buildlog 0.4.0__py3-none-any.whl → 0.5.0__py3-none-any.whl

buildlog/mcp/server.py

@@ -6,9 +6,16 @@ from mcp.server.fastmcp import FastMCP
 
 from buildlog.mcp.tools import (
     buildlog_diff,
+    buildlog_end_session,
+    buildlog_experiment_report,
     buildlog_learn_from_review,
+    buildlog_log_mistake,
+    buildlog_log_reward,
     buildlog_promote,
     buildlog_reject,
+    buildlog_rewards,
+    buildlog_session_metrics,
+    buildlog_start_session,
     buildlog_status,
 )
 
@@ -20,6 +27,15 @@ mcp.tool()(buildlog_promote)
 mcp.tool()(buildlog_reject)
 mcp.tool()(buildlog_diff)
 mcp.tool()(buildlog_learn_from_review)
+mcp.tool()(buildlog_log_reward)
+mcp.tool()(buildlog_rewards)
+
+# Session tracking tools (experiment infrastructure)
+mcp.tool()(buildlog_start_session)
+mcp.tool()(buildlog_end_session)
+mcp.tool()(buildlog_log_mistake)
+mcp.tool()(buildlog_session_metrics)
+mcp.tool()(buildlog_experiment_report)
 
 
 def main() -> None:

buildlog/mcp/tools.py

@@ -9,7 +9,20 @@ from dataclasses import asdict
 from pathlib import Path
 from typing import Literal
 
-from buildlog.core import diff, learn_from_review, promote, reject, status
+from buildlog.core import (
+    diff,
+    end_session,
+    get_experiment_report,
+    get_rewards,
+    get_session_metrics,
+    learn_from_review,
+    log_mistake,
+    log_reward,
+    promote,
+    reject,
+    start_session,
+    status,
+)
 
