Tigris Data Skills - Setup Guide¶

Source: tigrisdata/skills via skills.sh - 28.1K combined installs across 19 indexed listings; first seen 2026-10-31 (morning sweep)

GitHub: tigrisdata/skills - 2 stars, MIT, branch main, last pushed 2026-07-23; 18 SKILL.md files, one per directory under skills/, each paired with a matching .zip distribution package generated by a script in scripts/

Category: Globally distributed S3-compatible object storage: buckets, access control and SDK usage for agents

Quality Tier: 🟡 Yellow - one sampled skill carries a Socket warn and the repository has 2 GitHub stars, so the verdict sample is thin and not uniformly clean

Tigris maintains this as the skill set for its own object storage platform, and the coverage follows the product rather than a tutorial: file-storage is the onboarding path (CLI setup, access keys, the @tigrisdata/storage SDK, and framework guides for Next.js, Remix, Express, Rails, Django and Laravel), while tigris-object-operations, tigris-bucket-management, tigris-static-assets, tigris-image-optimization, tigris-lifecycle-management, tigris-egress-optimizer and tigris-backup-export each take one workload. The platform-specific skills sit on top: tigris-snapshots-forking and tigris-snapshots-recovery for point-in-time state, tigris-agent-kit for forks, per-agent workspaces with TTL, checkpoints and bucket-webhook coordination, and tigris-s3-migration for moving off S3, GCS or Azure Blob with shadow buckets and an active drain.

Two things make the repository broader than its category line. It ships conventional-commits and go-table-driven-tests, which are general engineering skills with no storage content at all, and it ships openclaw-backup, which backs up an agent's own state directory to a Tigris bucket with snapshots and forks as the recovery mechanism. The second is the most Hermes-relevant skill in the set even though it is the least installed.

Installation¶

The README documents one quick install for everything, then a manual copy for Claude Code and a paste route for claude.ai:

npx skills add tigrisdata/skills

cp -r skills/{skill-name} ~/.claude/skills/

For this profile the durable install is a clone plus a directory-by-directory copy. Copy the directories rather than the whole skills/ tree, because each skill also has a .zip beside it and the glob below matches directories only:

git clone --depth 1 https://github.com/tigrisdata/skills.git /tmp/tigris-skills
for d in /tmp/tigris-skills/skills/*/; do cp -r "$d" ~/.hermes/profiles/corpusiq/skills/; done

That installs 18 skills. If the profile only needs storage in one framework, take file-storage plus the workload skill that matches the task (tigris-object-operations for raw reads and writes, tigris-static-assets for a build pipeline, tigris-python-sdk when the codebase is Python and boto3), and skip the six skills below the 500 install line, which are narrower operational runbooks.

What It Provides¶

Skill	Installs	Use For
tigris-object-operations	3,409	Uploading, downloading, deleting, listing and inspecting objects with the Tigris SDK: `put` with multipart for large files, `get` as string, file or stream, `list` with prefix and pagination, `head` for metadata, and presigned URLs for temporary access.
tigris-security-access-control	3,391	Hardening a bucket: access key lifecycle and rotation, bucket policies and visibility, CORS rules, presigned URL security, and a security audit checklist.
file-storage	3,388	Getting started with Tigris file storage end to end: CLI authentication, bucket creation, access key creation and assignment, the SDK call surface, browser uploads via `handleClientUpload`, and framework guides for Next.js, Remix, Express, Rails, Django and Laravel.
tigris-sdk-guide	3,373	Choosing an SDK per language: when to use the Tigris-native SDKs and CLI, when to fall back to AWS S3-compatible SDKs, and how to configure the fallback when there is no other option.
tigris-image-optimization	3,367	Resizing, cropping and optimising images stored in Tigris: upload-time variant generation with Sharp, ImageMagick, Pillow or Intervention, responsive delivery through `next/image`, Active Storage or django-imagekit, and CDN delivery from a public bucket.
tigris-static-assets	3,352	Deploying CSS, JavaScript, fonts and build artifacts to Tigris: asset pipeline integration, cache-busting, `Cache-Control` configuration and global CDN delivery across the major frameworks.
tigris-s3-migration	3,350	Moving from AWS S3, Google Cloud Storage or Azure Blob with zero downtime: shadow buckets via `tigris buckets set-migration`, active drain via `tigris buckets migrate`, bulk copy, and an SDK endpoint swap.
tigris-agent-kit	3,335	Agent storage workflows on `@tigrisdata/agent-kit`: forks for isolated dataset copies, per-agent workspaces with TTL, checkpoints for snapshot and restore, and bucket-webhook coordination for multi-agent pipelines.
tigris-snapshots-forking	444	Point-in-time state for testing and sandboxes: `createBucketSnapshot`, `listBucketSnapshots`, reading from a snapshot version, instantiating a fork as an isolated copy, and deletion protection patterns.
tigris-bucket-management	126	Creating, listing, inspecting and deleting buckets: `createBucket` with access level, region and snapshot enablement, pagination on `listBuckets`, `getBucketInfo`, and `removeBucket` including force delete.
conventional-commits	100	Writing commit messages in the Conventional Commits format, with no storage content: a general version-control skill that happens to live in this repository.
go-table-driven-tests	93	Writing Go table-driven tests: table fields for name, input, want, wantErr, errCheck and setupEnv, integration guards that skip without credentials, `t.Helper()` helpers, custom error validation and `t.Parallel()`.
tigris-lifecycle-management	81	Automating object expiry and tier transitions: TTL-based deletion, cleanup rules for temporary files, and archival policies that control storage cost.
tigris-egress-optimizer	77	Diagnosing high storage costs: a four-step framework over caching, CDN delivery, presigned URLs and `Cache-Control` headers, aimed at egress and bandwidth.
tigris-backup-export	76	Backing up databases and exporting application data to Tigris: scheduled export pipelines, retention policies and restore workflows across the major frameworks.
tigris-snapshots-recovery	75	Recovering deleted or overwritten files on a snapshot-enabled bucket, restoring objects to an earlier state, and enabling snapshots before an incident rather than after one.
tigris-python-sdk	43	Using Tigris from Python: boto3 setup, Django uploads through django-storages, snapshots, forking, in-place rename, conditional writes with IfMatch and IfNoneMatch, and the Bundle API for batch machine-learning fetches via `tigris-boto3-ext`.
installing-tigris-storage	37	Indexed on skills.sh with no matching directory in the clone, so there is no `SKILL.md` to route to. The setup path it implies is covered by `file-storage`, which is the skill to install.
openclaw-backup	11	Backing up and restoring an agent state directory to a Tigris bucket: scheduled backups, safe SQLite handling, point-in-time rollback through bucket snapshots and zero-copy clones through forks.

All 19 indexed listings clear the threshold and appear above. One of them, installing-tigris-storage, has no directory in the tree, so the install copies 18.

Why This Matters for Hermes Agents¶

Object storage is where an agent's work outlives the session: uploaded files, generated assets, exported reports, database dumps, and the agent's own state. The skills here are written as operating instructions rather than API docs, so an agent asked to "upload this file" gets the full sequence, access key handling included, instead of a bare put call that fails on credentials.

The platform-specific skills are the differentiator against generic S3 knowledge. Forks and snapshots turn "give the agent a scratch copy of the production dataset" into a documented operation, tigris-agent-kit wraps that in per-agent workspaces with a TTL, and tigris-s3-migration covers the endpoint swap that makes the platform reachable from existing code. None of that transfers from an S3 tutorial.

openclaw-backup is the skill to read even if nothing else here is installed. It backs up an agent state directory to a bucket, handles SQLite safely rather than copying a live database file, and uses snapshots and forks for rollback, which is the same shape a Hermes profile needs for its own data/ and skill directories. At 11 installs it is the least-used skill in the repository and the most directly applicable to this profile.

Usage¶

You say	What happens
"Set up Tigris storage in this Next.js app"	`file-storage` walks the CLI setup (authenticate, create bucket, create and assign keys), then the SDK calls and the framework-specific upload path.
"Upload this report and give me a link that expires in an hour"	`tigris-object-operations` uploads with `put` and returns a `getPresignedUrl` result scoped to the requested lifetime.
"Our storage bill jumped, find the egress"	`tigris-egress-optimizer` runs its four-step framework over caching, CDN, presigned URLs and `Cache-Control`, then proposes fixes with an estimate.
"Give the test suite an isolated copy of the production bucket"	`tigris-snapshots-forking` captures a snapshot and instantiates a fork, so the tests read real data without touching the original.
"Move us off S3 without downtime"	`tigris-s3-migration` sets up a shadow bucket, drains writes through `tigris buckets migrate`, and swaps the SDK endpoint once the copy settles.
"Lock down public access on this bucket"	`tigris-security-access-control` covers bucket visibility, CORS rules, key rotation and presigned URL scope, with the audit checklist at the end.
"Back up the agent state directory before the upgrade"	`openclaw-backup` schedules the backup to a bucket, handles the SQLite files safely, and leaves a snapshot to roll back to.

