traceforge-toolkit 0.1.0__py3-none-any.whl

traceforge/parsers/copilot.py

@@ -0,0 +1,314 @@
+"""CopilotPreParser — extracts structured events from Copilot CLI data.
+
+Copilot CLI stores data in two locations:
+1. session-store.db: turns table with user_message + assistant_response (markdown)
+2. process-*.log: raw API messages with structured tool_use/tool_result JSON
+
+This parser handles BOTH:
+- Markdown parsing (via MarkdownPreParser base): extracts tool calls and
+  structured blocks from rendered assistant_response text
+- Log line parsing (parse_log_line): extracts structured API events from
+  embedded JSON in process logs
+
+Output: Structured event dicts suitable for copilot_markdown.yaml mapping.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Iterator
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from typing import Any
+
+import tree_sitter as ts
+
+from traceforge.parsers.base import (
+    MD_LANGUAGE,
+    Block,
+    MarkdownPreParser,
+    node_text,
+    try_parse_json,
+)
+
+# ─── Tree-sitter queries for Copilot response structure ──────────────────────
+
+_BLOCK_QUERY = ts.Query(
+    MD_LANGUAGE,
+    """
+    (atx_heading (atx_h1_marker) (inline) @h1_text) @h1
+    (atx_heading (atx_h2_marker) (inline) @h2_text) @h2
+    (atx_heading (atx_h3_marker) (inline) @h3_text) @h3
+    (fenced_code_block
+      (info_string) @lang
+      (code_fence_content) @code_content) @fenced
+    (paragraph) @para
+    (pipe_table) @table
+    (block_quote) @blockquote
+    (thematic_break) @hr
+    (list) @list_block
+    """,
+)
+
+# Pattern indices
+_PAT_H1 = 0
+_PAT_H2 = 1
+_PAT_H3 = 2
+_PAT_FENCED = 3
+_PAT_PARA = 4
+_PAT_TABLE = 5
+_PAT_BQ = 6
+_PAT_HR = 7
+_PAT_LIST = 8
+
+# ─── Log line patterns ───────────────────────────────────────────────────────
+
+# (removed in #45) The process-log path — parse_log_line + api_*/telemetry/
+# session_event extraction — targeted an obsolete Anthropic streaming log format
+# the current Copilot CLI no longer emits. The canonical high-fidelity path is
+# copilot.yaml over ~/.copilot/session-state/<id>/events.jsonl. Only the SQLite
+# turns path (parse_turn) remains here as a thin fallback.
+
+
+# ─── Turn state ──────────────────────────────────────────────────────────────
+
+
+@dataclass
+class _TurnState:
+    """Tracks state within a single turn parse."""
+
+    session_id: str
+    turn_index: int | None = None
+    timestamp: str = ""
+    sequence: int = 0
+
+
+# ─── Main parser ─────────────────────────────────────────────────────────────
+
+
+class CopilotPreParser(MarkdownPreParser):
+    """Extracts structured events from Copilot CLI stored data.
+
+    Two parsing modes:
+    1. parse_turn(row_dict) — for SqliteSource rows from the turns table
+       (uses MarkdownPreParser infrastructure for AST extraction)
+    2. parse_log_line(line) — for FileWatchSource lines from process logs
+       (parses structured JSON from raw API messages)
+    """
+
+    def __init__(self) -> None:
+        super().__init__()
+        self._turn_state: _TurnState | None = None
+
+    # ─── Base class hooks (for markdown AST parsing) ─────────────────────
+
+    def _get_query(self) -> ts.Query:
+        return _BLOCK_QUERY
+
+    def _classify_match(
+        self, pattern_index: int, captures: dict[str, list[ts.Node]], source: bytes
+    ) -> list[Block]:
+        blocks: list[Block] = []
+
+        if pattern_index in (_PAT_H1, _PAT_H2, _PAT_H3):
+            level = {_PAT_H1: 1, _PAT_H2: 2, _PAT_H3: 3}[pattern_index]
+            cap_prefix = {_PAT_H1: "h1", _PAT_H2: "h2", _PAT_H3: "h3"}[pattern_index]
+            text_nodes = captures.get(f"{cap_prefix}_text", [])
+            heading_nodes = captures.get(cap_prefix, [])
+            if heading_nodes and text_nodes:
+                blocks.append(
+                    Block(
+                        role="heading",
+                        byte_pos=heading_nodes[0].start_byte,
+                        text=node_text(text_nodes[0], source).strip(),
+                        level=level,
+                    )
+                )
+        elif pattern_index == _PAT_FENCED:
+            fenced_nodes = captures.get("fenced", [])
+            lang_nodes = captures.get("lang", [])
+            code_nodes = captures.get("code_content", [])
+            if fenced_nodes and code_nodes:
+                lang = node_text(lang_nodes[0], source).strip() if lang_nodes else ""
+                blocks.append(
+                    Block(
+                        role="code",
+                        byte_pos=fenced_nodes[0].start_byte,
+                        text=node_text(code_nodes[0], source),
+                        lang=lang,
+                    )
+                )
+        elif pattern_index == _PAT_PARA:
+            para_nodes = captures.get("para", [])
+            for node in para_nodes:
+                if node.parent and node.parent.type in ("document", "section"):
+                    blocks.append(
+                        Block(
+                            role="text",
+                            byte_pos=node.start_byte,
+                            text=node_text(node, source).strip(),
+                            node=node,
+                        )
+                    )
+        elif pattern_index == _PAT_TABLE:
+            table_nodes = captures.get("table", [])
+            for node in table_nodes:
+                blocks.append(
+                    Block(
+                        role="table",
+                        byte_pos=node.start_byte,
+                        text=node_text(node, source).strip(),
+                        node=node,
+                    )
+                )
+        elif pattern_index == _PAT_BQ:
+            bq_nodes = captures.get("blockquote", [])
