fix(spawn): pi profile silently ignores --model from profile config

[ruttybob]

Bug

When creating a profile for pi with a --model value, the model is silently ignored during spawn. The resulting command has no --model flag, so pi always uses its default model regardless of profile configuration.

Steps to Reproduce

# Create a pi profile with a specific model
clawteam profile set pi-anthropic \
  --agent pi \
  --model "anthropic/claude-sonnet-4.6"

# Spawn an agent using the profile
clawteam spawn tmux --profile pi-anthropic --team test --task "Do something"

Expected command: pi --model anthropic/claude-sonnet-4.6 -p "..."
Actual command: pi -p "..." ← --model is missing

Root Cause

In clawteam/spawn/profiles.py, _model_flag() maps agent names to their CLI model flag:

def _model_flag(agent: str) -> str | None:
    if agent in {"claude", "claude-code", "codex", "codex-cli", "gemini", "kimi"}:
        return "--model"
    return None  # ← "pi" is not in the set

apply_profile() calls _model_flag() and only appends --model when the return value is non-None:

if profile.model and not _command_has_model_arg(resolved_command):
    model_flag = _model_flag(agent)
    if model_flag:  # ← False for pi
        resolved_command.extend([model_flag, profile.model])

The pi-coding-agent does support --model (confirmed via pi --help):

--model <pattern>    Model pattern or ID (supports "provider/id" and optional ":<thinking>")

The adapter was added in PR #109 but _model_flag() was overlooked.

Fix

Add "pi" to the _model_flag() agent set, and add explicit return None entries in _base_url_env_var() and _api_key_target_env() for clarity (pi resolves these via --provider and provider-specific env vars).

PR: #132
Branch: fix/pi-profile-model-flag

Test Coverage

3 new tests added:

• test_apply_profile_pi_model_flag_is_injected — verifies --model is appended
• test_apply_profile_pi_model_not_duplicated_when_already_present — verifies no double --model
• test_apply_profile_pi_no_base_url_or_api_key_env_injection — verifies no unintended env var injection

[samyohuang-bit]

Fixed in 77bae51. Added 'pi', 'nanobot', 'openclaw' to _model_flag() mapping so --model from profile config is correctly passed during spawn.

[samyohuang-bit]

Fixed in 77bae51. Added 'pi', 'nanobot', 'openclaw' to _model_flag() mapping so --model from profile config is correctly passed during spawn.

def _model_flag(agent: str) -> str | None:
    if agent in {"claude", "claude-code", "codex", "codex-cli", "gemini", "kimi", "pi", "nanobot", "openclaw"}:
        return "--model"
    return None

All profile tests pass (8/8).

[ruttybob]

Thanks @samyohuang-bit

So my pr should be closed?

[tjb-tech]

Resolved on main. pi profiles now inject --model correctly, with explicit regression coverage in tests/test_profiles.py.