@jaguilar87/gaia 5.0.0-rc.2

package/tools/context/context_provider.py

@@ -0,0 +1,721 @@
+#!/usr/bin/env python3
+"""
+Context Provider for Claude Agent System
+
+Generates structured context payloads for agents based on:
+1. Agent contracts (context-contracts.json + cloud overlays)
+2. Universal rules (universal-rules.json)
+3. Historical episodes (episodic memory)
+
+Usage:
+    python3 context_provider.py <agent_name> [user_task] [--context-file PATH]
+"""
+
+import json
+import argparse
+import sys
+from pathlib import Path
+from typing import Dict, List, Any, Optional
+
+# Ensure the package root (gaia-ops-dev/) is on sys.path so that
+# `tools.memory.scoring` and `tools.memory.search_store` resolve when
+# this file runs as a subprocess with cwd=workspace-root.
+# Pattern: same as hooks/pre_tool_use.py line 22.
+_PACKAGE_ROOT = str(Path(__file__).resolve().parent.parent.parent)
+if _PACKAGE_ROOT not in sys.path:
+    sys.path.insert(0, _PACKAGE_ROOT)
+
+try:
+    from ._paths import resolve_config_dir
+    from .surface_router import (
+        build_investigation_brief,
+        classify_surfaces,
+        load_surface_routing_config,
+    )
+except ImportError:
+    from _paths import resolve_config_dir
+    from surface_router import (
+        build_investigation_brief,
+        classify_surfaces,
+        load_surface_routing_config,
+    )
+
+# Default paths
+DEFAULT_CONTEXT_PATH = Path(".claude/project-context/project-context.json")
+
+
+# ============================================================================
+# CONTRACTS DIRECTORY RESOLUTION
+# ============================================================================
+
+def get_contracts_dir():
+    """Determines the correct contracts directory based on execution context."""
+    return resolve_config_dir()
+
+
+DEFAULT_CONTRACTS_DIR = get_contracts_dir()
+
+
+# ============================================================================
+# UNIVERSAL RULES SYSTEM
+# ============================================================================
+
+DEFAULT_RULES_FILE = "universal-rules.json"
+
+
+def load_universal_rules(agent_name: str, rules_file: Optional[Path] = None) -> Dict[str, Any]:
+    """Load universal rules and agent-specific rules from JSON file."""
+    if rules_file is None:
+        rules_file = get_contracts_dir() / DEFAULT_RULES_FILE
+
+    if not rules_file.is_file():
+        print(f"Warning: Rules file not found: {rules_file}", file=sys.stderr)
+        return {"universal": [], "agent_specific": []}
+
+    try:
+        with open(rules_file, 'r', encoding='utf-8') as f:
+            rules_data = json.load(f)
+
+        universal = [r["rule"] for r in rules_data.get("rules", {}).get("universal", [])]
+        # agent_specific values may be a flat list [{rule:...}] or a nested
+        # dict {"rules": [{rule:...}]} -- handle both formats.
+        agent_raw = rules_data.get("rules", {}).get("agent_specific", {}).get(agent_name, [])
+        if isinstance(agent_raw, dict):
+            agent_raw = agent_raw.get("rules", [])
+        agent_specific = [r["rule"] for r in agent_raw]
+
+        total_rules = len(universal) + len(agent_specific)
+        if total_rules > 0:
+            print(f"Loaded {len(universal)} universal rules, {len(agent_specific)} agent-specific", file=sys.stderr)
+
+        return {
+            "universal": universal,
+            "agent_specific": agent_specific
+        }
+    except Exception as e:
+        print(f"Warning: Could not load rules: {e}", file=sys.stderr)
+        return {"universal": [], "agent_specific": []}
+
+
+# ============================================================================
+# CLOUD PROVIDER DETECTION
+# ============================================================================
+
+def detect_cloud_provider(project_context: Dict[str, Any]) -> str:
+    """Detects the cloud provider from project-context.json.
