highperfocused/multica
1
0
Fork 0
mirror of https://github.com/multica-ai/multica.git synced 2026-06-17 11:48:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

docs: add SWE-bench section to CLAUDE.md

Browse Source 
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Jiayuan Zhang
2026-02-15 18:32:04 +08:00
parent 10c57c0f7a
commit 45acb965ba
1 changed files with 20 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
CLAUDE.md
View File

@@ -190,6 +190,26 @@ Logged events: `run_start`, `run_end`, `llm_call`, `llm_result`, `tool_start`, `
Each line is a JSON object with `ts` (timestamp) and `event` (type), suitable for AI-assisted log analysis. Full event reference: `packages/core/src/agent/run-log.ts`.
## SWE-bench (Agent Benchmark)
Run the Multica agent against [SWE-bench](https://www.swebench.com/), the standard benchmark for evaluating AI coding agents on real GitHub issues.
```bash
# Download dataset
python scripts/swe-bench/download-dataset.py --dataset lite --limit 5
# Run agent against tasks
npx tsx scripts/swe-bench/run.ts --limit 5 --provider kimi-coding
# Analyze results
npx tsx scripts/swe-bench/analyze.ts
# Official evaluation (requires Docker)
bash scripts/swe-bench/evaluate.sh
```
Scripts are in `scripts/swe-bench/`. Full guide: `docs/swe-bench.md`.
## E2E Testing (Agent-Driven)
E2E tests are executed and analyzed by the Coding Agent (Claude Code), not by vitest. The Coding Agent runs the Multica agent via CLI, reads the structured run-log, and intelligently analyzes intermediate behavior and results.