Top 10 Skills tips from OpenClaw in the past 24 hours: 1. Anthropic engineer @trq212: Summary of internal practices for Skills (11,724 likes) Anthropic has hundreds of Skills in use, with the core conclusion being that project-level Skills (placed in project .claude/skills/) are 3 times more effective than global Skills due to more precise context. Skills are divided into three categories: process-type (standardized work steps), knowledge-type (supplementary domain background), and tool-type (encapsulated reusable commands). The more specific, the better; instead of saying "help me write good code," say "write unit tests with Jest, covering the happy path + 3 edge cases." 2. @itsolelehmann: Method to increase Skills success rate from 56% to 92% (2,265 likes) Most Skills have only a 70% success rate, and the solution is "start before bed, check results in the morning"—write an automated testing loop that allows the Agent to run the Skill hundreds of times overnight, automatically scoring and rewriting, so by morning the Skill has already iterated. Essentially, the polishing process of the Skill is also handed over to AI. 3. Google Cloud: The SKILL.md format is dead; content design is the real challenge (1,643 likes) 30+ tools have unified the SKILL.md format, and the format issue has been resolved. The current challenge is "content design"—pack a Skill with a FastAPI specification and a four-step document generation Skill; the underlying logic is completely different, but the format looks the same. Think clearly: is your Skill constraining model behavior, providing background knowledge, or giving it tools? 4. @dotey translated version: In-depth guide to Anthropic Skills (1,534 likes) Complete Chinese translation of @trq212's original text. Added a key detail: the quality of the Skill depends on your depth of understanding of the task, not on prompt skills. If you can't clearly articulate what the standard for "good" is, the Skill you write will be unstable. It is recommended to write acceptance criteria first, then write the Skill itself. 5. @chenchengpro: Matt Pocock open-sources 5 TypeScript engineering Skills (232 likes) Well-known TypeScript engineer Matt Pocock has open-sourced 5 Skills he uses daily, including /grill-me, which makes the Agent act as a "harsh code reviewer" actively finding flaws in your code. The core point: the real gap is not in what model is used, but in who encodes engineering experience into reusable Skills. 6. @yan5xu: The ceiling for making Skills and the correct posture (275 likes) Sharp analysis: Prompt-type Skills can be copied, while Script-type Skills have barriers. Simply selling Skills at the ceiling is selling copies; the real value lies in encapsulating your business processes with Skills, allowing the Agent to run your unique workflows. Skills are leverage, not the product itself. 7. @coreyhainesco: Marketing Skills v1.4.0 released (409 likes) Update of the Skills package specifically for marketing scenarios, adding the /lead-magnets command (to generate high-conversion lead strategies) and integrating Composio MCP, a server directly connected to HubSpot, Salesforce, Meta Ads, and LinkedIn. This is a typical case of making Skills into a vertical product. 8. @AI_Jasonyu: Anthropic hackathon Skills configuration collection (217 likes) There is a complete Skills configuration collection from the Anthropic hackathon on GitHub, including production-level Agents, Skills, Hooks, commands, rules, and MCP configurations, which is the result of 10 months of intensive use. Using it directly is much faster than figuring it out from scratch. 9. Hermes Agent v0.3.0 plugin architecture (821 likes) The new version from Nous Research has made Skills "first-class citizens"—packaging, sharing, and installing Skills is as simple as installing an npm package. It also added the /browser command to natively take over Chrome, allowing Skills to directly control real browser operations on the web, no longer just text processing. 10. @startupideaspod: Drive the entire AI team with folders (427 likes)...