v0.2.0 — WordPad workflows, cross-platform CI, agent handoff

What's new in v0.2.0

Windows GUI generation that survives restarts. Generated GUI commands now find the app by its executable and window class instead of a captured process ID or title, show the selectors they use, and refuse to act when more than one window matches.

WordPad document workflows. For native English WordPad, the generated harness now includes open, read, replace, append, bold, save-as and workflow. The workflow edits a private copy of the document, saves it through WordPad's real Save As dialog, reopens it, and checks the text, heading weight and file integrity. It refuses to overwrite files, refuses to act without confirmation, and rejects malformed RTF or ambiguous markers.

Optional coding-agent handoff. Auto-CLI can prepare evidence, logs and validation for Codex or Claude to build out a harness. In one real GUI-only handoff, the generated harness passed 18 tests. That's one run, not a benchmark.

CI and integrations. Hosted CI now runs 8 jobs: tests on Windows, Linux and macOS across Python 3.10 and 3.13, a package build-and-install check, and real REA, sqlite-utils and Datasette integrations.

Other changes

• Generated harnesses include the Apache-licensed CLI-Anything REPL helper, with attribution, and declare their GUI dependencies.
• Generated commands return structured JSON errors and can report target and runtime info without a database.
• Database file paths are resolved correctly, and failed probes keep their diagnostics.
• README, Windows, deep-workflow and CI guides have been added or updated, along with a corrected comparison to CLI-Anything.

Numbers

• Local suite: 278 passed, 4 skipped (up from 126 passed, 2 skipped in v0.1.0).
• The opt-in WordPad integration test passed twice on a local Windows desktop.
• Auto-CLI still has no new required runtime libraries.

Limits

• The WordPad test needs an interactive Windows desktop with English WordPad installed. Hosted CI doesn't run it.
• Other editors need their own document operations.
• Live GUI testing has covered Windows Notepad, Calculator and WordPad only.
• Auth, POST bodies and MCP still need to be done by hand.
• There are no benchmarks or success-rate claims.

Install

pip install "git+https://github.com/Northernator/Auto-CLI.git@v0.2.0"

Independent project, not affiliated with REA or CLI-Anything.

Full changelog: v0.1.0...v0.2.0

v0.1.0 — first public release

First public release of Auto-CLI: point it at software and get an agent-native CLI. It runs REA recon, classifies the automation surface, and generates a CLI-Anything-style harness with the evidence attached.

• Commands: doctor, probe, surface, scaffold, pipeline, verify, drive, complete, gui, auto, validate
• auto: recon → scaffold → verify → optional IPC drive → complete → optional GUI capture → validate (--repair adds one repair pass)
• complete: list/get/search/add/update/delete per readable DB table, plus one command per verified endpoint
• Windows GUI driving via UIA (optional uiautomation)
• Safety: reads use a read-only connection; writes require --yes and take a timestamped backup
• No required runtime dependencies; Python 3.10+

Known limits: Windows-first, tested only on Notepad and Calculator; Linux/macOS untested. MCP, auth, POST bodies and scripting surfaces need hand implementation. No agent benchmarks. Not on PyPI; install from GitHub. Not affiliated with REA or CLI-Anything.

Install: pip install "git+https://github.com/Northernator/Auto-CLI.git@v0.1.0"

Tests: 126 passed, 2 skipped, 0 failed (pytest tests/ -q, unit suite on Linux x64, Python 3.13.5, Node v22.23.4). Run with click and prompt-toolkit installed; adding them to the dev extras is in PR #2, not yet on main. The code in this release is unchanged.