, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 2, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24589, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 2, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22561, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 2, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31307, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 2, "status": "ok", "skills_on_disk": 25, "description_stats": { "n": 25, "median": 149, "mean": 153, "max": 418, "total_chars": 3822 }, "context": 23139, "delta": 852, "per_skill": 34.08, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 2, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22549, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "floor", "round": 3, "status": "ok", "skills_on_disk": 0, "context": 22287, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 3, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22404, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 3, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24589, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 3, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22561, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 3, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31307, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 3, "status": "ok", "skills_on_disk": 25, "description_stats": { "n": 25, "median": 149, "mean": 153, "max": 418, "total_chars": 3822 }, "context": 23139, "delta": 852, "per_skill": 34.08, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 3, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22549, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" } ], "the_correction_makes_my_model_worse": "Worth stating plainly because it cuts against me. On the 35-file walk this site's published skills model over-predicted mattpocock/skills by 60%. Priced against the 25 skills the plugin actually ships, the same model over-predicts by 64%. Correcting the count did not rescue the model; it made the miss slightly larger. The model's failure on this pack is therefore not an artefact of counting the wrong files.", "independent_reproduction": "The second harness is a SEPARATE SCRIPT in a SEPARATE PROCESS, into separate temporary project directories, about twenty minutes after the first: the raw artefacts are stamped 09:25:58-09:27:05 and 09:45:29-09:46:37 on 2026-08-15. Their floors are NOT identical, 22,283 then 22,287, a 4-token drift, and every delta is paired against its own round's floor, so four of six packs came back identical to the token across that drift. BE PRECISE ABOUT WHAT THAT DOES AND DOES NOT ESTABLISH. Both harnesses read the SAME clones under /tmp/skillpacks, created once at 09:24:29-09:24:38 before either ran; neither script clones anything. So the second run re-tests the measurement and the machine state twenty minutes on, and does NOT re-test the fixture. A mis-clone would make both runs wrong together and agree perfectly. Stronger than four rounds inside one process, weaker than a re-clone on another machine." }, "trending_scrape": { "date": "2026-08-15", "windows": [ "daily", "weekly", "monthly" ], "listings": 53, "distinct_repositories": 48, "rule": "case-insensitive /\\bskills?\\b/ against 'owner/name' plus the repository description", "matches": [ "Nutlope/hallmark", "TencentCloud/TencentDB-Agent-Memory", "addyosmani/agent-skills", "google/skills", "kangarooking/cangjie-skill", "mattpocock/skills", "tt-a1i/archify", "virgiliojr94/book-to-skill", "zhaoxuya520/reverse-skill" ], "match_count": 9, "the_rule_is_crude_in_both_directions": "It catches TencentCloud/TencentDB-Agent-Memory, which is a database memory hub and not a skill pack, and it misses cathrynlavery/diagram-design, which was number one on the daily board and ships a Claude Code skill without using the word. Both are stated in the post. The rule is published so a reader can disagree with it and recount.", "parser_note": "Scraped per ARTICLE, not by zipping a list of names against a list of descriptions. The first parser did the latter, and on this very board one repository shipped no description, so every repository after it was captioned with its neighbour's text. Descriptions were cross-checked against the GitHub API.", "repositories": [ { "repo": "cathrynlavery/diagram-design", "description": "29 editorial diagram types for Claude Code. Self-contained HTML + SVG. No shadows, no Mermaid-slop.", "language": "HTML" }, { "repo": "cactus-compute/needle", "description": "14MB foundation model for tiny devices; phones, wearables, smart home, and robots.", "language": "Python" }, { "repo": "megadose/holehe", "description": "holehe allows you to check if the mail is used on different sites like twitter, instagram and will retrieve information on sites with the forgotten password function.", "language": "Python" }, { "repo": "macro-inc/macro", "description": "Macro is a unified workspace for teams: email, chat, docs, tasks, agents, calls, and CRM \u2014 @-linked together with shared AI memory.", "language": "Rust" }, { "repo": "smicallef/spiderfoot", "description": "SpiderFoot automates OSINT for threat intelligence and mapping your attack surface.", "language": "Python" }, { "repo": "citrolabs/ego-lite", "description": "The fastest browser for AI agents to run browser automation, built for sharing your logged-in browser state with your AI agents, like Codex or Claude Code, without disturbing you. Zero cost, zero config.", "language": "JavaScript" }, { "repo": "holaboss-ai/holaOS", "description": "Open-source All in One AI agent workspace. Run any agent \u2014 Claude Code, Codex \u2014 across your tools (100+ integrations + MCP), apps, browser, and files, with shared memory. Built-in models or BYOK.", "language": "TypeScript" }, { "repo": "github/spec-kit", "description": "\ud83d\udcab Toolkit to help you get started with Spec-Driven Development", "language": "Python" }, { "repo": "lightningpixel/modly", "description": "Desktop app to generate 3D models from images or prompt using local AI \u2014 runs entirely on your GPU", "language": "TypeScript" }, { "repo": "infiniflow/ragflow", "description": "RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs", "language": "Go" }, { "repo": "cursor/plugins", "description": "Cursor plugin specification and official plugins", "language": "TypeScript" }, { "repo": "deepseek-ai/awesome-deepseek-agent", "description": null, "language": null }, { "repo": "semantica-agi/semantica", "description": "Graph-Native Infrastructure for Context and Accountable AI Systems", "language": "Python" }, { "repo": "rustdesk/rustdesk", "description": "An open-source remote desktop application designed for self-hosting, as an alternative to TeamViewer.", "language": "Rust" }, { "repo": "OpenCut-app/OpenCut", "description": "The open-source CapCut alternative", "language": "TypeScript" }, { "repo": "unslothai/unsloth", "description": "Local UI to run and train LLMs and diffusion models, including Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, DeepSeek-V4, FLUX and more.", "language": "Python" }, { "repo": "ToolJet/ToolJet", "description": "ToolJet is the open-source foundation of ToolJet AI - the enterprise app generation platform for building internal tools, dashboard, business applications, workflows and AI agents \ud83d\ude80", "language": "JavaScript" }, { "repo": "PrimeIntellect-ai/prime-agent", "description": "A self-improving RLM agent for coding workflows and long-running autonomous tasks.", "language": "TypeScript" }, { "repo": "google/skills", "description": "Agent Skills for Google products and technologies", "language": "Python" }, { "repo": "denoland/celld", "description": "self-hosted, distributed Durable Objects", "language": "Rust" }, { "repo": "vitali87/code-graph-rag", "description": "The ultimate RAG for your monorepo. Query, understand, and edit multi-language codebases with the power of AI and knowledge graphs", "language": "Python" }, { "repo": "NVIDIA-NeMo/Switchyard", "description": "Switchyard lets LLM applications route traffic across models and providers while preserving native OpenAI and Anthropic API compatibility - enabling flexible model selection, benchmarking, and cost/performance optimization.", "language": "Rust" }, { "repo": "cloudflare/computer", "description": "Give your agent a computer \ud83d\udc7e", "language": "TypeScript" }, { "repo": "addyosmani/agent-skills", "description": "Production-grade engineering skills for AI coding agents.", "language": "JavaScript" }, { "repo": "TencentCloud/TencentDB-Agent-Memory", "description": "TencentDB Agent Memory is a team-level memory hub for AI Agents \u2014 turning conversations, docs, and code into four reusable memory assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) that are governed, shared, and equipped across agents and frameworks.", "language": "TypeScript" }, { "repo": "huangruiteng/loopx", "description": "Lightweight loop engineering state kernel for long-running AI agent teams. Agent-loop agnostic across Codex, Claude Code, and other coding agents, with durable goals, quota-aware auto-wake, executable todos, evidence logs, and verifiable handoffs.", "language": "Python" }, { "repo": "3b1b/manim", "description": "Animation engine for explanatory math videos", "language": "Python" }, { "repo": "LadybirdBrowser/ladybird", "description": "Truly independent web browser", "language": "C++" }, { "repo": "TapXWorld/ChinaTextbook", "description": "\u6240\u6709\u5c0f\u521d\u9ad8\u3001\u5927\u5b66PDF\u6559\u6750\u3002", "language": "Roff" }, { "repo": "pingdotgg/t3code", "description": null, "language": "TypeScript" }, { "repo": "diegosouzapw/OmniRoute", "description": "Never stop coding. Free MIT AI gateway: one endpoint, 339 providers (90+ free), 1200+ models \u2014 Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 450+ contributors", "language": "TypeScript" }, { "repo": "koala73/worldmonitor", "description": "Real-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface", "language": "TypeScript" }, { "repo": "1jehuang/jcode", "description": "The most RAM efficient harness", "language": "Rust" }, { "repo": "zhaoxuya520/reverse-skill", "description": "Reverse Engineering / Authorized Penetration Testing / Security Research Skill Router Pack AI-powered routing + On-demand toolchain bootstrapping + Self-evolving knowledge base Supports Claude Code, Kiro, Cursor, Cline, and other AI coding clients \u9006\u5411/\u6e17\u900f/\u5b89\u5168\u6280\u80fd\u8def\u7531\u5305 - AI \u81ea\u52a8\u8def\u7531 + \u6309\u9700\u81ea\u4e3e\u5de5\u5177\u94fe + \u81ea\u52a8\u8fdb\u5316\u7ecf\u9a8c\u5e93 | \u652f\u6301 Claude Code / Kiro / Cursor / Cline \u7b49\u4ee3\u7801 AI \u5ba2\u6237\u7aef", "language": "PowerShell" }, { "repo": "stablyai/orca", "description": "Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and VPS.", "language": "TypeScript" }, { "repo": "HKUDS/DeepTutor", "description": "DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.", "language": "Python" }, { "repo": "mattpocock/skills", "description": "Skills for Real Engineers. Straight from my .agents directory.", "language": "Shell" }, { "repo": "earendil-works/pi", "description": "AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI", "language": "TypeScript" }, { "repo": "virgiliojr94/book-to-skill", "description": "Turn any technical book PDF into a Claude Code skill \u2014 ready to study, reference, and use while you work.", "language": "Python" }, { "repo": "every-app/open-seo", "description": "Open source alternative to Semrush and Ahrefs", "language": "TypeScript" }, { "repo": "tirth8205/code-review-graph", "description": "Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with benchmarked context reductions on reviews and large-repo workflows.", "language": "Python" }, { "repo": "Nutlope/hallmark", "description": "Anti-AI-slop design skill for Claude Code, Cursor, and Codex.", "language": "CSS" }, { "repo": "agegr/pi-web", "description": "Web UI for the pi coding agent", "language": "TypeScript" }, { "repo": "kangarooking/cangjie-skill", "description": "\u628a\u4e66\u3001\u957f\u89c6\u9891\u3001\u64ad\u5ba2\u7b49\u9ad8\u4ef7\u503c\u5185\u5bb9\u84b8\u998f\u6210\u53ef\u6267\u884c\u7684 Agent Skills", "language": "Python" }, { "repo": "MoonshotAI/kimi-code", "description": "Kimi Code CLI \u2014 The Starting Point for Next-Gen Agents", "language": "TypeScript" }, { "repo": "tt-a1i/archify", "description": "Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams\u2014self-contained HTML with motion and crisp export.", "language": "HTML" }, { "repo": "different-ai/openwork", "description": "The open-source alternative to Claude Cowork (powered by opencode)", "language": "TypeScript" }, { "repo": "opengeos/GeoLibre", "description": "A lightweight, cloud-native GIS platform for visualizing, exploring, and analyzing geospatial data. It runs in the web browser, on the desktop, on mobile, and inside Jupyter notebooks.", "language": "TypeScript" } ] } }, "known_limits": { "single_skill_packs_are_inside_the_noise": "The one-skill packs (Nutlope_hallmark, cathrynlavery_diagram-design, tt-a1i_archify) are at or below this harness's noise. This site's own skills corpus already records that a single skill cannot be resolved against the floor, and that limit is carried forward unchanged rather than quietly dropped.", "the_model_gap_is_unexplained": "The model misses in both directions and no cause is offered. Candidates not tested here include the description field being truncated in the listing, real descriptions tokenising differently from the prose fixture the model was fitted on, and the listing compacting sooner at longer description lengths. Naming one would be publishing a confound's direction from reasoning.", "the_model_is_applied_outside_its_published_range_for_some_packs": "The model is declared valid for descriptions of 30 to 300 characters. Packs whose median description exceeds that are flagged per pack with model_applied_outside_its_published_range, and for those the comparison is indicative only. That limit was published with the model and is being honoured, not ignored.", "the_injection_control_detects_only_loud_payloads": "The arrival control shows that no description changed the literal output: all 56 replies across BOTH harnesses were exactly OK. It does NOT show that no description influenced the model. This site has measured exactly that gap: in prompt-injection-ISOLATED.json (the 20-run isolated re-verification, not the 30-run trial) a loud payload was obeyed 0 of 20 times while a QUIET one was followed 6 of 6 by Haiku and 0 of 6 by Opus, so a plausible-looking instruction can be acted on without the output looking wrong. Opus is the model this harness ran on and it resisted that payload every time, which is reassuring and is not the same as a positive control. There is no positive control here: no arm contains a known-steering description, so nothing demonstrates the check would fire. Treat it as a smoke alarm for blatant payloads, not a clean bill of health for the 163 third-party descriptions in the shipped fixture (173 in the first harness, which carried ten skills mattpocock does not ship).", "per_skill_cost_does_not_order_by_description_length": "google/skills has a median description 1.6x longer than addyosmani/agent-skills and costs 15% LESS per skill (81.26 against 95.92). So across these three packs the per-skill cost does not order by description length, and no single-variable explanation is offered.", "a_pack_is_a_moving_target": "Each pack is measured at one commit on one day. These repositories are trending, which means they are changing fast; the commit is recorded so the figure can be reproduced.", "listing_cost_only": "This prices having the pack installed. It does not price using it: a skill body is charged when the skill is invoked, and nothing here invokes one.", "one_machine_one_model": "Opus on Claude Code 2.1.233, project scope only.", "a_repo_is_not_always_what_the_plugin_ships": "Counting SKILL.md files on disk is not the same as counting what a plugin installs, and the three multi-skill packs here show three different shapes. mattpocock/skills has a plugin.json naming 25 of its 35 skills explicitly, so the file walk over-counted. addyosmani/agent-skills points its manifest at a directory, so everything ships and the walk was right. google/skills has NO plugin.json at all: what it ships is a marketplace.json listing 16 plugins living in 16 OTHER repositories, none of them google/skills, so it enumerates nothing measured here. The 9,020 figure therefore prices one defensible reading, copying every skill in the repository, and the repository is not homogeneous: {'ads': 13, 'cloud': 96, 'analytics': 2}. Someone who wants the Ads skills installs 13, not 111, and does not pay 9,020. Check the manifest before pricing anyone's pack, and check what kind of manifest it actually is.", "the_corrected_figure_has_no_cross_harness_replication": "Four packs are identical to the token across both independent harnesses. mattpocock/skills is NOT one of them and cannot be: the two runs measured different fixtures, 35 walked files and then the 25 the manifest ships. So the corrected 852 rests on four rounds in the second harness alone, with zero spread inside them and no cross-harness replication behind it. The withdrawn 1,209 had exactly the same backing. Stated because the correction is otherwise easy to read as better-evidenced than it is.", "the_second_harness_shares_the_fixture": "The two harness runs share one clone set on one machine twenty minutes apart. They control for process state and transient machine state, and for nothing about the repositories themselves. Any figure resting on 'it reproduced across both harnesses' is a claim about measurement stability, not about the fixture being right." }, "runs": [ { "pack": "floor", "round": 0, "status": "ok", "skills_on_disk": 0, "context": 22283, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 0, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 21879, "delta": -404, "per_skill": -404.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 0, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24585, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 0, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22557, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 0, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31303, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 0, "status": "ok", "skills_on_disk": 35, "description_stats": { "n": 35, "median": 149, "mean": 151, "max": 418, "total_chars": 5274 }, "context": 23492, "delta": 1209, "per_skill": 34.54, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 0, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22545, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "floor", "round": 1, "status": "ok", "skills_on_disk": 0, "context": 22283, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 1, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22400, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 1, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24585, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 1, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22557, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 1, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31303, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 1, "status": "ok", "skills_on_disk": 35, "description_stats": { "n": 35, "median": 149, "mean": 151, "max": 418, "total_chars": 5274 }, "context": 23492, "delta": 1209, "per_skill": 34.54, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 1, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22545, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "floor", "round": 2, "status": "ok", "skills_on_disk": 0, "context": 22283, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 2, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22400, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 2, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24585, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 2, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22557, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 2, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31303, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 2, "status": "ok", "skills_on_disk": 35, "description_stats": { "n": 35, "median": 149, "mean": 151, "max": 418, "total_chars": 5274 }, "context": 23492, "delta": 1209, "per_skill": 34.54, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 2, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22545, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "floor", "round": 3, "status": "ok", "skills_on_disk": 0, "context": 22283, "delta": 0, "reply_exact_ok": true }, { "pack": "Nutlope_hallmark", "round": 3, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 260, "mean": 260, "max": 260, "total_chars": 260 }, "context": 22400, "delta": 117, "per_skill": 117.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "addyosmani_agent-skills", "round": 3, "status": "ok", "skills_on_disk": 24, "description_stats": { "n": 24, "median": 249, "mean": 271, "max": 485, "total_chars": 6505 }, "context": 24585, "delta": 2302, "per_skill": 95.92, "reply_exact_ok": true, "reply": "OK" }, { "pack": "cathrynlavery_diagram-design", "round": 3, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 579, "mean": 579, "max": 579, "total_chars": 579 }, "context": 22557, "delta": 274, "per_skill": 274.0, "reply_exact_ok": true, "reply": "OK" }, { "pack": "google_skills", "round": 3, "status": "ok", "skills_on_disk": 111, "description_stats": { "n": 111, "median": 397, "mean": 430, "max": 1021, "total_chars": 47748 }, "context": 31303, "delta": 9020, "per_skill": 81.26, "reply_exact_ok": true, "reply": "OK" }, { "pack": "mattpocock_skills", "round": 3, "status": "ok", "skills_on_disk": 35, "description_stats": { "n": 35, "median": 149, "mean": 151, "max": 418, "total_chars": 5274 }, "context": 23492, "delta": 1209, "per_skill": 34.54, "reply_exact_ok": true, "reply": "OK" }, { "pack": "tt-a1i_archify", "round": 3, "status": "ok", "skills_on_disk": 1, "description_stats": { "n": 1, "median": 652, "mean": 652, "max": 652, "total_chars": 652 }, "context": 22545, "delta": 262, "per_skill": 262.0, "reply_exact_ok": true, "reply": "OK" } ] }