Problem

The agent has run_tests as an internal tool but users can't drive it directly. Test failures during agent work get surfaced as raw cargo test / pytest / npm test output, which is hard to parse in a transcript.

Proposed solution

/test [path] slash command:

1. Auto-detect the project's test runner from manifest (Cargo.toml → cargo, package.json → npm/pnpm/yarn, pyproject.toml → pytest, etc.).
2. Run the test command. Stream output live.
3. Parse failure output structurally:
   • File:line of the assertion
   • Expected vs actual values
   • Backtrace (folded by default, expandable)
4. Render failures as cards (like tool results) with file:line as a clickable opener (Ctrl+Click → open in $EDITOR / code -g <file>:<line>).
5. Summary line at the end: "12 passed, 1 failed, 0 ignored — 4.2s".

Companion: file:line clickability everywhere

Stack traces in tool errors, sandbox denials, etc. all get parsed for path:line patterns and rendered as openable links.

Acceptance criteria

• /test runs the right command per project type (Rust / Node / Python at minimum).
• Failure card renders with structured fields (file, line, message).
• At least one click target opens an editor (env-driven: $EDITOR, code -g, idea --line, etc.).
• Summary line on completion.
• Works inside a sub-agent invocation.

Related

• crates/tui/src/tools/run_tests.rs — existing internal tool.
• crates/tui/src/commands/ — where the new slash command lives.