Add Benchmarks & evals use case (Terminal-Bench via Harbor on E2B) - #321
Open
tomasvarga wants to merge 2 commits into
Open
Add Benchmarks & evals use case (Terminal-Bench via Harbor on E2B)#321tomasvarga wants to merge 2 commits into
tomasvarga wants to merge 2 commits into
Conversation
- Quickstart is now tabbed: Terminal-Bench (registry dataset) and ReactBench (local task dirs from a cloned repo), both verified end-to-end on E2B - ReactBench tab explains --allow-agent-host (offline tasks + per-phase network policy) and --override-memory-mb (build-time resource limits), with a Warning documenting the two temporary Dockerfile edits - New 'Bring your own eval' section: harbor init scaffold, task layout, SWE-bench-style fail-to-pass/pass-to-pass testing, and eval validation with oracle/nop control runs before spending on agents - All commands and output blocks are verbatim from verified runs
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Adds a new Use cases → Benchmarks & evals page (
docs/use-cases/benchmarks.mdx) plus its nav entry indocs.json.The page covers running coding-agent benchmarks on E2B via Harbor (the Terminal-Bench harness, which ships an E2B environment backend): why per-task sandbox isolation matters for eval integrity, a verified quickstart running the Terminal-Bench 2 sample set with Claude Code, and pointers to scaling up (full TB2, other registry datasets,
--skillfor with/without-skill comparisons, and e2b-dev/swe-bench).Verified
All commands were run end-to-end on 2026-07-29 with harbor 0.20.0:
pip install 'harbor[e2b]'→harbor run -d terminal-bench-sample@2.0 -a claude-code -m claude-sonnet-5 -e e2b -n 3Known limitation (documented in a Note callout)
Harbor currently hardcodes a 24h sandbox timeout at creation (
harbor/environments/e2b.py), which fails with400: Timeout cannot be greater than 1 hourson plans with a 1h sandbox lifetime cap. The page documents this; a Harbor upstream fix to make it configurable is a sensible follow-up.