+
+    Detection priority:
+      1. metadata.cloud_provider (explicit user/scanner setting)
+      2. infrastructure.cloud_providers[0].name (v2 scanner section)
+      3. metadata.project_id presence -> gcp
+      4. Fallback -> gcp
+    """
+    metadata = project_context.get("metadata", {})
+    if "cloud_provider" in metadata:
+        provider = metadata["cloud_provider"].lower()
+        if provider == "multi-cloud":
+            print("Multi-cloud detected, using GCP contracts as primary", file=sys.stderr)
+            return "gcp"
+        return provider
+
+    sections = project_context.get("sections", {})
+
+    # v2: read from infrastructure.cloud_providers
+    infra = sections.get("infrastructure", {})
+    if isinstance(infra, dict):
+        cloud_providers = infra.get("cloud_providers", [])
+        if isinstance(cloud_providers, list) and cloud_providers:
+            primary = cloud_providers[0]
+            if isinstance(primary, dict):
+                name = primary.get("name", "")
+                if name:
+                    provider = name.lower()
+                    if provider == "multi-cloud":
+                        return "gcp"
+                    return provider
+
+    if "project_id" in metadata:
+        return "gcp"
+
+    print("Could not detect cloud provider, defaulting to GCP", file=sys.stderr)
+    return "gcp"
+
+
+def load_provider_contracts(cloud_provider: str, contracts_dir: Path = DEFAULT_CONTRACTS_DIR) -> Dict[str, Any]:
+    """
+    Loads context contracts using the base+cloud merge strategy.
+
+    Strategy:
+    1. Load base contracts from context-contracts.json (cloud-agnostic)
+    2. Load cloud overrides from cloud/{provider}.json and merge (extend) read/write lists
+    3. If base contracts missing → error (contracts are the single source of truth)
+    """
+    base_file = contracts_dir / "context-contracts.json"
+    cloud_file = contracts_dir / "cloud" / f"{cloud_provider}.json"
+
+    # --- Step 1: Load base contracts ---
+    if not base_file.is_file():
+        print(f"Error: Contract file not found at {base_file}", file=sys.stderr)
+        sys.exit(1)
+
+    try:
+        with open(base_file, 'r', encoding='utf-8') as f:
+            base_contracts = json.load(f)
+        print(f"Loaded base contracts from {base_file}", file=sys.stderr)
+    except json.JSONDecodeError as e:
+        print(f"Error: Invalid JSON in {base_file}: {e}", file=sys.stderr)
+        sys.exit(1)
+
+    # --- Step 2: Merge cloud-specific overrides ---
+    if cloud_file.is_file():
+        try:
+            with open(cloud_file, 'r', encoding='utf-8') as f:
+                cloud_overrides = json.load(f)
+            print(f"Loaded {cloud_provider.upper()} cloud overrides from {cloud_file}", file=sys.stderr)
+
+            for agent_name, agent_overrides in cloud_overrides.get("agents", {}).items():
+                if agent_name in base_contracts.get("agents", {}):
+                    existing_read = base_contracts["agents"][agent_name].get("read", [])
+                    existing_write = base_contracts["agents"][agent_name].get("write", [])
+                    extra_read = [s for s in agent_overrides.get("read", []) if s not in existing_read]
+                    extra_write = [s for s in agent_overrides.get("write", []) if s not in existing_write]
+                    base_contracts["agents"][agent_name]["read"] = existing_read + extra_read
+                    base_contracts["agents"][agent_name]["write"] = existing_write + extra_write
+                else:
+                    base_contracts["agents"][agent_name] = agent_overrides
+
+        except json.JSONDecodeError as e:
+            print(f"Warning: Invalid JSON in {cloud_file}: {e} — skipping cloud overrides", file=sys.stderr)
+    else:
+        print(f"No cloud overrides found at {cloud_file}, using base contracts only", file=sys.stderr)
+
+    return {
+        "version": base_contracts.get("version", "unknown"),
+        "provider": cloud_provider,
+        "agents": base_contracts.get("agents", {})
+    }
+
+
+def load_project_context(context_path: Path) -> Dict[str, Any]:
+    """Loads the project context from the specified JSON file."""
