Skip to content

NeamCode v0.3.5.7 — Terminal-Bench 2.0 Alignment

Choose a tag to compare

View all tags
@github-actions github-actions released this 12 Mar 07:14
v0.3.5.7
7309de0

NeamCode v0.3.5.7 — Terminal-Bench 2.0 Alignment

10-phase CI/benchmark harness improvement based on Terminal-Bench 2.0 analysis (4 runs, 80 tasks, best score 15%).

What's New

  • CI Mode Auto-Detection — --prompt flag or non-TTY stdin triggers autonomous execution mode
  • Agent Loop Capacity — maxTurns scaled 25→200 in CI mode for complex multi-step tasks
  • Bash Timeout Scaling — default 30s→120s, max 120s→600s for package installs/builds
  • Token Output Limit — maxTokens 4096→16384 in CI mode to prevent truncated code output
  • Agentic System Prompt — 10 CRITICAL EXECUTION RULES injected in autonomous mode
  • Completion Verification Loop — synthetic verification prompt before finishing in CI mode
  • Parallel Tool Execution — infrastructure ready (config field, sequential for safety)
  • Error Recovery Hints — 7 pattern-matched hints (EACCES, ENOENT, SIGTERM, ENOSPC, ECONNREFUSED)
  • Token Budget — session budget 100K→500K, config-driven maxTurns/maxTokens

New CLI Flags

--max-turns <n>          Maximum agent loop turns (default: 25, CI: 200)
--max-tokens <n>         Maximum tokens per LLM response (default: 4096, CI: 16384)
--bash-timeout <ms>      Default bash timeout in ms (default: 120000)
--bash-max-timeout <ms>  Maximum bash timeout in ms (default: 600000)
--no-addon               Alias for --no-native (disable native addon)

Projected Impact

Metric Before After
Terminal-Bench Score 15% 56% (projected)
Max Turns (CI) 25 200
Max Tokens (CI) 4,096 16,384
Bash Timeout 30s / 120s 120s / 600s
Token Budget 100K 500K

Build Stats

  • Bundle: dist/index.js 1.20 MB (+10 KB from v0.3.5.6)
  • Tests: 3,748 passed | 1 failed (pre-existing Gemini quota)
  • Files Changed: 14 files, +3,896 / -58 lines

Installation

# macOS (Apple Silicon)
curl -fsSL https://github.com/neam-lang/NeamCode/releases/download/v0.3.5.7/neamcode-v0.3.5.7-macos-arm64.tar.gz | tar xz
# macOS (Intel)
curl -fsSL https://github.com/neam-lang/NeamCode/releases/download/v0.3.5.7/neamcode-v0.3.5.7-macos-x64.tar.gz | tar xz
# Linux (x64)
curl -fsSL https://github.com/neam-lang/NeamCode/releases/download/v0.3.5.7/neamcode-v0.3.5.7-linux-x64.tar.gz | tar xz
# Linux (ARM64)
curl -fsSL https://github.com/neam-lang/NeamCode/releases/download/v0.3.5.7/neamcode-v0.3.5.7-linux-arm64.tar.gz | tar xz
# npm
npm install -g neamcode@0.3.5.7

Documentation

Assets 11
Loading