 
 def _validate_skill_ids(skill_ids: list[str]) -> list[str]:
@@ -140,3 +153,255 @@ def buildlog_learn_from_review(
     """
     result = learn_from_review(Path(buildlog_dir), issues, source)
     return asdict(result)
+
+
+def buildlog_log_reward(
+    outcome: str,
+    rules_active: list[str] | None = None,
+    revision_distance: float | None = None,
+    error_class: str | None = None,
+    notes: str | None = None,
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """Log a reward signal for bandit learning.
+
+    Call this after agent work to provide feedback on the outcome.
+    This enables learning which rules are effective in which contexts.
+
+    Args:
+        outcome: Type of feedback:
+            - "accepted": Work was accepted as-is (reward=1.0)
+            - "revision": Work needed changes (reward=1-distance)
+            - "rejected": Work was rejected entirely (reward=0.0)
+        rules_active: List of rule IDs that were in context during the work
+        revision_distance: How much correction was needed (0-1, 0=minor tweak, 1=complete redo)
+        error_class: Category of error if applicable (e.g., "missing_test", "validation_boundary")
+        notes: Optional notes about the feedback
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with reward_id, reward_value, total_events
+
+    Example:
+        # Work was accepted
+        buildlog_log_reward(outcome="accepted", rules_active=["arch-123", "wf-456"])
+
+        # Work needed revision
+        buildlog_log_reward(
+            outcome="revision",
+            revision_distance=0.3,
+            error_class="missing_test",
+            notes="Forgot to test error path"
+        )
+
+        # Work was rejected
+        buildlog_log_reward(outcome="rejected", notes="Completely wrong approach")
+    """
+    # Validate outcome
+    if outcome not in ("accepted", "revision", "rejected"):
+        return {
+            "reward_id": "",
+            "reward_value": 0.0,
+            "total_events": 0,
+            "message": "",
+            "error": f"Invalid outcome: {outcome}. Must be 'accepted', 'revision', or 'rejected'",
+        }
+
+    result = log_reward(
+        Path(buildlog_dir),
+        outcome=outcome,  # type: ignore[arg-type]
+        rules_active=rules_active,
+        revision_distance=revision_distance,
+        error_class=error_class,
+        notes=notes,
+        source="mcp",
+    )
+    return asdict(result)
+
+
+def buildlog_rewards(
+    limit: int | None = None,
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """Get reward events with summary statistics.
+
+    Returns recent reward events and aggregate statistics useful for
+    understanding learning progress.
+
+    Args:
+        limit: Maximum number of events to return (most recent first)
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with:
+            - total_events: Total count of reward events
+            - accepted: Count of accepted outcomes
+            - revisions: Count of revision outcomes
+            - rejected: Count of rejected outcomes
+            - mean_reward: Average reward value
+            - events: List of recent events (limited)
+
+    Example:
+        buildlog_rewards(limit=10)  # Get 10 most recent events with stats
+    """
+    result = get_rewards(Path(buildlog_dir), limit)
+
+    # Convert events to dicts
+    return {
+        "total_events": result.total_events,
+        "accepted": result.accepted,
+        "revisions": result.revisions,
+        "rejected": result.rejected,
+        "mean_reward": result.mean_reward,
+        "events": [e.to_dict() for e in result.events],
+    }
+
+
+# -----------------------------------------------------------------------------
+# Session Tracking MCP Tools (Experiment Infrastructure)
+# -----------------------------------------------------------------------------
+
+
+def buildlog_start_session(
+    error_class: str | None = None,
+    notes: str | None = None,
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """Start a new experiment session.
+
+    Begins tracking for a learning experiment. Captures the current
+    set of active rules to measure learning over time.
+
+    Args:
+        error_class: Error class being targeted (e.g., "missing_test")
+        notes: Notes about this session
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with session_id, error_class, rules_count, message
+
+    Example:
+        buildlog_start_session(error_class="missing_test")
+    """
+    result = start_session(
+        Path(buildlog_dir),
+        error_class=error_class,
+        notes=notes,
+    )
+    return asdict(result)
+
+
+def buildlog_end_session(
+    entry_file: str | None = None,
+    notes: str | None = None,
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """End the current experiment session.
+
+    Finalizes the session and calculates metrics including:
+    - Total mistakes logged
+    - Repeated mistakes (from prior sessions)
+    - Rules added during session
+
+    Args:
+        entry_file: Corresponding buildlog entry file, if any
+        notes: Additional notes to append
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with session_id, duration_minutes, mistakes_logged,
+        repeated_mistakes, rules_at_start, rules_at_end, message
+
+    Example:
+        buildlog_end_session(entry_file="2026-01-21.md")
+    """
+    result = end_session(
+        Path(buildlog_dir),
+        entry_file=entry_file,
+        notes=notes,
+    )
+    return asdict(result)
+
+
+def buildlog_log_mistake(
+    error_class: str,
+    description: str,
+    corrected_by_rule: str | None = None,
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """Log a mistake during the current session.
+
+    Records the mistake and checks if it's a repeat of a prior mistake
+    (from earlier sessions). This enables measuring repeated-mistake rates.
+
+    Args:
+        error_class: Category of error (e.g., "missing_test")
+        description: Description of the mistake
+        corrected_by_rule: Rule ID that should have prevented this
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with mistake_id, session_id, was_repeat, similar_prior, message
+
+    Example:
+        buildlog_log_mistake(
+            error_class="missing_test",
+            description="Forgot to add unit tests for new helper function"
+        )
+    """
+    result = log_mistake(
+        Path(buildlog_dir),
+        error_class=error_class,
+        description=description,
+        corrected_by_rule=corrected_by_rule,
+    )
+    return asdict(result)
+
+
+def buildlog_session_metrics(
+    session_id: str | None = None,
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """Get metrics for a session or all sessions.
+
+    Returns mistake rates and rule changes for analysis.
+
+    Args:
+        session_id: Specific session ID, or None for aggregate metrics
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with session_id, total_mistakes, repeated_mistakes,
+        repeated_mistake_rate, rules_at_start, rules_at_end, rules_added
+
+    Example:
+        buildlog_session_metrics()  # Aggregate metrics
+        buildlog_session_metrics(session_id="session-20260121-140000")
+    """
+    result = get_session_metrics(
+        Path(buildlog_dir),
+        session_id=session_id,
+    )
+    return asdict(result)
+
+
+def buildlog_experiment_report(
+    buildlog_dir: str = "buildlog",
+) -> dict:
+    """Generate a comprehensive experiment report.
+
+    Returns summary statistics, per-session breakdown, and error class analysis.
+
+    Args:
+        buildlog_dir: Path to buildlog directory
+
+    Returns:
+        Dict with:
+            - summary: Overall statistics
+            - sessions: Per-session breakdown
+            - error_classes: Breakdown by error class
+
+    Example:
+        buildlog_experiment_report()
+    """
+    return get_experiment_report(Path(buildlog_dir))