+    if not context_path.is_file():
+        print(f"Error: Context file not found at {context_path}", file=sys.stderr)
+        sys.exit(1)
+    with open(context_path, 'r', encoding='utf-8') as f:
+        return json.load(f)
+
+
+# ============================================================================
+# CONTEXT EXTRACTION
+# ============================================================================
+
+def get_relevant_sections(
+    sections: Dict[str, Any],
+    contract_keys: List[str],
+    surface_routing: Optional[Dict[str, Any]] = None,
+    routing_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """Filter sections by surface relevance, with fallback to all readable sections.
+
+    Args:
+        sections: All available sections from project-context.json.
+        contract_keys: The agent's permitted read keys (from context-contracts).
+        surface_routing: The routing result from classify_surfaces().
+        routing_config: The full surface-routing.json config (has contract_sections per surface).
+
+    Returns:
+        Filtered dict of sections. Falls back to all readable sections when:
+        - No surface_routing or routing_config provided
+        - No active surfaces detected
+        - Surface has no contract_sections defined
+        - Intersection of surface sections and agent permissions is empty
+    """
+    all_readable = {k: sections[k] for k in contract_keys if k in sections}
+
+    if not surface_routing or not routing_config:
+        return all_readable
+
+    active_surfaces = surface_routing.get("active_surfaces", [])
+    if not active_surfaces:
+        return all_readable
+
+    surfaces_cfg = routing_config.get("surfaces", {})
+
+    # Collect relevant sections from all active surfaces
+    relevant: set = set()
+    for surface in active_surfaces:
+        surface_config = surfaces_cfg.get(surface, {})
+        surface_sections = surface_config.get("contract_sections", [])
+        relevant.update(surface_sections)
+
+    if not relevant:
+        # Surfaces have no contract_sections defined -- inject all (fallback)
+        return all_readable
+
+    # Filter: agent permissions AND surface relevance
+    filtered = {k: sections[k] for k in contract_keys if k in sections and k in relevant}
+
+    if not filtered:
+        # Nothing matched -- inject all (fallback)
+        return all_readable
+
+    omitted = set(all_readable.keys()) - set(filtered.keys())
+    if omitted:
+        print(
+            f"Surface gating: {len(filtered)} sections injected, "
+            f"{len(omitted)} omitted ({', '.join(sorted(omitted))})",
+            file=sys.stderr,
+        )
+    else:
+        print(
+            f"Surface gating: all {len(filtered)} readable sections match active surfaces",
+            file=sys.stderr,
+        )
+
+    return filtered
+
+
+def get_contract_context(
+    project_context: Dict[str, Any],
+    agent_name: str,
+    provider_contracts: Dict[str, Any],
+    surface_routing: Optional[Dict[str, Any]] = None,
+    routing_config: Optional[Dict[str, Any]] = None,
+) -> Dict[str, Any]:
+    """Extracts the contract-defined context sections for a given agent.
+
+    When surface_routing and routing_config are provided, sections are filtered
+    to only those relevant to the active surface(s).  Falls back to returning
+    all readable sections when routing is unavailable or yields an empty set.
+    """
+    agent_contract = provider_contracts.get("agents", {}).get(agent_name)
+    if not agent_contract:
+        print(f"ERROR: Invalid agent '{agent_name}'. Available: {list(provider_contracts.get('agents', {}).keys())}", file=sys.stderr)
+        sys.exit(1)
+
+    contract_keys = agent_contract.get("read", [])
+
+    sections = project_context.get("sections", {})
+    if not sections:
+        raise KeyError("project-context.json must contain a 'sections' object.")