Verification¶

The raw path for the most-installed skill returned HTTP 200 on branch main when this guide was written on 2026-10-31:

curl -sL -o /dev/null -w "%{http_code} tigris-object-operations\n" \
  https://raw.githubusercontent.com/tigrisdata/skills/main/skills/tigris-object-operations/SKILL.md

Expect 200. After the copy lands, confirm the count and that no .zip packages came along:

ls -d ~/.hermes/profiles/corpusiq/skills/tigris-*/ \
      ~/.hermes/profiles/corpusiq/skills/file-storage \
      ~/.hermes/profiles/corpusiq/skills/openclaw-backup \
      ~/.hermes/profiles/corpusiq/skills/conventional-commits \
      ~/.hermes/profiles/corpusiq/skills/go-table-driven-tests | wc -l
head -4 ~/.hermes/profiles/corpusiq/skills/tigris-object-operations/SKILL.md

The count should print 18, and the header should show name: tigris-object-operations with a one-line trigger-led description and no license field, since the licence here is the repository LICENSE file rather than a per-skill declaration.

Security¶

One sampled skill carries a warn and the project is barely starred: tigris-image-optimization returns a Socket Warn, and the repository has 2 GitHub stars, so the sample behind these verdicts is thin and one entry in it is not clean.

skills.sh verdicts for sampled skills (verified 2026-10-31) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
tigris-object-operations	Pass	Pass	Pass
tigris-image-optimization	Pass	Warn	Pass
tigris-bucket-management	Pass	Pass	Pass
tigris-backup-export	Pass	Pass	Pass
openclaw-backup	Pass	Pass	Pass

The mechanical surface is small. Every skill is markdown, there is no package to install and no postinstall hook, and the only executable in the tree is scripts/generate-deployment-packages.sh, a bash script that builds the .zip distributions from the skill directories and is not something an install runs. The two skills worth reading before you run them are the ones that touch credentials and state: tigris-security-access-control manages access keys and bucket policies, and openclaw-backup reads an entire agent state directory, including SQLite databases, and writes it to a bucket.

The credentials question is the one to settle first. Tigris access runs on access keys created through the CLI, and the skills describe creating, assigning and rotating those keys, so the agent will be handling secret material by design. Keep keys in environment variables or a local config the agent reads by reference, never echoed into a prompt or written into a generated file, and scope each key to the bucket it needs. A public bucket plus an over-broad key is the failure mode this suite will happily build for you if the request is phrased casually.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified 2026-10-31: 28,128 combined installs across 19 indexed listings; 2 GitHub stars; MIT; last pushed 2026-07-23. Counts drift over time.
The index runs ahead of the tree. Nineteen listings are indexed but the clone holds 18 directories: installing-tigris-storage resolves to nothing under skills/, and the setup path it implies is covered by file-storage.
Every skill directory has a .zip twin in skills/, 18 archives in total, built by scripts/generate-deployment-packages.sh. A blanket cp -r skills/. copies 36 entries and leaves archives in the profile's skill directory, which is why the install loop above matches directories only.
Install counts are not a ranking. Eight skills cluster between 3,335 and 3,409 installs and then the series falls off a cliff to 444 and below, which reads as one bulk import rather than eight independent adoption decisions; openclaw-backup sits at 11 while being the most directly useful skill in the repository for an agent profile.
The star count is 2, so third-party scrutiny is minimal. The verdicts on skills.sh are a sample, and a two-star repository has few eyes on changes between the last push and the next one.
Three skills overlap on snapshot behaviour: tigris-snapshots-forking, tigris-snapshots-recovery and the snapshot options inside tigris-bucket-management. An agent asked to restore a file can plausibly route to any of them, and only the recovery skill is written around restoring rather than creating.
Two skills have nothing to do with storage. conventional-commits and go-table-driven-tests are general engineering skills bundled here, so a storage-only install can skip them without losing anything the category line promises.

Skills Catalog | Skills Marketplace

Pro Tips¶

Install file-storage first whatever else you take. It is the only skill that covers authentication, bucket creation and access key assignment in one place, and every other storage skill assumes that setup already happened.
Skip tigris-sdk-guide on a fresh install unless the codebase is already on an AWS SDK. It answers "which SDK do I use per language", a question the agent usually resolves from the installed packages.
Copy directories, not the tree, so the 18 .zip packages stay out of the profile's skill folder; the loop in the installation section matches */ for exactly that reason.
Read openclaw-backup before building any backup for this profile's own state. Its safe SQLite handling and snapshot-based rollback are the pattern, even if the target bucket and schedule differ.
Ask for a fork rather than a copy when an agent needs production-shaped test data. Snapshots plus forks give an isolated, instant copy, and it is the one capability here that has no clean S3 equivalent.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