+            for node in bq_nodes:
+                blocks.append(
+                    Block(
+                        role="quote",
+                        byte_pos=node.start_byte,
+                        text=node_text(node, source).strip(),
+                        node=node,
+                    )
+                )
+        elif pattern_index == _PAT_HR:
+            hr_nodes = captures.get("hr", [])
+            for node in hr_nodes:
+                blocks.append(Block(role="hr", byte_pos=node.start_byte))
+        elif pattern_index == _PAT_LIST:
+            list_nodes = captures.get("list_block", [])
+            for node in list_nodes:
+                if node.parent and node.parent.type in ("document", "section"):
+                    blocks.append(
+                        Block(
+                            role="list",
+                            byte_pos=node.start_byte,
+                            text=node_text(node, source).strip(),
+                            node=node,
+                        )
+                    )
+
+        return blocks
+
+    def _process_blocks(self, blocks: list[Block], source: bytes) -> Iterator[dict[str, Any]]:
+        """Convert sorted blocks into event dicts."""
+        for block in blocks:
+            yield from self._emit_block_event(block)
+
+    def _emit_block_event(self, block: Block) -> Iterator[dict[str, Any]]:
+        """Convert a single block into event dicts."""
+        state = self._turn_state
+
+        if block.role == "code":
+            lang = block.lang.lower()
+            if lang in ("powershell", "bash", "sh", "python", "cmd"):
+                yield self._turn_event(
+                    state,
+                    "tool_call",
+                    {
+                        "tool_name": lang,
+                        "command": block.text.strip(),
+                    },
+                )
+            elif lang in ("json", "yaml", "toml"):
+                parsed = try_parse_json(block.text)
+                if parsed:
+                    yield self._turn_event(
+                        state,
+                        "structured_output",
+                        {
+                            "format": lang,
+                            "data": parsed,
+                        },
+                    )
+                else:
+                    yield self._turn_event(
+                        state,
+                        "code_block",
+                        {
+                            "language": lang,
+                            "content": block.text.strip(),
+                        },
+                    )
+            else:
+                yield self._turn_event(
+                    state,
+                    "code_block",
+                    {
+                        "language": lang or "unknown",
+                        "content": block.text.strip(),
+                    },
+                )
+        elif block.role == "text":
+            if block.text:
+                yield self._turn_event(state, "assistant_text", {"content": block.text})
+        elif block.role == "heading":
+            yield self._turn_event(
+                state,
+                "section_heading",
+                {
+                    "level": block.level,
+                    "title": block.text,
+                },
+            )
+        elif block.role == "table":
+            yield self._turn_event(state, "table_output", {"content": block.text})
+        elif block.role == "quote":
+            yield self._turn_event(state, "quoted_output", {"content": block.text})
+        elif block.role == "list":
+            if block.text:
+                yield self._turn_event(state, "assistant_text", {"content": block.text})
+
+    # ─── Mode 1: Parse a turn row from SQLite ────────────────────────────
+
+    def parse_turn(self, row: dict[str, Any]) -> Iterator[dict[str, Any]]:
+        """Parse a row from the Copilot turns table into event dicts.
+
+        Expected row keys: session_id, turn_index, user_message,
+        assistant_response, timestamp
+        """
+        session_id = row.get("session_id", "unknown")
+        turn_index = row.get("turn_index")
+        timestamp = row.get("timestamp", datetime.now(timezone.utc).isoformat())
+
+        self._turn_state = _TurnState(
+            session_id=session_id,
+            turn_index=turn_index,
+            timestamp=timestamp,
+        )
+
+        # Emit user message event
+        user_msg = row.get("user_message")
+        if user_msg:
+            yield self._turn_event(
+                self._turn_state,
+                "user_message",
+                {
+                    "content": user_msg,
+                    "turn_index": turn_index,
+                },
+            )
+
+        # Parse assistant response with tree-sitter (uses base class infra)
+        assistant_resp = row.get("assistant_response")
+        if assistant_resp:
+            source = assistant_resp.encode("utf-8")
+            tree = self._ts_parser.parse(source)
+            blocks = self._extract_blocks(tree, source)
+            yield from self._process_blocks(blocks, source)
+
+    # ─── Event construction ──────────────────────────────────────────────
+
+    @staticmethod
+    def _turn_event(
+        state: _TurnState | None, event_type: str, payload: dict[str, Any]
+    ) -> dict[str, Any]:
+        event: dict[str, Any] = {
+            "type": event_type,
+            "timestamp": state.timestamp if state else datetime.now(timezone.utc).isoformat(),
+            "session_id": state.session_id if state else "unknown",
+        }
+        if state and state.turn_index is not None:
+            event["turn_index"] = state.turn_index
+        event.update(payload)
+        return event