+
+    return get_relevant_sections(
+        sections, contract_keys,
+        surface_routing=surface_routing,
+        routing_config=routing_config,
+    )
+
+
+def get_context_update_contract(
+    agent_name: str,
+    provider_contracts: Dict[str, Any]
+) -> Dict[str, Any]:
+    """Return the SSOT contract agents should use for CONTEXT_UPDATE decisions."""
+    agent_contract = provider_contracts.get("agents", {}).get(agent_name)
+    if not agent_contract:
+        print(f"ERROR: Invalid agent '{agent_name}'. Available: {list(provider_contracts.get('agents', {}).keys())}", file=sys.stderr)
+        sys.exit(1)
+
+    return {
+        "readable_sections": agent_contract.get("read", []),
+        "writable_sections": agent_contract.get("write", []),
+        "source": "config/context-contracts.json + config/cloud/{provider}.json",
+    }
+
+
+# ============================================================================
+# EPISODIC MEMORY
+# ============================================================================
+
+try:
+    from tools.memory.scoring import rank_episodes as _rank_episodes
+    _HAS_SCORING = True
+except ImportError:
+    try:
+        import importlib, sys as _sys
+        _scoring = importlib.import_module("tools.memory.scoring")
+        _rank_episodes = _scoring.rank_episodes
+        _HAS_SCORING = True
+    except ImportError:
+        _rank_episodes = None
+        _HAS_SCORING = False
+
+try:
+    from tools.memory.search_store import search as fts5_search
+except ImportError:
+    fts5_search = None
+
+
+def _estimate_tokens(text: str) -> int:
+    """Rough token estimate: 1 token ≈ 4 characters."""
+    return len(text) // 4
+
+
+def _build_memory_index_table(index_episodes: List[Dict[str, Any]]) -> str:
+    """Build a compact markdown table of all memory sources for Layer 1."""
+    from datetime import datetime, timezone
+    lines = ["## Memory Index", "", "| # | Title | Type | Score | Age |", "|----|-------|------|-------|-----|"]
+    for i, ep in enumerate(index_episodes, 1):
+        title = ep.get("title", "")[:40]
+        ep_type = ep.get("type", "unknown")
+        score = ep.get("relevance_score", ep.get("_score", 0.0))
+        # Calculate age from timestamp field
+        ts = ep.get("timestamp", "")
+        try:
+            if ts:
+                ep_time = datetime.fromisoformat(ts.replace("Z", "+00:00"))
+                age = (datetime.now(timezone.utc) - ep_time).days
+                age_str = f"{age}d"
+            else:
+                age_str = "?d"
+        except Exception:
+            age_str = "?d"
+        lines.append(f"| {i} | {title} | {ep_type} | {score:.2f} | {age_str} |")
+    return "\n".join(lines)
+
+
+def _fallback_keyword_score(episode: Dict[str, Any], user_task: str) -> float:
+    """Keyword-based relevance scoring fallback when scoring module is unavailable."""
+    task_lower = user_task.lower()
+    task_words = set(task_lower.split())
+    score = 0.0
+    for tag in episode.get("tags", []):
+        if tag.lower() in task_lower:
+            score += 0.4
+    title_words = set(episode.get("title", "").lower().split())
+    common_words = task_words & title_words
+    if common_words:
+        score += 0.3 * (len(common_words) / max(len(title_words), 1))
+    return score * episode.get("relevance_score", 0.5)
+
+
+def load_relevant_episodes(
+    user_task: str,
+    max_episodes: int = 2,
+    max_tokens: Optional[int] = None,
+) -> Dict[str, Any]:
+    """Load relevant historical episodes using 2-layer progressive disclosure.
+
+    Layer 1 (always): A compact markdown table of all scored memory sources
+    (~200 tokens), returned under the ``memory_index`` key.
