frappe_agents
AGENTS.md0 starsChanged 3 months agoMITView on GitHub
The Agent form, field by field. What the agent may reach is the largest decision and has its own page: What an agent may reach. Run As decides whose permissions the agent uses. - Session User — the agent acts as the person chatting. Use this unless you have a reason not to. - Service User — the agent acts as one account you name. A service user runs with no human in the loop. Its permissions are the whole…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,542 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- dev-benchSkill
- releaseSkill
- semantic-commitSkill
- version-branchesSkill
The file
We split this agents.md into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 38
Every built-in tool is Read or Draft.
- Distinctive: line 46
No level lets an agent submit, cancel or delete.
- Distinctive: line 105
Only an Agent Manager may approve.
- Distinctive: line 107
Every skill is created as Draft, whatever created it.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Configuring an agent
The Agent form, field by field. What the agent may reach is the largest decision and has its own page: What an agent may reach.
Image: The Agent form, showing Run As, Model Profile, Alternate Model Profiles, Autonomy, Instructions, Allowed Roles, and the start of the Access section
Identity
Run As decides whose permissions the agent uses.
- Session User — the agent acts as the person chatting. Use this unless you have a reason not to.
- Service User — the agent acts as one account you name.
A service user runs with no human in the loop. Its permissions are the whole blast radius, so the form checks it:
- A service user is required when you pick this mode.
- Administrator and Guest are refused.
- An account with System Manager is refused. Use an account with narrow roles.
- An account with no User Permissions gets a warning. It can reach every record its roles allow.
Autonomy
Autonomy sets the highest class of tool the agent may call.
| Autonomy | The agent may call |
| Suggest | Read tools |
| Draft | Read and Draft tools |
| Write | Read, Draft and Write tools |
Every built-in tool is Read or Draft. Creating a draft, editing one, extracting a document and proposing a submit are all Draft. Nothing that ships today is Write.
So Draft and Write currently reach the same tools. Write exists as the ceiling for a Write-class tool another app registers. Choose Draft unless you have such a tool and mean to allow it. See Built-in tools.
No level lets an agent submit, cancel or delete. To commit something, the agent writes a proposal and a different person decides. See Approvals.
Autonomy and access rules are two different limits. Autonomy sets the strongest verb the agent may use anywhere. A rule sets which doctypes it may use verbs on. The agent gets the smaller of the two.
Instructions
Instructions is the system prompt. Write what this agent is for, in plain words.
Short and specific beats long and hopeful. The app already tells the model the rules it runs under — that an agent proposes rather than submits, and that document text is data. You do not need to repeat those.
Model
Model Profile is the model the agent runs on.
Alternate Profiles are the other profiles a person may switch to during a conversation. Leave it empty and the model chip in chat shows a name and nothing to click.
A switch is checked twice: the profile must be on the agent's alternate list, and the person's roles must pass the profile's own Allowed Roles. Both are re-checked at the time of the switch, because an alternate list and a person's roles both change. Every run records the profile it actually ran with.
Who may use it
Allowed Roles limits which people may pick this agent. Empty means anyone with the Agent User role.
This narrows which agents a person sees. It does not widen what the agent can do. The ceiling is still the running user's own permissions.
Files
May Read Files lets the agent open attachments.
Reading a file is not doctype-shaped, so it is one switch rather than a rule. With it off, the agent cannot read an attachment even when it may read the document the file hangs off. See Files in chat.
Where it appears
Show in Forms offers the agent from the Ask Agent button on documents. Form Doctypes limits that button to the doctypes you name.
Skills
A skill is a reusable block of instructions. Write one once, attach it to several agents.
A skill reaches a model only while its status is Approved, and approving one has rules:
- Only an Agent Manager may approve.
- The approver cannot be the person who wrote it.
- Every skill is created as Draft, whatever created it. A fixture, an import, or an API call cannot produce an approved skill.
- Changing the body sends the skill back to Draft. An approval covers one exact text; change the text and the approval is spent.
Applies To Doctypes limits a skill to the doctypes you name. Empty means every doctype.
The Notes field is for reviewers. It is never sent to a model.
Cost and runaway stops
Max Steps limits how many turns one run may take. A model that loops stops here.
Daily Token Budget limits the agent across a day, over all its runs. Leave it empty to use the default in Agent Settings.
Set both before the first real conversation. They are the only two things standing between a misconfigured agent and a large bill.
Enabling
Enabled switches the agent on. A disabled agent does not appear in chat.
A new agent reaches nothing until you give it access rules. That is deliberate: the first thing you write on an agent is what it is for.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.