Problem

Permission prompt detection in get_status() fails in two ways depending on the approach:

1. PR fix: prevent permission prompt pattern from matching stale prompts #69 (current main): Pattern [ \t]* between [y/n/t]: and idle prompt can't bridge newlines. Active permission prompts are never detected — get_status() returns IDLE instead of WAITING_USER_ANSWER. Messages delivered during active prompts corrupt the terminal.

2. PR fix: Handle CLI prompts with trailing text #61 (previous approach): Pattern \s* bridges newlines but also matches stale (already-answered) prompts. After the user answers y, the old [y/n/t]: text remains in the terminal buffer. get_status() keeps returning WAITING_USER_ANSWER even after the command completes.

Both approaches fail because they try to combine the permission pattern and idle pattern into a single regex, which can't distinguish active from stale prompts.

Solution

Decouple permission detection from idle detection using line-based counting:

1. Match [y/n/t]: independently (non-greedy, no idle pattern suffix)
2. Find the last match (handles re-rendered prompts)
3. Count lines containing the idle prompt after the last [y/n/t]:
4. ≤1 idle line = active prompt → WAITING_USER_ANSWER

5. 1 idle lines = stale prompt → fall through to other status checks

This works because:

• Active prompt: CLI renders [y/n/t]: then at most one idle prompt on the next line
• Stale prompt: user answered, agent produced output, multiple idle prompts follow
• \r redraws stay on the same line (no \n), so they don't inflate the count

Applied to both kiro_cli.py and q_cli.py.

Test Results

Unit Tests (24 tests — test_permission_prompt_detection.py)

* #69 stale tests pass but only because permission prompts are never detected (right result, wrong reason).

Integration Tests (7 tests — test_kiro_cli_integration.py, real kiro-cli)

* #69 P5/P6 passes for wrong reason (nothing ever matches).

Integration tests support CAO_TEST_WATCH=1 to open a Terminal.app window attached to the tmux session for live debugging, and auto-dump terminal output on failure.

Files Changed

• src/cli_agent_orchestrator/providers/kiro_cli.py — line-based counting in get_status()
• src/cli_agent_orchestrator/providers/q_cli.py — same fix
• test/providers/test_permission_prompt_detection.py — 24 unit tests with 6 fixture files
• test/providers/test_kiro_cli_integration.py — 7 integration tests against real kiro-cli
• test/providers/fixtures/ — 6 fixture files with real ANSI terminal captures