+
+    Layer 2 (selective): Full content of top-N episodes ranked by
+    ``tools.memory.scoring.rank_episodes()``, loaded only within the
+    remaining token budget.
+
+    Parameters
+    ----------
+    user_task:
+        Free-text description of the user's current task.
+    max_episodes:
+        Legacy cap on the number of full episodes to include (Layer 2).
+    max_tokens:
+        Total token budget for the episodic memory block.  Reads from
+        ``GAIA_MEMORY_TOKEN_BUDGET`` env var when not supplied explicitly.
+        Defaults to 2000.
+    """
+    import os as _os
+
+    if max_tokens is None:
+        env_budget = _os.environ.get("GAIA_MEMORY_TOKEN_BUDGET")
+        if env_budget:
+            try:
+                max_tokens = int(env_budget)
+            except ValueError:
+                max_tokens = 2000
+        else:
+            max_tokens = 2000
+
+    try:
+        index_file = Path(".claude/project-context/episodic-memory/index.json")
+        if not index_file.exists():
+            return {}
+
+        with open(index_file) as f:
+            index = json.load(f)
+
+        all_index_episodes = index.get("episodes", [])
+        if not all_index_episodes:
+            return {}
+
+        # --- Layer 1: Memory Index -- compact markdown table (~200 tokens, always included) ---
+        layer1_text = _build_memory_index_table(all_index_episodes)
+        layer1_tokens = _estimate_tokens(layer1_text)
+        remaining_budget = max_tokens - layer1_tokens
+
+        # --- Score and rank episodes: hybrid FTS5 + keyword fallback ---
+        # Build a lookup map from episode id to index entry for fast access
+        ep_by_id = {ep["id"]: ep for ep in all_index_episodes if "id" in ep}
+
+        # Try FTS5 first; results are BM25-ranked (better quality)
+        fts5_ids: List[str] = []
+        if fts5_search is not None:
+            try:
+                fts5_results = fts5_search(user_task, max_results=max_episodes * 3)
+                fts5_ids = [r["episode_id"] for r in fts5_results if "episode_id" in r]
+                print(
+                    f"FTS5 search returned {len(fts5_ids)} candidates for retrieval",
+                    file=sys.stderr,
+                )
+            except Exception as _fts_err:
+                print(
+                    f"Warning: FTS5 search failed (non-fatal): {_fts_err}",
+                    file=sys.stderr,
+                )
+                fts5_ids = []
+
+        # Build ranked list: FTS5 hits first, then fill with keyword/scoring results
+        fts5_id_set = set(fts5_ids)
+
+        # Keyword/scoring baseline (used to fill gaps and as full fallback)
+        if _HAS_SCORING and _rank_episodes is not None:
+            keyword_ranked = _rank_episodes(all_index_episodes, user_task)
+        else:
+            keyword_ranked = sorted(
+                [dict(ep, _score=_fallback_keyword_score(ep, user_task)) for ep in all_index_episodes],
+                key=lambda x: x["_score"],
+                reverse=True,
+            )
+
+        # If FTS5 found candidates, prepend them (with decay scoring if available)
+        if fts5_ids:
+            fts5_episodes = [ep_by_id[eid] for eid in fts5_ids if eid in ep_by_id]
+            if _HAS_SCORING and _rank_episodes is not None:
+                fts5_episodes = _rank_episodes(fts5_episodes, user_task)
+            else:
+                # Assign a generous score so they sort above keyword results
+                fts5_episodes = [dict(ep, _score=max(ep.get("_score", 0.0), 0.5)) for ep in fts5_episodes]
+            # Fill remaining slots with keyword results not already in FTS5 set
+            keyword_fill = [ep for ep in keyword_ranked if ep.get("id") not in fts5_id_set]
+            ranked = fts5_episodes + keyword_fill
+        else:
+            # FTS5 not available or returned nothing — fall back entirely to keyword
+            ranked = keyword_ranked
+
+        # --- Layer 2: full content of top episodes within remaining budget ---
+        full_episodes = []
+        tokens_used = 0
+        for ep in ranked:
+            if len(full_episodes) >= max_episodes:
+                break
+            score = ep.get("_score", 0.0)
+            if score <= 0.05:
+                continue
+            if remaining_budget <= 0:
+                break
+
+            full_ep = load_full_episode(ep["id"], index_file.parent)
