feat: initial bootstrap — structure, task_brief, blackboard, adapter bases, escalation, prompts

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

adapters/base/llm.py

@@ -0,0 +1,50 @@
+"""
+adapters/base/llm.py
+Abstract base class for all LLM adapters.
+"""
+from abc import ABC, abstractmethod
+
+
+class LLMAdapter(ABC):
+    """
+    Contract that every LLM provider adapter must fulfil.
+
+    Capability strings
+    ------------------
+    "reasoning-heavy"  — tasks requiring deep chain-of-thought (e.g. T1, T2)
+    "capable"          — general-purpose capable model (e.g. T3, T4)
+    "fast-cheap"       — high-volume, low-latency tasks (e.g. T5 quick checks)
+    """
+
+    @abstractmethod
+    def complete(self, prompt: str, capability: str, context: dict) -> str:
+        """
+        Send a prompt to the model and return the text response.
+
+        Parameters
+        ----------
+        prompt      : The full prompt string (system + user combined or just user).
+        capability  : One of "reasoning-heavy" | "capable" | "fast-cheap".
+        context     : Arbitrary key/value bag passed to the adapter (e.g. system
+                      prompt override, temperature, max_tokens).
+
+        Returns
+        -------
+        The model's text completion as a plain string.
+        """
+        ...
+
+    @abstractmethod
+    def resolve_model(self, capability: str) -> str:
+        """
+        Map a capability string to the concrete model identifier for this provider.
+
+        Parameters
+        ----------
+        capability : One of "reasoning-heavy" | "capable" | "fast-cheap".
+
+        Returns
+        -------
+        Model identifier string (e.g. "claude-opus-4-6", "gpt-4o").
+        """
+        ...