fix: count only non-comment lines in command-sequence detection

[Osamaali313]

What

EnhancedStreamingMessageParser._isCommandSequence decides whether a fenced code block is a shell command sequence (to execute) or a file (to write), by what fraction of lines look like commands. The numerator already excludes comment/empty lines, but the denominator used lines.length (every line):

const commandLikeLines = lines.filter(
  (line) => line.length > 0 && !line.startsWith('#') && (this._isSingleLineCommand(line) || this._isSimpleCommand(line)),
);
// If more than 70% of non-comment lines are commands, treat as command sequence
return commandLikeLines.length / lines.length > 0.7;

The comment says "70% of non-comment lines", but the math divides by all lines. So comment lines dilute the ratio. A normal command snippet with explanatory comments:

```bash
# Install dependencies
npm install
# Start the dev server
npm run dev
```

is 2/2 = 100% commands by the intended rule, but computes 2/4 = 50%, falls below the 0.7 threshold, and is misclassified as a file to write instead of shell commands to run — the exact "code written to files instead of executed" failure this enhanced parser was built to address.

Fix

Compute the ratio over non-comment, non-empty lines (matching the comment/intent), and guard the all-comment case against 0/0 = NaN:

const nonCommentLines = lines.filter((line) => line.length > 0 && !line.startsWith('#'));
if (nonCommentLines.length === 0) {
  return false;
}
const commandLikeLines = nonCommentLines.filter(
  (line) => this._isSingleLineCommand(line) || this._isSimpleCommand(line),
);
return commandLikeLines.length / nonCommentLines.length > 0.7;

Testing

Added a regression test to message-parser.spec.ts (a comment-interleaved bash block is detected as shell). It fails on main (the block isn't recognized → onActionOpen never fires for a shell action) and passes with the fix.

vitest run app/lib/runtime/message-parser.spec.ts
✓ 46 tests passed

pnpm typecheck and pnpm lint pass (pre-commit hook green).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.