Skip to content

crew: validate-roster support for codex/gemini/ollama/anthropic-api backends #67

@jdidion

Description

@jdidion

tools/validate-roster currently only introspects the cursor backend (via cursor-agent models). Other backends get a "skipping validation" warning.

Gap

Backends without catalog introspection:

  • codex — CLI is a stub on the backend side too; need to verify the real codex CLI's "list models" command and wire it in.
  • gemini — unclear if Google's gemini CLI has a list-models subcommand. Check docs.
  • ollamaollama list lists locally installed models. Easy.
  • anthropic-api — no public list-models endpoint documented; may need to skip permanently or hardcode known models.

Impact

Users hit a late-stage backend failure (e.g. codex run --model gpt-4.3 exits 1 with "model not found") instead of getting a clear parse-time error. The same footgun that motivated validate-roster in the first place, scoped per-backend.

Proposed

Implement the catalog fetch per backend and wire into the tool's case \"$backend\" in ... esac. Start with ollama (easiest) and codex (highest value since gpt-* defaults route to codex-first). Gemini and anthropic-api can stay as warning-skip if their CLIs don't expose listing.

Originally flagged in PR #61's follow-ups list and in tools/validate-roster's inline docs.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions