Solaris Dev Shop

Find the right specialist skill, route a task, and carry its evidence requirements into your own agent.

61 portable specialist skills across 12 departments, plus a chief-of-staff skill and an optional deterministic Python router. Free under MIT. No model calls, account or API key is needed to try the router.

The router produces an assignment packet with one accountable lead and, when needed, multiple supporting specialists. To do the work, open those skills in your own agent or agents with the tools and permissions that task needs. The router does not start agents or grant access to your accounts.

Native plugins

Use the native plugin repository for host-specific installation instructions, tested client versions and publication status. The packages keep the same canonical workflows and companion resources together and record their exact source revision. Native format support, model-driven results and provider-directory approval have separate evidence; check the recorded scope for your client and operating system.

Try it

Requires Git and Python 3.10+. This first run leaves your PATH unchanged:

git clone https://github.com/Shaisolaris/solaris-dev-shop.git
cd solaris-dev-shop
python3 control-plane/meta_control_plane.py skills playwright
python3 control-plane/meta_control_plane.py intake --explain "Write a test plan for the client portal login regression"

On Windows, use python in place of python3.

The intake selects solaris.qa-engineer and returns:

{
  "decision": "assign",
  "assignment": {
    "primary_capability": "solaris.qa-engineer",
    "skill_path": "employees/quality-security/qa-engineer/SKILL.md"
  },
  "performed_work": false
}

This is an excerpt; the full packet includes authority bounds, required evidence, confidence and runner-up. Confidence measures keyword evidence, not the probability that a result is correct. An unclear goal asks for clarification. Recognized high-risk wording triggers a human-review gate. Risk matching is lexical and does not grant permission for actions that are not flagged.

Next: open the returned SKILL.md in your agent, keep its companion files together, and supply the task and workspace. Follow one complete handoff, browse the 61 skills, or use the Actions playground without a local install.

Watch the 65-second narrated walkthrough · Transcript, captions and reproducible commands

Choose a skill

After the optional ./install.sh setup, the same commands are available as solaris-intake:

solaris-intake skills accessibility --department quality-security
solaris-intake skills "landing page" --json
solaris-intake intake --explain "review this code"

Search covers public names, summaries and tags. All query terms must match. Results include the source-tree path and GitHub link; searching does not install or execute a skill. The Python wheel contains the router and catalog, while the complete skill folders live in this repository.

Work with multiple specialists

One assignment can include several specialists working toward the same outcome:

python3 control-plane/meta_control_plane.py intake --explain "Build a FastAPI REST API backend; design Figma wireframes; write a Playwright test plan"

The packet names solaris.backend-developer as the accountable lead and includes solaris.ui-ux-designer and solaris.qa-engineer in supporting. Each supporting role has its own source path in supporting_handoffs. The host agent can coordinate their work within the packet's scope, using parallel workers when it supports them and the tasks are independent. A single host can also use the skills in sequence.

The router recognizes bounded task separators such as and, semicolons and newlines; it does not infer a team from every competing keyword match. The default limit is five supporting roles. All roles share the original authority bounds, and performed_work: false means no specialist has run yet. See the multi-specialist handoff and routing limits.

What is included

Use a skill directly when you already know the specialist you need. Use intake when the task spans departments or you want a reviewable handoff. Architecture and privacy explain the boundaries and opt-in local follow-up context.

Executable workflows

Selected specialists now include local tools for recoverable delivery, data reconciliation, experiment readouts, grounded-answer checks, duplicate-safe APIs, offline synchronization and evidence-bound review. Start with the capability guide and runnable examples. These helpers extend the existing roles; the router still assigns work rather than executing it.

Install and portability

The source quickstart above is the smallest trial. ./install.sh adds a managed solaris-intake command after checking the checkout. Install instructions cover Windows, a local wheel, uninstall and copying skills into a host agent.

Provider-neutral means the skill text and routing packet do not require a particular model API. It does not mean every host supports the same tools or that outputs have equal quality across models. Review the selected skill's tools, references and scope before use.

Quality and evidence

• A reproducible first-task handoff
• Routing benchmark and limits
• Multi-domain example and safe escalation example
• Known limits and FAQ
• Structural audit: python3 employees/quality-security/assurance/quality_os.py audit --html report.html

CI runs the full capability suite on supported Linux/macOS runtimes. Windows checks the router, assurance behavior, explicit platform boundaries and an installed wheel outside the checkout; POSIX-dependent helpers do not claim Windows execution support. Catalog and YAML checks cover all 61 specialists plus coordinator discovery metadata. The structural audit checks documented output contracts and QA gates; it does not certify real-world skill outputs. The routing benchmark measures local Python routing, not model quality or autonomous task completion.

Around the community

Solaris Dev Shop is listed on DevShelf, SilentStars and other community directories. See verified public listings for entries and their evidence. Directory inclusion does not imply an endorsement. Project facts and media provide reusable, evidence-backed descriptions.

AI Agents Directory now has a live Solaris Dev Shop listing, verified October 15, 2026. Its status is Standard / New. The publisher backlink below does not imply an editorial award:

Postmake's free listing route requires its badge. This backlink is in place for that application; a Postmake listing has not yet been confirmed:

Contribute

• Start small: follow the contributor quickstart, then choose an unclaimed good first issue or help wanted task.
• Found a routing miss? Open a bug with the command, expected specialist and observed result. Use synthetic examples without credentials or private data.
• Want to expand the library? Propose a new specialist with its missing use case, boundaries and worked example. The current 61-skill roster can grow through reviewed contributions.
• Improving the router? Run python3 -m unittest discover -s tests -v and python3 control-plane/meta_control_plane.py self-test.
• See the roadmap, current issues or show what you built.

Contributors

License

MIT. See LICENSE and NOTICE.md. Do not commit passwords, API keys, tokens or private account details.