Skip to content

generator: --append mode for automated board PRs#6

Merged
lakshvantb merged 1 commit into
mainfrom
generator-append-mode
Jul 22, 2026
Merged

generator: --append mode for automated board PRs#6
lakshvantb merged 1 commit into
mainfrom
generator-append-mode

Conversation

@lakshvantb

Copy link
Copy Markdown
Collaborator

Adds two flags to scripts/generate_model_rows.py so the eval workflow can build board PRs without ad-hoc append code:

  • --append PUBLIC_DIR — writes the generated table + cost rows into the release CSVs, preserving each file's own line endings (table LF, cost CRLF), refusing duplicates, and re-verifying its own rows after writing (round-trip --verify).
  • --links-meta '{...}' — with --append, inserts the modelLinks.js entry (url, organization, displayName, openweight/reasoner flags) before the map's closing brace; skips if an entry exists.

Validation

Ran --append for mimo-v2.5-pro on a clean main checkout and byte-diffed against the hand-built #5 branch: all three files identical. Re-running refuses with "row already exists". check_data.js green (41/41, no orphans).

Companion: the livebench-private workflow step that calls this (board-PR automation).

🤖 Generated with Claude Code

--append PUBLIC_DIR writes the generated table/cost rows directly into the
release CSVs, preserving each file's own line endings (table LF, cost CRLF)
and refusing duplicate model rows; --links-meta JSON additionally inserts
the modelLinks.js entry (url/organization/displayName + openweight/reasoner
flags). After writing it re-verifies its own rows against the CSVs.

Byte-for-byte validated against the hand-built add-mimo-v2.5-pro branch:
all three files identical; duplicate append refused; check_data green.
This is the piece the eval workflow's board-PR step calls.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant