What exactly is broken in popular skills: references to files that are not bundled, actions taken without human approval, processes with no stopping condition. Everything here is measured from the skill text rather than judged by a model, so every line can be checked.
ProcedureGitHub
Train sparse autoencoders to interpret model features.
- 13 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureDockerGitHub
Vector search engine for production RAG systems.
- 13 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHub
Fast vector similarity search at billion scale.
- 10 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureKubernetes
Serverless GPU cloud for ML jobs and model APIs.
- 17 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ReferenceGitHub
Reference-free preference alignment, simpler than DPO.
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHubDockerwill not run
RL post-training for LLMs with Megatron and SGLang.
- References files that are not bundled: examples/search-r1/, scripts/models/
- 14 top-level sections: this looks like several domains in one skill
- 2 referenced file(s) missing: examples/search-r1/, scripts/models/
NousResearch/hermes-agent · ★ 245 156
IntegrationNotion
Notion API + ntn CLI: pages, databases, markdown, Workers.
- 10 top-level sections: this looks like several domains in one skill
- The response is described with custom markup (4 tags): a typed call is more reliable
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
Integrationwill not run
Search/download GIFs from Tenor via curl + jq.
- References files that are not bundled: url
- 1 referenced file(s) missing: url
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
Procedure
Search arXiv papers by keyword, author, category, or ID.
- 15 top-level sections: this looks like several domains in one skill
- Does not say what the result is
- No condition that starts the skill
NousResearch/hermes-agent · ★ 245 156
AnalyzerGitHubwill not run
Validated interactive HTML diagrams, upstream-maintained.
- References files that are not bundled: scripts/check-update.mjs
- 1 referenced file(s) missing: scripts/check-update.mjs
- Does not say what the result is
NousResearch/hermes-agent · ★ 245 156
ProcedureGitHubwill not run
Draw code changes as animated architecture/data-flow SVGs.
- References files that are not bundled: .pr-lens/<view>.svg
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- The response is described with custom markup (6 tags): a typed call is more reliable
NousResearch/hermes-agent · ★ 245 156
Referencewill not run
Meta-cognitive self-learning system - Automated skill evolution based on predictive coding and value-driven mechanisms.
- References files that are not bundled: MCP_CONFIG.md, ../MEMORY.md
- 2 referenced file(s) missing: MCP_CONFIG.md, ../MEMORY.md
- Does not say what the result is
clawhub · ↓ 13 878 ★ 35
Integration
Manage Things 3 via the `things` CLI on macOS (add/update projects+todos via URL scheme; read/search/list from the local Things database). Use when a user asks Clawdbot to add a ta
- The response is described with custom markup (17 tags): a typed call is more reliable
- Does not say what the result is
- No condition that starts the skill
clawhub · ↓ 16 836 ★ 20
ProcedureGitHubwill not run
The first security skill to install after setting up OpenClaw — powered by Tencent Zhuque Lab. Works like an antivirus for your AI environment: audits installed skills, scans skill
- References files that are not bundled: 参考reference链接
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 25 top-level sections: this looks like several domains in one skill
clawhub · ↓ 3 913 ★ 125
IntegrationStripeNotionwill not run
Google Workspace Admin SDK integration with managed OAuth. This is a write-capable administrative integration for users, groups, organizational units, roles, and domain settings. O
- References files that are not bundled: references/<source>/triggers.md, references/{source}/triggers.md
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 14 top-level sections: this looks like several domains in one skill
clawhub · ↓ 18 597 ★ 21
ProcedureSlackDiscordwill not run
Postiz is a tool to schedule social media and chat posts to 28+ channels X, LinkedIn, LinkedIn Page, Reddit, Instagram, Facebook Page, Threads, YouTube, Google My Business, TikTok,
- References files that are not bundled: COMMAND_LINE_GUIDE.md, examples/EXAMPLES.md, examples/basic-usage.sh
- 11 top-level sections: this looks like several domains in one skill
- 8 referenced file(s) missing: COMMAND_LINE_GUIDE.md, examples/EXAMPLES.md, examples/basic-usage.sh
clawhub · ↓ 12 454 ★ 34
AnalyzerGitHub
OpenClaw Token Optimizer v3.2.0 — practical cost-control toolkit for OpenClaw agents. Lazy context loading, Sonnet/Opus-aware routing, heartbeat scheduling, local token budgets, ca
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 10 top-level sections: this looks like several domains in one skill
- Does not say what the result is
clawhub · ↓ 12 822 ★ 30
Procedurewill not run
- References files that are not bundled: references/routing-guide.md
- 1 referenced file(s) missing: references/routing-guide.md
- Does not say what the result is
clawhub · ↓ 3 992 ★ 122
Analyzerwill not run
Evaluate skill quality, find the weakest dimension, and apply directed improvements. Also tracks usage to spot idle or risky skills. Use when: first session after install, or user
- References files that are not bundled: scripts/hud-extra.js
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 10 top-level sections: this looks like several domains in one skill
clawhub · ↓ 3 455 ★ 124
Reference
Fast file-name and content search using `fd` and `rg` (ripgrep).
- Prose only: no discrete steps
- Does not say what the result is
clawhub · ↓ 20 208 ★ 19
ProcedureGitHub
Query and manage Linear issues, projects, and team workflows.
- Does not say what the result is
- No condition that starts the skill
clawhub · ↓ 17 570 ★ 21
IntegrationGitHubwill not run
一键发布 Markdown 到微信公众号草稿箱。基于 wenyan-cli,支持多主题、代码高亮、图片自动上传。
- References files that are not bundled: images/photo.jpg, /Users/bruce/photo.jpg
- 10 top-level sections: this looks like several domains in one skill
- 2 referenced file(s) missing: images/photo.jpg, /Users/bruce/photo.jpg
clawhub · ↓ 9 494 ★ 39
AnalyzerPlaywright
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, a
- Does not say what the result is
- No condition that starts the skill
anthropics/skills · ★ 176 119
Procedurewill not run
Stop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claud
- References files that are not bundled: URL
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- 1 referenced file(s) missing: URL
anthropics/skills · ★ 176 119
GeneratorPowerPointSlack
Use this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsi
- Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- The skill tells the model to perform an irreversible action with no human approval
- Does not say what the result is
anthropics/skills · ★ 176 119