Conversation
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A loop that fails to launch shows as stopped and never replies, so polling for the reply alone waited out the whole turn timeout on every scenario retry. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds
Tools/clawarena-graphcode, which runs a GraphCode loop as the main agent (manager) of ClawArena-Team for private trials, plusdocs/benchmarks/clawarena-team-trial.md(feasibility, comparability, pool hosting, cost). Nothing is added to ClawArena, nothing is submitted, no model was called.How it plugs in
ClawArena scores management from its own tool traces, so GraphCode joins as the
mainprovider rather than replacing the harness:provider: graphcode): each harness model call writesturn-NNNN.json; the loop answersreply-NNNN.jsonwith tool calls or a final answer; the harness still executes, sandboxes and scores everyCreateSubagent/RunSubagent/Workflowgraphcode node send(a goal loop's Stop hook would keep it busy between turns)clawarena-graphcode-dry-runVerification
CLAWARENA_TEAM_DATA)s_observability_incident630efd8a)claudehidden, zero spend)Trial status
Not leaderboard-comparable: a scaffold around a model is a disclosed deviation; the meaningful result is a paired run (GraphCode vs the same model via the stock
anthropicprovider) on one pool. Blocked on a subagent pool (no NVIDIA GPU; hosted gemma-4-31b exists, hosted e4b with audio unconfirmed) and on approved spend. Estimate, ±2×: paired smoke ≈ $5–15, paired full run ≈ $180–630, nearly all main-agent API. The report is force-added because/docs/*is gitignored.