Conversation
Tools/harbor-graphcode installs graphcode, graphcoded and zmx into a task container, creates one goal loop per task and waits for it to resolve, reusing Harbor's ClaudeCode install and trajectory conversion. The done check is either the task's own test.sh (an oracle) or the session's own node done. build-linux-bundle.sh builds the binaries in swift:6.2; docs/benchmarks/terminal-bench-trial.md holds the trial plan and cost estimate, not yet run. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
done_check=tests makes the task's own test.sh the loop's predicate, an oracle plain claude-code never gets, so it stays an opt-in upper-bound arm and done_check=agent becomes the default. 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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/harbor-graphcode, a Harbor installed agent that runs each Terminal-Bench task as one GraphCode goal loop, so a private trial can compare it with Harbor's plainclaude-codeagent on the same model. Nothing is added to Harbor and nothing is submitted anywhere.What it does
ClaudeCode.install, uploadsgraphcode-linux-<arch>.tar.gzand unpacks graphcode, graphcoded and zmx into/opt/graphcode/binbypassPermissions, every model alias pinned to Harbor's-m), creates one goal loop with the instruction as its goal, and pollsgraphcode statusuntil the loop resolves/logs/agent/sessions/projects, so the inherited converter writestrajectory.jsonagent(no predicate; the loop resolves ongraphcode node done), the fair arm; opt-intestsmakes the task's owntest.shthe predicate, an oracle, so only an upper boundVerification
bash -n; build scriptssh -nstoppedswift:6.2Trial status
No trial has run: no Docker and no
ANTHROPIC_API_KEY.docs/benchmarks/terminal-bench-trial.mdhas the plan, the prerequisites and a cost estimate (smoke ≈ $16, 15-task subset ≈ $120). It is force-added because/docs/*is gitignored.