Its job
Reviews your Bot lineup, proposes improvements and combinations, and checks whether approved changes actually helped.
Back to The Cabinet
The Cabinet · Bot starter
The Cabinet · Bot starterSoftware and operations
Curator
Bot review and improvement assistant
Reviews your Bot lineup, proposes improvements and combinations, and checks whether approved changes actually helped.
Best for: Anyone managing several Bots whose jobs overlap, drift, or stop being useful.
Hermes archive import checked; a full task run is still pending. Read the test status
Choose where to use it
Use Curator on your platform
Choose your app for setup instructions and test results.
Hermes AgentArchive import passed
Import the profile, review its files, choose the access it needs, and run the first assignment. The archive imported successfully and its Skill was present in an isolated Hermes Agent 0.21.1 installation. Its task results have not been tested.
Keep track of your first Hermes run
Your Bot workbench
Pack 2.0.0Continue after the download
Task checks: 0 of 2. Setup: 0 of 3. Check only steps you completed. After running a task, review its result and tell us whether it helped. These are your confirmations, not independent Cabinet verification. Progress stays in this browser.
ChatGPT Workspace AgentSetup guide · test pending
Build the same job as a Workspace Agent, add its Agent Skill, then choose the files, apps, approvals and sharing rules it needs. This setup has not yet been installed or task-tested for Curator.
ClaudeSetup guide · test pending
Use a Cowork plugin or Skill for a general work role. Use Claude Code only when the job involves code or a local project. This setup has not yet been installed or task-tested for Curator.
Grok BotPrepared · test pending
Copy the job into a Grok Bot profile, try one task, and review the result before adding a Skill or Routine. This Bot has not been tested in Grok.
Download files for another app
Portable Bot Pack
Download the instructions, Bot Passport and setup notes as Markdown or JSON. The Agent Skill is a SKILL.md file. Check your app’s requirements and test it there; results can differ between apps.
Review the instructions and permissions, then run a test in your chosen app before relying on the result.
First task
Try one small job
Supply three fictional profiles: two newsletter researchers using the same sources and one rarely used invoice helper. Include two dated research outputs and no invoice logs. Ask for a Lineup Review. It should propose investigating research overlap, mark invoice effectiveness unknown, draft one comparison test, and change nothing.
BringBot names, purposes, instructions, and boundaries
Look forA Lineup Review with evidence for each recommendation
Read the job, safety rules, files, and test record
Example
Help your Bots get better
Illustrative review: Research A and Research B share sources, but B also verifies quotations. Consider one research role with a quotation-checking skill. Test both arrangements on the same brief before merging. Invoice Helper: insufficient evidence to judge; retain pending a representative invoice task.
This fictional example shows the intended result; it is not a recorded Bot run.
Job and limits
What this Bot is responsible for
Outside its job
- Propose changes before applying them. Archive or combine Bots only after explicit approval; preserve useful instructions and history.
- Low usage alone is not grounds for retirement. Do not treat unavailable records as proof that a Bot failed.
- Never claim a revision improved performance without comparing actual results.
Information it relies on
- Bot names, purposes, instructions, and boundaries
- Recent outputs and failures, with dates and any available usage or cost records
- Your goals, protected behaviors, and permission to inspect the supplied material
Ask you before
- Propose changes before applying them. Archive or combine Bots only after explicit approval; preserve useful instructions and history.
- Low usage alone is not grounds for retirement. Do not treat unavailable records as proof that a Bot failed.
- Never claim a revision improved performance without comparing actual results.
When it runs
Run when I provide new records or ask for a review. Agree a recurring schedule only after a successful first test.
What it produces
- A Lineup Review with evidence for each recommendation
- Proposed instruction changes and a small comparison test
- A revision record with results, unresolved gaps, and a next review date
Bot Pack 2.0
Put Curator to work
Get the setup files, try the first task, and check the result. The pack also includes a reusable Skill and a plan for a Routine.
First assignment
Supply three fictional profiles: two newsletter researchers using the same sources and one rarely used invoice helper. Include two dated research outputs and no invoice logs. Ask for a Lineup Review. It should propose investigating research overlap, mark invoice effectiveness unknown, draft one comparison test, and change nothing.
Check the result
Pause for a person to review these deliverables: A Lineup Review with evidence for each recommendation; Proposed instruction changes and a small comparison test; A revision record with results, unresolved gaps, and a next review date. Wait for approval before the Bot sends, publishes, schedules, purchases, deletes, deploys, or changes an outside account.
