@maximoai/maximo-syntax-cli 0.1.52 → 0.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +12 -0
  2. package/dist/cli.mjs +12910 -10266
  3. package/package.json +1 -1
package/README.md CHANGED
@@ -63,6 +63,18 @@ After installation, you can invoke the CLI using any of the following commands:
63
63
 
64
64
  All five commands are equivalent and natively supported.
65
65
 
66
+ ## What's new in 0.2.0
67
+
68
+ - **Code Mode**: the `Code` tool lets the agent write a short script that calls its tools as functions (`read`, `edit`, `grep`, `` $`cmd` ``, `browser.*`, any MCP tool), so a ten-step chore is one round trip and only the result enters the context. Every call inside still goes through your permission rules and hooks.
69
+ - **Built-in Browser**: real Chrome over DevTools, with no extra download. Pages arrive as compact outlines with element refs, and each click or keystroke returns only what changed. It is always loaded, so a browser task starts at once.
70
+ - **Lessons**: when the agent finds what works after a failure, it records a one-line lesson with `Learn`. The best lessons go into future sessions, so the same mistake isn't paid for twice. Lessons work in any language.
71
+ - **Lean Context**: in long conversations, stale tool output (old snapshots, files read again, repeated commands) is swapped for one-line placeholders, but only when that saves more than the prompt-cache miss costs.
72
+ - **Leaner prompt**: the same rules in about half the fixed tokens per request (14.2K → 7.5K).
73
+ - **Fewer failed tool calls**: Maximo models get each tool's real list of required parameters, placeholder values such as `null` for an unused option no longer cost a round trip, and a failed command now shows its error output instead of only its exit code.
74
+ - **Faster, cheaper Maximo Atlas**: Atlas reasons at `low` effort unless you pick more (`/effort high`, `xhigh` or `max`), and the prompt cache now carries across sessions.
75
+
76
+ Live on Maximo Atlas 1.5, 0.2.0 finished the same real tasks at 56% lower cost and in 32% less time than 0.1.52. Benchmarks and how to reproduce them are in [`bench/`](bench/README.md). Everything can be switched off: `MAXIMO_SYNTAX_CODE_MODE=0`, `MAXIMO_SYNTAX_BROWSER=0`, `MAXIMO_SYNTAX_LESSONS=0`, `MAXIMO_SYNTAX_LEAN_CONTEXT=0`, `MAXIMO_SYNTAX_LEAN_PROMPT=0`.
77
+
66
78
  ## Scrolling (app-managed by default)
67
79
 
68
80
  Interactive sessions use **fullscreen alt-screen scroll** by default (`displayMode: