Add support for AI agent environment variables to quiet test output

[robobun]

Summary

This PR adds support for AI agent environment variables to enable quiet test output in Bun's test runner. When an AI agent is detected through environment variables like CLAUDECODE=1, REPL_ID=1, or IS_CODE_AGENT=1, the test runner reduces verbosity by only showing failures while still maintaining accurate test summary counts.

Changes

• Uses existing Output.isAIAgent() function in src/output.zig which detects:
  • CLAUDECODE=1 - Claude Code
  • REPL_ID=1 - Replit
  • IS_CODE_AGENT=1 - Generic AI agent flag
  • (Future support for Gemini, Codex, and Cursor background agents)
• Modified test output functions in src/cli/test_command.zig:
  • writeTestStatusLine() - Skip non-failure status indicators in quiet mode
  • printTestLine() - Skip non-failure test line output (with full JUnit support)
• Added comprehensive test suite with snapshots in test/cli/test/claudecode-flag.test.ts

Behavior

When AI agent is detected:

• Only test failures are displayed with their full output
• Passing, skipped, and todo test indicators are hidden
• Summary statistics are still shown at the end
• JUnit reporter output remains unaffected
• Error messages are more concise (e.g., "0 test files matching..." instead of friendly hints)

When no AI agent is detected:

• All test output is shown normally (existing behavior)

Example Output

Normal mode (no AI agent):

test.js:
(pass) passing test [1.00ms]
(fail) failing test [2.00ms]
(skip) skipped test
(todo) todo test

 1 pass
 1 skip
 1 todo
 1 fail

Quiet mode (AI agent detected):

test.js:
(fail) failing test [2.00ms]

 1 pass
 1 skip
 1 todo
 1 fail

Test plan

• All existing tests pass
• New comprehensive test suite with snapshots
• Comparison tests between AI agent and normal modes
• Debug build compilation successful
• JUnit reporter functionality preserved
• Summary statistics accuracy maintained

This feature is useful for AI coding assistants where reduced output verbosity improves readability and context efficiency while still providing essential failure information.

🤖 Generated with Claude Code

[robobun]

^{Updated 8:31 PM PT - Jul 17th, 2025}

@Jarred-Sumner, your commit 25d749a is building: #20874

[RiskyMH]

this probably should also force disable ansi (if it somehow gets enabled, as its a waste of its tokens)

[RiskyMH]

it should just always run this code here then? not too good to duplicate the code

[mizulu]

1. This feature seem vaguely familiar
   https://github.com/oven-sh/bun/compare/claude/implement-agent-flag

there they use --agent
which I think is slightly better, as it is more generic(for the day you switch to another agent)

2. some test runners implements flags to control that even more generically
   which is what we should have for bun test runner

Jest : https://jestjs.io/docs/cli#--onlyfailures

--onlyFailures

Vitest

--silent   // to remove all outputs 
--silent passed-only  //  let errors passthrough 

--no-color
https://vitest.dev/guide/cli.html#color

I think we need those explicit flags so we have more control future wise
as this feature might not be just agent specific, and can be useful
for users as well, especially the --onlyFailures

after we have proper options for the test runner,
then feel free to add some AGENT env with some convenience defaults