Cao Provider

CAO Provider Creator

Guide for creating a new CLI agent provider for CLI Agent Orchestrator. A "provider" is an adapter that lets CAO interact with a specific CLI-based AI agent through tmux.

What You're Building

A provider translates between CAO's unified interface and a specific CLI tool's terminal output. It needs to:

1. Launch the CLI tool in a tmux window with the right flags
2. Detect status by parsing terminal output (IDLE, PROCESSING, COMPLETED, ERROR, WAITING_USER_ANSWER)
3. Extract responses from the terminal buffer after the agent finishes
4. Clean up when the terminal is deleted

Before You Start

Gather this information about the target CLI:

• What command launches it? (e.g., claude, kiro-cli chat, codex)
• What does the idle prompt look like? (e.g., > , ❯ , ask a question)
• What does the processing state look like? (e.g., spinner characters, "Thinking...")
• How are responses formatted? (e.g., preceded by ⏺, inside a box, plain text)
• Does it support --dangerously-skip-permissions or similar flags?
• Does it have a REPL mode or is it single-shot?
• How does it handle MCP servers? (CLI flags, config file, agent JSON)
• Does it use alt-screen (full-screen TUI) or scrollback (inline output)? This fundamentally changes status detection logic — see lesson #16
• What's the exit command? (/exit, /quit, Ctrl+C)

Step-by-Step Implementation

Step 1: Add to ProviderType enum

File: src/cli_agent_orchestrator/models/provider.py

class ProviderType(str, Enum):
    # ... existing providers ...
    NEW_CLI = "new_cli"

The value string is used everywhere — in API requests, database, config. Use snake_case.

Step 2: Create the provider class

File: src/cli_agent_orchestrator/providers/new_cli.py

Read references/provider-template.md for the full annotated template. The key sections:

Regex patterns — Define at module level, not inside methods. You need patterns for:

• ANSI code stripping (reuse r"\x1b\[[0-9;]*m")
• Idle prompt detection (what the prompt looks like when waiting for input)
• Processing detection (spinners, "Thinking...", progress indicators)
• Response markers (how agent responses start — e.g., ⏺ for Claude Code)
• Permission/confirmation prompts (if the CLI asks Y/n questions)

Status detection priority — The order in get_status() matters. Read references/lessons-learnt.md for the critical "stale buffer" lesson. The recommended pattern:

1. Strip ANSI codes from terminal output
2. Check WAITING_USER_ANSWER first (permission prompts need immediate attention)
3. Check COMPLETED (response marker + idle prompt both present in recent lines)
4. Check IDLE (just idle prompt, no response marker)
5. Check PROCESSING (spinner/thinking indicator in recent lines only)
6. Default to ERROR

Message extraction — Find the last response boundary in the terminal output and extract everything between it and the next prompt. Always strip ANSI codes from the final extracted text.

Step 3: Register in ProviderManager

File: src/cli_agent_orchestrator/providers/manager.py

Add the import and elif branch:

from cli_agent_orchestrator.providers.new_cli import NewCliProvider

# In create_provider():
elif provider_type == ProviderType.NEW_CLI.value:
    provider = NewCliProvider(
        terminal_id, tmux_session, tmux_window, agent_profile, allowed_tools
    )

Step 4: Add to PROVIDERS_REQUIRING_WORKSPACE_ACCESS

File: src/cli_agent_orchestrator/cli/commands/launch.py

If the provider executes code or accesses the filesystem, add it:

PROVIDERS_REQUIRING_WORKSPACE_ACCESS = {
    # ... existing ...
    "new_cli",
}

Step 5: Tool restriction enforcement

There are three approaches depending on the CLI's capabilities. Read docs/tool-restrictions.md for full context.

Hard enforcement via CLI flags (e.g., Claude Code, Copilot CLI): Add the provider to TOOL_MAPPING in src/cli_agent_orchestrator/utils/tool_mapping.py to translate CAO vocabulary to native tool names.