+            if not full_ep:
+                continue
+
+            episode_entry = {
+                "id": full_ep["id"],
+                "title": full_ep["title"],
+                "type": full_ep["type"],
+                "relevance": round(score, 4),
+                "lessons_learned": full_ep.get("lessons_learned", [])[:2],
+                "resolution": full_ep.get("resolution", "")[:200],
+            }
+            entry_text = json.dumps(episode_entry)
+            entry_tokens = _estimate_tokens(entry_text)
+
+            if tokens_used + entry_tokens > remaining_budget:
+                break
+
+            full_episodes.append(episode_entry)
+            tokens_used += entry_tokens
+
+        result: Dict[str, Any] = {
+            "memory_index": layer1_text,
+        }
+
+        if full_episodes:
+            result["episodes"] = full_episodes
+            result["summary"] = f"Found {len(full_episodes)} relevant historical episodes"
+            print(
+                f"Added {len(full_episodes)} historical episodes to context "
+                f"(budget={max_tokens}, used≈{layer1_tokens + tokens_used})",
+                file=sys.stderr,
+            )
+        else:
+            print(
+                f"Memory index built ({len(all_index_episodes)} entries, "
+                f"no full episodes within score/budget threshold)",
+                file=sys.stderr,
+            )
+
+        # --- Retrieval strengthening: update retrieval_count + last_retrieved ---
+        # Failure here must never block context injection.
+        try:
+            if full_episodes:
+                import tempfile as _tempfile
+                from datetime import datetime as _datetime
+
+                selected_ids = {ep["id"] for ep in full_episodes}
+                now_iso = _datetime.utcnow().strftime("%Y-%m-%dT%H:%M:%SZ")
+
+                updated = False
+                for entry in index.get("episodes", []):
+                    if entry.get("id") in selected_ids:
+                        entry["retrieval_count"] = entry.get("retrieval_count", 0) + 1
+                        entry["last_retrieved"] = now_iso
+                        updated = True
+
+                if updated:
+                    index_path = index_file.resolve()
+                    index_dir = index_path.parent
+                    fd, tmp_path = _tempfile.mkstemp(
+                        dir=str(index_dir), suffix=".tmp", prefix="index_"
+                    )
+                    try:
+                        with _os.fdopen(fd, "w", encoding="utf-8") as tf:
+                            json.dump(index, tf, indent=2)
+                        _os.rename(tmp_path, str(index_path))
+                        print(
+                            f"Retrieval strengthening: updated {len(selected_ids)} episode(s)",
+                            file=sys.stderr,
+                        )
+                    except Exception:
+                        try:
+                            _os.unlink(tmp_path)
+                        except Exception:
+                            pass
+                        raise
+        except Exception as _rs_err:
+            print(
+                f"Warning: retrieval_count update failed (non-fatal): {_rs_err}",
+                file=sys.stderr,
+            )
+
+        return result
+
+    except Exception as e:
+        print(f"Warning: Could not load episodic memory: {e}", file=sys.stderr)
+        return {}
+
+
+def load_full_episode(episode_id: str, memory_dir: Path) -> Optional[Dict[str, Any]]:
+    """Load full episode details from JSONL file."""
+    try:
+        episodes_file = memory_dir / "episodes.jsonl"
+        if episodes_file.exists():
+            with open(episodes_file) as f:
+                for line in f:
+                    try:
+                        episode = json.loads(line)
+                        if episode.get("id") == episode_id:
+                            return episode
+                    except Exception:
+                        continue
+    except Exception:
+        pass
+    return None
+
+
+# ============================================================================
+# MAIN FUNCTION
+# ============================================================================
+
+def main():
+    """Main function to generate and print the context payload."""
+    import os as _os
+
+    parser = argparse.ArgumentParser(
+        description="Generates a structured context payload for a Claude agent."
+    )
+    parser.add_argument("agent_name", help="The name of the agent being invoked.")
+    parser.add_argument("user_task", nargs="?", default="General inquiry",
+                        help="The user's task or query for the agent.")
+    parser.add_argument(
+        "--context-file",
+        type=Path,
+        default=DEFAULT_CONTEXT_PATH,
+        help=f"Path to the project-context.json file. Defaults to '{DEFAULT_CONTEXT_PATH}'"
+    )
+    parser.add_argument(
+        "--memory-token-budget",
+        type=int,
+        default=None,
+        help=(
+            "Token budget for episodic memory injection. "
+            "Overrides GAIA_MEMORY_TOKEN_BUDGET env var. Default: 2000."
+        ),
+    )
+
+    args = parser.parse_args()
+
+    # Resolve memory token budget: CLI arg > env var > default
+    memory_token_budget: Optional[int] = args.memory_token_budget
+    if memory_token_budget is None:
+        env_budget = _os.environ.get("GAIA_MEMORY_TOKEN_BUDGET")
+        if env_budget:
+            try:
+                memory_token_budget = int(env_budget)
+            except ValueError:
+                memory_token_budget = 2000
+        else:
+            memory_token_budget = 2000
+
+    # Load project context
+    project_context = load_project_context(args.context_file)
+
+    # Detect cloud provider and load contracts
+    cloud_provider = detect_cloud_provider(project_context)
+    provider_contracts = load_provider_contracts(cloud_provider)
+
+    # Compute surface routing BEFORE extracting sections so we can gate by surface
+    surface_routing_config = load_surface_routing_config()
+    surface_routing = classify_surfaces(
+        args.user_task,
+        current_agent=args.agent_name,
+        routing_config=surface_routing_config,
+    )
+
+    # Extract contracted sections (surface-gated when routing is available)
+    contract_context = get_contract_context(
+        project_context, args.agent_name, provider_contracts,
+        surface_routing=surface_routing,
+        routing_config=surface_routing_config,
+    )
+    context_update_contract = get_context_update_contract(args.agent_name, provider_contracts)
+
+    # Load historical episodes (2-layer progressive disclosure)
+    historical_context = load_relevant_episodes(args.user_task, max_tokens=memory_token_budget)
+
+    # Load universal rules
+    rules_context = load_universal_rules(args.agent_name)
+    investigation_brief = build_investigation_brief(
+        args.user_task,
+        args.agent_name,
+        contract_context,
+        routing_config=surface_routing_config,
+        routing=surface_routing,
+    )
+
+    # Build final payload
+    final_payload = {
+        "project_knowledge": contract_context,
+        "write_permissions": context_update_contract,
+        "rules": rules_context,
+        "surface_routing": surface_routing,
+        "investigation_brief": investigation_brief,
+        "metadata": {
+            "cloud_provider": cloud_provider,
+            "contract_version": provider_contracts.get("version", "unknown"),
+            "historical_episodes_count": len(historical_context.get("episodes", [])),
+            "rules_count": len(rules_context.get("universal", [])) + len(rules_context.get("agent_specific", [])),
+            "surface_routing_version": surface_routing_config.get("version", "unknown"),
+            "active_surfaces_count": len(surface_routing.get("active_surfaces", [])),
+            "surface_routing_confidence": surface_routing.get("confidence", 0.0),
+        }
+    }
+
+    # Add historical context if episodes found
+    if historical_context:
+        final_payload["historical_context"] = historical_context
+
+    print(json.dumps(final_payload, indent=2))
+
+
+if __name__ == "__main__":
+    main()