Skip to content

NanoTeams 1.7.1 — Autovisor Presets: hands-off repo audits

Choose a tag to compare

View all tags
@jmstajim jmstajim released this 18 Jul 21:38
· 16 commits to main since this release
1.7.1
3b06867

✨ Highlights

Point NanoTeams' autonomous manager — the Autovisor — at your repo and it sweeps the whole thing for bugs, security holes, missing tests, or stale docs, then hands back a triaged findings report. New in 1.7.1: pick its mission from five ready-made presets instead of writing a prompt.

And whether or not you touch the Autovisor: this release fixes the tool-call failures that make small local models flaky — quoted numbers that failed as "Missing required argument," silent wrong-branch runs, and list_files output bloating your context window.

🆕 Added

Autovisor Presets — a hands-off code auditor, one preset away. Pick a mission and the Autovisor sweeps your repo and hands you a triaged findings report — no prompt-writing, no fixes. Every preset is find-and-record only: the manager works through your repo incrementally, spawns worker tasks that read and report, and the findings pile up in reports/<topic>-findings.md. The manager itself carries only management and read-only file/git tools — write_file, edit_file, delete_file and git-write are excluded by construction — and every preset instructs its workers to write nothing but the report. You triage every finding and decide each fix. Never turned the Autovisor on? These presets are the easiest way to start. They appear in both the first-enable Setup pane and Settings → Autovisor; pick one and tweak it, or write your own.

  • ✦ Bug Hunter — a severity-tagged list of defects (crashes, wrong behavior, edge cases) in reports/bug-findings.md.
  • ✦ Test Coverage Guardian — every untested behavior and weak suite, each with a suggested test name to write next.
  • ✦ Code Quality Janitor — duplication, dead code, and structural debt, reported module by module.
  • ✦ Docs Maintainer — cross-checks the docs against the code and reports drift, gaps, and stale guides.
  • ✦ Security Auditor — hardcoded secrets, injection (shell, SQL, path traversal), unsafe deserialization, and missing validation. Secrets are recorded by location only — their values are never copied into the report.

🛠 Improved

  • Sloppy tool arguments no longer break valid calls. If you've watched a local model fail a call for no obvious reason, this was often why. Local models routinely quote their numbers and booleans ("501", "false") or send a one-item list as a bare string. NanoTeams now coerces string-encoded integers, booleans, and string arrays at one shared seam. A valid call no longer fails with "Missing required argument" — and no longer silently runs the wrong branch (edit_file rewriting one match instead of all, search scanning the whole tree).

  • list_files is leaner and correct.

    • Leaner results: the redundant per-entry name and type fields are gone, leaving bare path strings — roughly 30% smaller on a subdirectory listing and 55% from the repo root, where path and name used to be identical. Real context-window headroom on a local model where every token counts.
    • Entry paths are now relative to the work-folder root, so a listed file feeds straight into read_file / edit_file.
    • depth is 0-indexed (depth: 0 = this folder's contents), the way models actually spell it.

    Fixes a case where a fresh folder's very first listing reported "empty."

  • Addressing the work folder as / now works. Models often address the work folder chroot-style, and list_files(path: "/") used to fail with a permission error and burn an LLM round trip. Any root-style path now resolves to the work-folder root.

  • Long model names no longer blow out the UI. Long LM Studio slugs had stretched the status bar edge-to-edge and pushed the Refresh button off every LLM-override row.

🐛 Fixed

  • Autovisor stuck-task detection — a clock mismatch let genuinely hung tasks slip under the "stuck" threshold, so a wedged task could sit there burning time and tokens while the manager never noticed. Now caught on schedule.
  • Update checks — the same clock bug delayed NanoTeams' periodic check for a newer release on GitHub. It now runs on its interval.
  • Reliability — corrected loop-detection timing and sandbox root-path handling, with expanded test coverage.

System Requirements

  • macOS 15.0 or later
  • LM Studio 0.4.0 or later
  • Apple Silicon recommended
  • macOS 26 or later required for private voice dictation

Install

  1. Download NanoTeams.app.zip and extract it.
  2. Drag NanoTeams.app into /Applications.
  3. If macOS blocks the app on first launch: System Settings → Privacy & Security → Open Anyway.
  4. Launch LM Studio, load a recommended model, then launch NanoTeams.

Build from Source

git clone https://github.com/jmstajim/NanoTeams.git
cd NanoTeams
xcodebuild -project NanoTeams.xcodeproj -scheme NanoTeams -configuration Release build

Links

Assets 3
Loading