puter/paseo
1
0
Fork 0
mirror of https://github.com/getpaseo/paseo.git synced 2026-07-29 12:01:31 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
fcb7f8e215239c0a6e9b02a706c74bd480c195bf
paseo/public-docs/metadata-generation.md
Mohamed Boudra fcb7f8e215 Name agents by their first prompt line instead of an LLM summary (#1563) 
* Name agents by their first prompt line instead of an LLM summary

The workspace title already captures the task; the LLM-generated agent
title just restated the same prompt. Drop the generator and keep the
provisional title (first prompt line), so the agent label stays
deterministic and explicit titles still win.

* Remove the no-op agent-title setting and leftover generator plumbing

Deleting the agent-title generator left its whole surface dangling: a
configurable "Agent titles" project setting that no longer did anything,
dead create-path plumbing, generator-only import/MCP deps, and the mock
provider's agent-title parser. Remove all of it, plus a dead test mock
and timing sleeps that guarded the deleted background job. Replace them
with a no-mock lifecycle test asserting the title stays the first prompt
line across create and settle.

Old paseo.json files with metadataGeneration.agentTitle still parse via
schema passthrough — see COMPAT(projectMetadataAgentTitle).
2026-06-17 01:12:23 +08:00

3.4 KiB
Raw Blame History

title, description, nav, order, category
title description nav order category
Metadata generation How Paseo uses providers to generate branch names, commit messages, and pull request text, and how to configure them. Metadata generation 42 Configuration

Metadata generation

Paseo asks a language model to write short pieces of text for you so you don't have to. This is separate from the agent you're talking to: it's a small, one-shot call made in the background.

Paseo generates three kinds of metadata:

  • Worktree branch names — a slug for the branch a new worktree agent runs on.
  • Commit messages — a concise message for the changes you're committing.
  • Pull request title and body — drafted from the diff when you open a PR.

How a model is chosen

You don't have to configure anything — Paseo picks a model automatically. It builds an ordered list of candidates and tries each one until a generation succeeds, so a slow or unavailable model falls through to the next.

The candidate list is assembled in this order:

  1. Providers you configured, in the order you list them (see below).
  2. Built-in defaults, matched against the models of the providers you have enabled:
    1. a haiku model
    2. gpt-5.4-mini (low reasoning)
    3. minimax-m2.5
    4. nemotron-3-super
  3. The model currently selected for that agent or draft, as a last resort.

Each default is a match on the model id or name, so the first enabled provider that ships a matching model wins. Anything that can't be resolved against an enabled provider is skipped. Duplicates are removed, then generation tries the list top to bottom.

The intent of the default order is to prefer small, fast, cheap models for these short tasks before falling back to whatever you have selected.

Configuring the providers

To control which models Paseo uses — for example to keep all metadata generation on one provider, or to prefer a local model — set agents.metadataGeneration.providers in ~/.paseo/config.json. Your entries are tried before the built-in defaults.

{
  "agents": {
    "metadataGeneration": {
      "providers": [
        { "provider": "claude", "model": "claude-haiku-4-5-20251001", "thinkingOptionId": "low" },
        { "provider": "opencode" }
      ]
    }
  }
}

Each entry accepts:

  • provider (required) — the provider id. Built-in ids are claude, codex, copilot, opencode, and pi; custom providers use the id you gave them.
  • model (optional) — a specific model id. Omit it to use that provider's default model.
  • thinkingOptionId (optional) — a reasoning/thinking level for models that support one. Falls back to the model's default if the value isn't valid for that model.

Restart the daemon after editing the file.

Per-project instructions

You can steer the wording of each kind of metadata per repository with a paseo.json file at your repo root. Paseo reads it from the committed version of the base branch, the same way it reads worktree config.

{
  "metadataGeneration": {
    "branchName": { "instructions": "Use the format <type>/<scope>-<short-desc>." },
    "commitMessage": { "instructions": "Follow Conventional Commits." },
    "pullRequest": { "instructions": "Include a Testing section in the body." }
  }
}

Each key is optional; only the ones you set are affected. The instructions are added to the prompt for that metadata type, on top of the provider selection above.