feat(roles): add claude_cli provider — run kernel via Claude Code CLI (#38)

Adds a third LLM provider alongside anthropic and openai. Set
`llm.provider: claude_cli` in evolution.yml to drive planner / evaluator
/ goal_evaluator / strategist through the local `claude` binary instead
of an API key.

Why: the bundled examples (oss_fix_demo) already use claude CLI for the
executor, but the planning roles still required an Anthropic API key.
With this, a Claude Pro subscription is enough to run the full kernel —
lowers the bar for evaluators / readers who try the README quickstart.

Implementation:
- Each role gets a `_call_claude_cli(prompt, model)` that shells out to
  `claude -p --model <model> --output-format json`, reads `result` from
  the JSON envelope. Planner / evaluator also read `total_cost_usd` and
  total tokens so the kernel-level cost guards keep working.
- All four roles route through the existing `provider` switch; default
  behavior unchanged (provider defaults stay anthropic / openai).
- No new dependency — uses stdlib subprocess + json.

Defaults: opt-in only. Existing configs continue to work bit-for-bit.

Targets a v1.2.0 minor release.

Author: Protocol-zero-0

roles/evaluator.py

@@ -41,6 +41,23 @@ def _call_llm(prompt: str, cfg: dict) -> tuple[str, int, float]:
         tokens = resp.usage.total_tokens
         return resp.choices[0].message.content, tokens, tokens * 3e-6
 
+    if provider == "claude_cli":
+        import subprocess
+        proc = subprocess.run(
+            ["claude", "-p", "--model", model, "--output-format", "json"],
+            input=prompt,
+            capture_output=True,
+            text=True,
+        )
+        if proc.returncode != 0:
+            raise RuntimeError(f"claude -p failed (exit {proc.returncode}): {proc.stderr[:200]}")
+        data = json.loads(proc.stdout)
+        text = data.get("result", "")
+        usage = data.get("usage", {})
+        tokens = usage.get("input_tokens", 0) + usage.get("output_tokens", 0)
+        cost = float(data.get("total_cost_usd", 0.0))
+        return text, tokens, cost
+
     raise ValueError(f"unknown llm.provider: {provider!r}")
 
 

roles/goal_evaluator.py

@@ -38,6 +38,21 @@ def _call_openai(prompt: str, model: str, api_key_env: str) -> str:
     return resp.choices[0].message.content
 
 
+def _call_claude_cli(prompt: str, model: str) -> str:
+    """Call Claude via the local Claude Code CLI (no API key required)."""
+    import subprocess
+    proc = subprocess.run(
+        ["claude", "-p", "--model", model, "--output-format", "json"],
+        input=prompt,
+        capture_output=True,
+        text=True,
+    )
+    if proc.returncode != 0:
+        raise RuntimeError(f"claude -p failed (exit {proc.returncode}): {proc.stderr[:200]}")
+    data = json.loads(proc.stdout)
+    return data.get("result", "")
+
+
 def main() -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument("--input", required=True)
@@ -81,6 +96,8 @@ def main() -> None:
         text = _call_anthropic(prompt, model, api_key_env)
     elif provider == "openai":
         text = _call_openai(prompt, model, api_key_env)
+    elif provider == "claude_cli":
+        text = _call_claude_cli(prompt, model)
     else:
         print(f"error: unknown llm.provider: {provider!r}", file=sys.stderr)
         sys.exit(1)

roles/planner.py

@@ -45,6 +45,29 @@ def _call_openai(prompt: str, model: str, api_key_env: str) -> tuple[str, int, f
     return resp.choices[0].message.content, tokens, cost
 
 
+def _call_claude_cli(prompt: str, model: str) -> tuple[str, int, float]:
+    """Call Claude via the local Claude Code CLI (no API key required).
+
+    Uses --output-format json to recover the response text plus real
+    token / cost numbers, so kernel-level cost guards keep working.
+    """
+    import subprocess
+    proc = subprocess.run(
+        ["claude", "-p", "--model", model, "--output-format", "json"],
+        input=prompt,
+        capture_output=True,
+        text=True,
+    )
+    if proc.returncode != 0:
+        raise RuntimeError(f"claude -p failed (exit {proc.returncode}): {proc.stderr[:200]}")
+    data = json.loads(proc.stdout)
+    text = data.get("result", "")
+    usage = data.get("usage", {})
+    tokens = usage.get("input_tokens", 0) + usage.get("output_tokens", 0)
+    cost = float(data.get("total_cost_usd", 0.0))
+    return text, tokens, cost
+
+
 def main() -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument("--input", required=True)
@@ -104,6 +127,8 @@ def main() -> None:
         text, tokens, cost = _call_anthropic(prompt, model, api_key_env)
     elif provider == "openai":
         text, tokens, cost = _call_openai(prompt, model, api_key_env)
+    elif provider == "claude_cli":
+        text, tokens, cost = _call_claude_cli(prompt, model)
     else:
         print(f"error: unknown llm.provider: {provider!r}", file=sys.stderr)
         sys.exit(1)

roles/strategist.py

@@ -38,6 +38,21 @@ def _call_openai(prompt: str, model: str, api_key_env: str) -> str:
     return resp.choices[0].message.content
 
 
+def _call_claude_cli(prompt: str, model: str) -> str:
+    """Call Claude via the local Claude Code CLI (no API key required)."""
+    import subprocess
+    proc = subprocess.run(
+        ["claude", "-p", "--model", model, "--output-format", "json"],
+        input=prompt,
+        capture_output=True,
+        text=True,
+    )
+    if proc.returncode != 0:
+        raise RuntimeError(f"claude -p failed (exit {proc.returncode}): {proc.stderr[:200]}")
+    data = json.loads(proc.stdout)
+    return data.get("result", "")
+
+
 def main() -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument("--input", required=True)
@@ -80,6 +95,8 @@ def main() -> None:
         text = _call_anthropic(prompt, model, api_key_env)
     elif provider == "openai":
         text = _call_openai(prompt, model, api_key_env)
+    elif provider == "claude_cli":
+        text = _call_claude_cli(prompt, model)
     else:
         print(f"error: unknown llm.provider: {provider!r}", file=sys.stderr)
         sys.exit(1)