[Blog](/blog/.md) / [Engineering](/blog/tags/engineering/.md) # Agent skills automate drudgery so you don't have to do it Xe Iaso · February 17, 2026 · 10 min read [![Xe Iaso](https://avatars.githubusercontent.com/u/529003?v=4)](https://xeiaso.net) [Xe Iaso](https://xeiaso.net) Senior Cloud Whisperer ![A robot arm carefully organizing colorful skill badges on a workspace surface](/blog/assets/images/hero-image-863a31ddc6804fc2303f1befd18a7253.webp) When you are automating things, the dream is to be able to *offload* the task to the automation so you don't have to do it by hand anymore. At Tigris we're excited about [agent skills](https://agentskills.io/home) because they let us describe how to do things in simple human-friendly language and then put that documentation to work with AI agents. We suspect that this will also make spinning up new teammates easier as they join our team. If you want to check out our agent skills library, take a look on [skills.sh](http://skills.sh): ## Agent skills are cool as heck[​](#agent-skills-are-cool-as-heck "Direct link to Agent skills are cool as heck") I'm gonna admit, when I first heard of agent skills, I thought they were superfluous and a waste of time. We already have MCP, and I tend to prefer minimalism when possible. Adding skills seemed like a waste of time because I wasn't convinced that it would actually make things better. After using them for a while, I think I've found the right niche that they fit into. Agent skills are halfway between high level documentation for humans and low level scripts to automate tasks. They best fit into those shapes of problems that you can *describe* and *validate* easily, but would take a lot of annoying or fiddly code to implement properly. ### The shape where skills are needed[​](#the-shape-where-skills-are-needed "Direct link to The shape where skills are needed") As an example of the right shape of problem, consider the problem of image format optimization. In our documentation and blog repositories we ban using png files for images that get included into pages. We store all our documentation, blogposts, and static assets in a git repository and git [is notoriously bad at handling large files properly](https://stackoverflow.com/questions/29393447/why-cant-git-handle-large-files-and-large-repos). As a result we end up converting image files to webp with `ffmpeg`: ``` ffmpeg -i image.png image.webp ``` This can make images go from filling entire 3d-printed save icons to 50,000 kilobytes. If the images are smaller, they'll load faster and you can enjoy what we write more easily. At first I thought this was the right shape of problem for a script. If you think about it, the core elements are there: * Clearly defined problem with obvious solutions (convert images, replace references, delete old files). * Trivially verifiable output (no png files, images load properly in the websites). * Easy to use APIs for parsing and converting the relevant files. However this doesn't really pan out in practice. There are many ways to include an image in a markdown file and adding edge cases for all of them tends to get annoying fast, especially when using multi-line HTML tags like our formatter does. As a result, I made the [convert image to webp skill](https://github.com/tigrisdata/tigris-blog/blob/main/.claude/skills/convert-images-to-webp/SKILL.md) in our blog repo. This automates the entire process and creates trivial to review patches. This makes copying blogposts over from Google Docs into our blog repo take single digit minutes instead of upwards of an hour at the worst case. ## Our skill library[​](#our-skill-library "Direct link to Our skill library") As I was experimenting with other skills and making my own, I realized that we needed to share them so that everyone else can benefit from the best of the best. As a result, we now have the [Tigris Agent Skills Library](https://skills.sh/tigrisdata/skills) and you can install the skills with `npx skills add`: ``` npx skills add https://github.com/tigrisdata/skills ``` ### Tigris SDK skills[​](#tigris-sdk-skills "Direct link to Tigris SDK skills") The following skills help your agents use the [Tigris SDK](https://www.tigrisdata.com/docs/sdks/tigris/) better: * [Installing Tigris Storage](https://skills.sh/tigrisdata/skills/installing-tigris-storage): Installing and setting up the Tigris SDK. * [Tigris Object Operations](https://skills.sh/tigrisdata/skills/tigris-object-operations): How to do basic object operations in TypeScript. * [Tigris Bucket Management](https://skills.sh/tigrisdata/skills/tigris-bucket-management): How to create, list, inspect, and remove buckets in Tigris. * [Tigris Snapshots and Forking](https://skills.sh/tigrisdata/skills/tigris-snapshots-forking): How to snapshot and fork buckets in Tigris. These skills help agents a lot because the Tigris SDK is intended to be very intuitive and AI agents don't always have language server integrations. This restates parts of the documentation in authoritative text that AI agents can easily understand, hook into, and use to help guide you through taking advantage of everything that Tigris has to offer. They're also fully suitable for humans to learn from, if a bit dry. #### Object operations[​](#object-operations "Direct link to Object operations") The object operations skill covers the bread and butter of working with Tigris: uploading, downloading, deleting, listing, and generating presigned URLs. Every method in the Tigris SDK returns a `TigrisStorageResponse`, so the skill drills into agents that they need to check `result.error` before touching `result.data`. This sounds obvious, but without the skill, agents love to just assume everything worked and plow ahead. The skill also covers patterns that trip agents up in practice. Large file uploads need multipart configuration. Listing objects requires pagination when you have more than a page of results. Downloads come in three flavors (string, file, stream) and picking the wrong one leads to confusing bugs. After installing the skill, you can just ask your agent to do what you need: ``` > Add a route that lets users upload avatars to the "profiles" bucket and returns a presigned URL for the uploaded image. ``` The agent picks up the skill automatically, knows to use multipart for large files, checks `result.error` before `result.data`, and generates a presigned URL with the right expiration. Without the skill, you'd spend half your time correcting the agent's assumptions about the API. #### Snapshots and forking in practice[​](#snapshots-and-forking-in-practice "Direct link to Snapshots and forking in practice") The snapshots and forking skill is worth calling out specifically because it teaches agents about one of Tigris's most powerful features. Snapshots capture your entire bucket at a point in time, and forks create instant isolated copies using copy-on-write. Even for terabytes of data, forking is essentially free—no data copying required. This is huge for AI agent workflows. Think about it: an agent can snapshot a production bucket, fork it into an isolated sandbox, run whatever experiments it needs to, and the source data is completely untouched. If the agent deletes everything in the fork, the original bucket doesn't care. This makes it safe to let agents operate on real data without the risk of catastrophic mistakes. The skill walks agents through the full lifecycle. Just tell your agent what you need: ``` > Snapshot the "assets" bucket, then fork it into a sandbox I can use for testing the migration script. ``` The agent knows to create a named snapshot, fork from it, and—critically—it knows that the bucket needs `enableSnapshot: true` from creation. Without the skill, agents hit a wall trying to enable snapshots on an existing bucket and waste your time debugging an API error that has a simple but non-obvious fix. ### Conventional Commits[​](#conventional-commits "Direct link to Conventional Commits") Tigris uses [Conventional Commits](https://www.conventionalcommits.org/en/v1.0.0/) so that we can easily glance at commits and pull requests to get a rough idea of what they do. If AI models are prompted to write conventional commits in your [AGENTS.md](https://agents.md/) file, they'll usually get it somewhat right, but we can do better. The [conventional-commits](https://skills.sh/tigrisdata/skills/conventional-commits) skill makes AI models get it right *every* time. Even better, the description is filled with the right keywords to make agents latch onto the skills whenever you ask to commit something: ``` name: conventional-commits description: > Use when creating git commits, writing commit messages, or following version control workflows. ``` This results in commits that look like this: ``` chore: format code with prettier Run npx prettier -w to format code. Assisted-by: GLM 4.7 via Claude Code Signed-off-by: Xe Iaso ``` This is much better than the worst I've seen on some teams where you get multi-thousand line commits with messages like "implemented metrics rollups". You can install just the conventional commits skill with this command: ``` npx skills add https://github.com/tigrisdata/skills --skill conventional-commits ``` ### Go table-driven tests[​](#go-table-driven-tests "Direct link to Go table-driven tests") When writing tests in [Go](https://go.dev), the idiomatic style is to use a [table-driven test](https://go.dev/wiki/TableDrivenTests) where you define a table of inputs and expected outputs. This allows you to easily add test cases without having to copy and paste your environment setup and teardown code every time. Every test gets run as its own subtest so you can see each test's result at a glance. The table-driven tests skill teaches agents how to best implement table-driven tests based on advice from the greatest minds in Go programming. In my experience this has made tests that are performant, correct, and easy to debug. You can install just the table-driven tests skill with this command: ``` npx skills add https://github.com/tigrisdata/skills --skill go-table-driven-tests ``` ## Other cool things we're doing with skills[​](#other-cool-things-were-doing-with-skills "Direct link to Other cool things we're doing with skills") We've been messing around with Claude Code skills for other things internally on my team at Tigris. When I write something I usually use a chorus of four style guides to do a once-over to find glaring issues: * A style guide based on [the patterns of my blog](https://xeiaso.net/blog), this pass mostly focuses on things that speak *up* to the reader and avoiding overly corporate phrasing. * [SEO & AEO Best Practices](https://skills.sh/sanity-io/agent-toolkit/seo-aeo-best-practices) to make sure that my writing has active voice (being moderately femme of centre makes me more likely to speak in passive voice) and is well understood by search/answer engines. * [Humanizer](https://skills.sh/softaworks/agent-toolkit/humanizer) because I am exposed to a lot of AI writing due to my job function. I've kinda adopted some of the AI writing patterns without realizing it and having something call it out helps me not repeat that framing. * [Writing Clearly and Concisely](https://skills.sh/obra/the-elements-of-style/writing-clearly-and-concisely), the eternally relevant style guide by Strunk & White for writing clearly and cutting fluff ruthlessly. This has some interesting side effects, notably AI models will tell me how much "myself" my writing is. Having a robot tell you that you need to write more like yourself is so excessively cyberpunk that I don't even know where to begin to explain it. It really does help when you're going through some mild medical stress like I am right now though. I've also made a skill [to help organize notes and brain dumps into deep dive posts](https://github.com/tigrisdata/tigris-blog/blob/main/.claude/skills/build-with-me-outline/SKILL.md). Maybe this will be useful, I have to do more experimentation to know for sure. ## Supercharge your agents with skills[​](#supercharge-your-agents-with-skills "Direct link to Supercharge your agents with skills") Agent skills won't solve every problem, but for that middle ground of "describable, verifiable, annoying to code" they're incredibly powerful. They've already saved us hours on image optimization and commit hygiene. As we discover more use cases, we'll add them to the library. Please be sure to keep an eye out for them! We'll post on [our Discord](https://community.tigrisdata.com/) as we discover and mint good ones. If working on things like this sounds neat to you, [we're hiring](https://www.tigrisdata.com/careers/). You'll get to work alongside me as we reinvent the future of how people interact with object storage. Otherwise, go out and use Tigris to change the world via B2B Saas! Ready to supercharge your AI agents? Install the Tigris Agent Skills Library and help your agents automate tasks like image optimization, conventional commits, and Tigris SDK operations. [Explore Tigris Skills](https://skills.sh/tigrisdata/skills) **Tags:** * [Engineering](/blog/tags/engineering/.md) * [AI](/blog/tags/ai/.md)