Hard enforcement via agent JSON (e.g., Kiro CLI): The CLI reads allowedTools from the agent profile. No TOOL_MAPPING entry needed — CAO passes vocabulary directly.

Soft enforcement via system prompt (e.g., Kimi CLI, Codex): No native restriction mechanism. CAO prepends restriction instructions to the system prompt. No TOOL_MAPPING entry needed.

Only add a TOOL_MAPPING entry if the CLI has its own native tool names that differ from CAO's vocabulary.

Step 6: Handle startup prompts

Many CLIs show cascading prompts on first launch (workspace trust, permission bypass, terms acceptance). Handle these in initialize() or a dedicated _handle_startup_prompts() method using a polling loop — not a single check. See references/lessons-learnt.md #17 for the stabilization loop pattern. Also consider shell warm-up (#14) and TERM variable compatibility (#15).

Step 7: Write unit tests

File: test/providers/test_new_cli_unit.py

Read references/test-guide.md for the full test structure. Minimum coverage:

1. Initialization — successful start, shell timeout, CLI timeout, agent profiles
2. Status detection — IDLE, PROCESSING, COMPLETED, WAITING_USER_ANSWER, ERROR, empty output
3. Message extraction — successful extraction, edge cases, error handling
4. Regex patterns — verify each pattern matches expected terminal output
5. Edge cases — ANSI codes, Unicode, long outputs, multiple responses

Use unittest.mock.patch to mock tmux_client. Create fixture files in test/providers/fixtures/.

Step 8: Write e2e tests

Add test classes to existing e2e test files and a fixture in test/e2e/conftest.py. Read references/test-guide.md for the full list of e2e test classes to add.

Step 9: Validate with assign + handoff orchestration

This is the canonical multi-agent e2e test. It exercises assign (non-blocking), handoff (blocking), send_message (async inbox), and status detection under concurrent load. Use the examples/assign/ profiles:

cao install examples/assign/data_analyst.md
cao install examples/assign/report_generator.md
cao install examples/assign/analysis_supervisor.md
cao launch --agents analysis_supervisor --provider new_cli --auto-approve

Test flow: Supervisor assigns 3x data_analyst workers in parallel + handoff 1x report_generator (blocking) → analysts send_message results back to supervisor → supervisor combines template + results into final report.

If any step fails, investigate:

• Assign fails: Status detection not recognizing IDLE after analyst finishes, or per-directory lock conflict (see lesson #19)
• Handoff times out: COMPLETED not detected — check stale buffer (lesson #1) or alt-screen detection (lesson #16)
• send_message not delivered: Supervisor not reaching IDLE state, blocking message delivery — check startup prompt loop (lesson #17)
• Concurrent failures: Race conditions in shared config files (lesson #19) or TERM env issues (lesson #15)

See test/e2e/test_assign.py for the automated version. Reference: https://github.com/awslabs/cli-agent-orchestrator/tree/feature/kimi-cli/examples/assign

Step 10: Documentation

Create docs/new-cli.md with prerequisites, launch examples, agent profile format, known limitations, and troubleshooting. Update README.md provider table and CHANGELOG.md.

File Checklist

When your provider is complete, verify you've touched all these files:

• src/cli_agent_orchestrator/models/provider.py — ProviderType enum
• src/cli_agent_orchestrator/providers/new_cli.py — Provider class
• src/cli_agent_orchestrator/providers/manager.py — Import + elif branch
• src/cli_agent_orchestrator/cli/commands/launch.py — PROVIDERS_REQUIRING_WORKSPACE_ACCESS
• src/cli_agent_orchestrator/utils/tool_mapping.py — TOOL_MAPPING (only if CLI needs translation)
• test/providers/test_new_cli_unit.py — Unit tests
• test/providers/fixtures/new_cli_*.txt — Test fixtures
• test/e2e/conftest.py — require_new_cli fixture
• test/e2e/test_*.py — E2E test classes
• docs/new-cli.md — Provider documentation
• README.md — Provider table
• CHANGELOG.md — New provider entry