Skip to content

v0.3.13 — Autonomous Model Selection

Choose a tag to compare

View all tags
@Keesan12 Keesan12 released this 26 Jun 16:46
v0.3.13
3025d36
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Pre-run governance enforcement + Windows nested-write fix + Autonomous model selection

Governance is now enforced at every code path

martin run blocks at the gate if no estimate receipt exists. Every code path — CLI, MCP, auto-bootstrap, JSON mode — requires an estimate before execution. No more silent bypasses.

  • martin estimate "..." writes a timestamped receipt
  • martin gate blocks Bash/Edit/Write until estimate + doctor are recorded
  • martin run fails with actionable message if estimate was skipped
  • Auto-bootstrap (JSON mode) automatically records an estimate receipt

Windows nested-write fix

File writes inside VS Code terminals and agent panels on Windows were silently blocked. The npm CLI shim created an extra cmd.exe process hop that prevented OS-level workspace-write permissions from propagating. Fixed by resolving the shim to its real node <script>.js target directly. Falls back safely on any format mismatch.

New failure class: sandbox_write_blocked — surfaces clearly instead of falling through to generic no_progress.

Autonomous model selection

martin run picks haiku/sonnet/opus automatically:

  • Simple task → claude-haiku, gpt-4o-mini, gemini-2.5-flash
  • Architectural work → claude-sonnet-4-6, gpt-4.1
  • Security + migration → claude-opus-4-6, o3

Works across Claude, Codex, Gemini, and all OpenAI-compatible endpoints.

Install

npm install -g martin-loop@0.3.13

Then wire governance into your IDE:

martin mcp install --host claude   # Claude Code
martin mcp install --host codex    # Codex
martin mcp install --host gemini   # Gemini CLI
martin mcp install --host cursor   # Cursor
Assets 2
Loading