Keep noisy exploration out of the main conversation.
Claude Code Atlas · free subagent library
Download a specialist.
Run it for free.
Choose a focused helper for research, review, testing, debugging, or documentation. Download the complete file, place it in your project, and ask Claude Code to use it.
THE SHORT VERSIONOne bounded job. Its own context. A concise result returned.
No account, payment, repository upload, or ceremonial comment beneath a video required.
Give a specialist only the tools needed for its job.
Define what evidence and conclusion must come back.
Three steps. No platform to join.
Download. Place. Ask.
The files are plain Markdown. You keep them, inspect them, edit them, and remove them whenever you like.
- 01
Download the agent
Choose one of the ready-made specialists below and download its complete
.mdfile. - 02
Put it in your project
Place the file inside
.claude/agents/. Create the folders if they do not already exist. - 03
Ask Claude to use it
For example: “Use the code-reviewer subagent to review the changes I just made.”
Review an agent’s instructions and permissions before you run it. Free does not mean infallible.
Ready-made and free to keep
Eight useful specialists.
None with a blank cheque.
Each agent starts with the smallest practical permission set. Download it as supplied, or open it in the builder and change the brief before you install it.
Read-only researcher
Investigates a codebase and returns evidence without changing files.
- Use when
- You need file discovery, dependency tracing, or an answer supported by repository evidence.
- Do not use when
- The answer is already in the main conversation or the task requires implementation.
Code reviewer
Reviews a completed change for correctness, risk, and maintainability.
- Use when
- A meaningful change is ready for an independent read-only review.
- Do not use when
- You still need the same worker to implement the change it is reviewing.
Bug investigator
Traces a reproducible failure and proposes the most likely cause.
- Use when
- You have observed behaviour, expected behaviour, reproduction steps, or logs.
- Do not use when
- The problem statement is still too vague to identify a testable failure.
Test planner
Finds risk boundaries and proposes tests before implementation begins.
- Use when
- You want acceptance criteria and meaningful coverage before code is changed.
- Do not use when
- The task is a trivial, already-covered mechanical edit.
Security checker
Reviews a bounded surface for trust, secrets, permissions, and abuse paths.
- Use when
- A change touches authentication, input handling, data access, secrets, or external tools.
- Do not use when
- You need a formal penetration test or compliance certification.
Documentation verifier
Checks a product claim against current first-party documentation.
- Use when
- Commands, APIs, product behaviour, or version-specific claims may have changed.
- Do not use when
- The question can be answered solely from stable repository code.
Accessibility reviewer
Reviews interface code and rendered behaviour against practical accessibility checks.
- Use when
- A UI feature is ready for keyboard, semantics, contrast, and responsive review.
- Do not use when
- No interface or rendered behaviour is available to inspect.
Performance auditor
Finds measurable loading, rendering, and bundle bottlenecks.
- Use when
- A page or workflow has a defined performance concern and can be measured.
- Do not use when
- There is no baseline, route, or workload against which to compare.
Subagent Lab
Decide. Build. Check.
Use only the part you need. Everything runs in this browser; Sirrius Atlas does not receive your configuration.
Mark every statement that is true
Use the right context boundary
Not everything called an agent
is the same thing.
Choose based on how long the work runs, whether it must report into the current conversation, and whether workers need to coordinate with each other.
Subagent
Best useOne focused side task inside the current session
What comes backA result to the parent conversation
Forked skill
Best useA reusable procedure that should run in isolated context
What comes backThe skill’s result to the invoking flow
Background agent
Best useA separate, longer-running session
What comes backA session you monitor or attach to
Agent team
Best useSeveral sessions that communicate and coordinate
What comes backCoordinated work across teammates
Product facts are based on current official Claude Code documentation. Version-specific fields should still be checked against your installed release.
Read the official subagent documentationSubagents · questions answered
Before you install a specialist.
The context boundary, file location, permission model, and privacy details worth understanding before you copy a definition.
01Are the downloadable subagents really free?
Yes. The ready-made subagent files are free to download, inspect, edit, and use. No account or email address is required. You still need access to Claude Code, whose own plan or usage charges are separate from Sirrius Atlas.
02What is a Claude Code subagent?
A subagent is a specialized AI assistant that handles a focused task in its own context window and returns a result to the parent conversation. A custom subagent can have its own instructions, tool access, model, permissions, skills, memory, and execution settings.
03When should I use a subagent?
Use a subagent when a bounded side task would fill the main conversation with file reads, search results, logs, or other working context that will not be needed later. Keep small, tightly connected work in the main conversation.
04Does the Subagent Lab access my repository?
No. The generator and checker run entirely in your browser. Sirrius Atlas does not receive your configuration, access your repository, execute the definition, or connect to Claude Code.
05Where do custom subagent files go?
Project-specific subagents go in .claude/agents/ and can be committed with the repository. Personal subagents available across projects go in ~/.claude/agents/ on your machine.
06How do I run a downloaded subagent?
Place its Markdown file in your project’s .claude/agents/ folder, start Claude Code in that project, and ask Claude to use the named subagent for a bounded task. For example: “Use the code-reviewer subagent to review the changes I just made.”
07What is the safest tool configuration?
Start with the smallest allowlist that can complete the task. A research or review agent commonly needs Read, Grep, and Glob, and sometimes Bash. Add Edit, Write, skills, web access, or MCP servers only when the delegated job requires them.
08Is a subagent the same as an agent team?
No. A subagent handles a focused task within a session and reports a result to the parent. Agent teams are multiple sessions that can communicate and coordinate. Background agents and forked skills create different context and lifecycle boundaries again.
Need an agent built around your business?
Keep the free files.
Bring us the complicated part.
Sirrius Office designs custom agents and automations around real workflows, existing tools, and boundaries your team can understand.