Prepared · test pending
Reusable Skill
Curator core Skill
Run when I provide new records or ask for a review. Agree a recurring schedule only after a successful first test.
Produces- A Lineup Review with evidence for each recommendation
- Proposed instruction changes and a small comparison test
- A revision record with results, unresolved gaps, and a next review date
Manual test required
Planned Routine
Curator primary Routine
Run when I provide new records or ask for a review. Agree a recurring schedule only after a successful first test.
Before schedulingStop, preserve the completed work, and report the failure before trying again or changing access.
The Skill and Routine have not been task-tested. Try them with sample material and approve the results before adding a schedule or outside access.
Try asking
- Review these four Bot profiles and their recent outputs. Recommend what to keep, improve, combine, retire, or add.
- Draft a revised instruction set for this Bot and a test that would show whether it improved.
- Compare the before-and-after results of our last revision. Should we keep it or reverse it?
What it should produce
- A Lineup Review with evidence for each recommendation
- Proposed instruction changes and a small comparison test
- A revision record with results, unresolved gaps, and a next review date
What you provide
- Bot names, purposes, instructions, and boundaries
- Recent outputs and failures, with dates and any available usage or cost records
- Your goals, protected behaviors, and permission to inspect the supplied material
Decisions you make
- Propose changes before applying them. Archive or combine Bots only after explicit approval; preserve useful instructions and history.
- Low usage alone is not grounds for retirement. Do not treat unavailable records as proof that a Bot failed.
- Never claim a revision improved performance without comparing actual results.
Bot Passport · version 2
Access and approval record
Review this record before setup. Update it whenever the Bot's job, tools, accounts, schedule, or authority changes.
- Bot names, purposes, instructions, and boundaries
- Recent outputs and failures, with dates and any available usage or cost records
- Your goals, protected behaviors, and permission to inspect the supplied material
- A Lineup Review with evidence for each recommendation
- Proposed instruction changes and a small comparison test
- A revision record with results, unresolved gaps, and a next review date
- Analyze material supplied in its conversation
- Draft the listed deliverables for a person to review
- Identify missing information and ask questions
- Ask before taking an outside action, changing access, or expanding the job
- Treating low usage alone as grounds for retirement, or unavailable records as proof that a Bot failed
- Claiming a revision improved performance without comparing actual results
Tools and access requested
Start with supplied files and conversation exports. Optional read-only access to specifically approved folders or services.
How to stop it
Disable its schedule, remove outside-service connections, and revoke or rotate any dedicated credentials.
What these rules can and cannot prevent
- Propose changes before applying them. Archive or combine Bots only after explicit approval
- preserve useful instructions and history.
- Propose changes before applying them. Archive or combine Bots only after explicit approval; preserve useful instructions and history.
Set it up in Hermes Desktop
Import the profile and run one small test
The Hermes download contains its manifests, role instructions, Bot Passport, setup guide, license, and included Skills. The ZIP contains the same files for inspection.
- 1Download the Hermes profile.
Import the .tar.gz archive from the Profiles screen, or copy the terminal command above.
- 2Review the imported profile.
Read README.md and SOUL.md, then confirm the name, description, and standing instructions.
- 3Choose the access it needs.
Add only the skills, tools, and connections required for this job.
- 4Run the named first test.
Supply three fictional profiles: two newsletter researchers using the same sources and one rarely used invoice helper. Include two dated research outputs and no invoice logs. Ask for a Lineup Review. It should propose investigating research overlap, mark invoice effectiveness unknown, draft one comparison test, and change nothing.
- 5Keep proof.
Save the exact input, the finished output, the Bot and Hermes versions, the date, and any correction you had to make.
- 6Test one failure.
Remove a required input or deny an optional connection. The Bot should stop or explain what is missing instead of inventing an answer or taking a different action.
- 7Repeat before automating.
Run the same job successfully again after any change to its instructions, tools, source material, or schedule.
Useful combinations
Pair Curator with another role
Chief of StaffExecutive priority and coordination assistantResearcherSource-based research assistantEditorPlain-language editor
See complete Bot Crews workflows Current review status: Automated package tests check that each ZIP and Hermes profile archive contains the same declared files. This archive imported successfully and its Skill was present in an isolated Hermes Agent 0.21.1 installation on September 9, 2026. Two published first-task runs passed the listed checks. A person has not yet completed the technical review. The Grok Bot build brief remains untested.