traceforge/phase/__init__.py

@@ -0,0 +1,50 @@
+"""The trained, session-aware phase classifier — production inference.
+
+The per-event workflow-stage (planning / implementation / verification /
+exploration) predictor. Unlike the deterministic majority-vote
+:class:`traceforge.tracking.PhaseTracker` (whose segmentation output is now just
+one *feature* among many), this is a supervised classifier trained on the
+labelling corpus and is the only phase producer in this path.
+
+Public API:
+
+* :class:`PhaseInferencer` — apply a trained bundle to live ``SessionEvent``s
+  and stamp ``metadata.phase``.
+* :func:`load` — load a persisted bundle (env var / packaged default).
+* :func:`featurize_session_events` — the portable, causal session featuriser
+  (shared with the research training pipeline).
+* :class:`SegmentationParams`, :class:`NeighborParams` — feature hyperparameters.
+
+Requires ``model2vec`` + ``scikit-learn`` (core dependencies), imported
+lazily so the base package stays quick to import.
+"""
+
+from __future__ import annotations
+
+from .event_rows import event_to_feature_row
+from .features import (
+    NeighborParams,
+    feature_set_blocks,
+    featurize_session_events,
+)
+from .inference import (
+    DEFAULT_FEATURE_SET,
+    PhaseModel,
+    load,
+    predict_session,
+)
+from .inferencer import PhaseInferencer
+from .segmentation import SegmentationParams
+
+__all__ = [
+    "DEFAULT_FEATURE_SET",
+    "NeighborParams",
+    "PhaseInferencer",
+    "PhaseModel",
+    "SegmentationParams",
+    "event_to_feature_row",
+    "feature_set_blocks",
+    "featurize_session_events",
+    "load",
+    "predict_session",
+]

traceforge/phase/data/potion-base-8M/README.md

@@ -0,0 +1,99 @@
+---
+library_name: model2vec
+license: mit
+model_name: potion-base-8M
+tags:
+- embeddings
+- static-embeddings
+- sentence-transformers
+---
+
+# potion-base-8M Model Card
+
+This [Model2Vec](https://github.com/MinishLab/model2vec) model is a distilled version of a Sentence Transformer. It uses static embeddings, allowing text embeddings to be computed orders of magnitude faster on both GPU and CPU. It is designed for applications where computational resources are limited or where real-time performance is critical. Model2Vec models are the smallest, fastest, and most performant static embedders available. The distilled models are up to 50 times smaller and 500 times faster than traditional Sentence Transformers.
+
+
+## Installation
+
+Install model2vec using pip:
+```
+pip install model2vec
+```
+
+## Usage
+
+### Using Model2Vec
+
+The [Model2Vec library](https://github.com/MinishLab/model2vec) is the fastest and most lightweight way to run Model2Vec models.
+
+Load this model using the `from_pretrained` method:
+```python
+from model2vec import StaticModel
+
+# Load a pretrained Model2Vec model
+model = StaticModel.from_pretrained("potion-base-8M")
+
+# Compute text embeddings
+embeddings = model.encode(["Example sentence"])
+```
+
+### Using Sentence Transformers
+
+You can also use the [Sentence Transformers library](https://github.com/UKPLab/sentence-transformers) to load and use the model:
+
+```python
+from sentence_transformers import SentenceTransformer
+
+# Load a pretrained Sentence Transformer model
+model = SentenceTransformer("potion-base-8M")
+
+# Compute text embeddings
+embeddings = model.encode(["Example sentence"])
+```
+
+### Distilling a Model2Vec model
+
+You can distill a Model2Vec model from a Sentence Transformer model using the `distill` method. First, install the `distill` extra with `pip install model2vec[distill]`. Then, run the following code:
+
+```python
+from model2vec.distill import distill
+
+# Distill a Sentence Transformer model, in this case the BAAI/bge-base-en-v1.5 model
+m2v_model = distill(model_name="BAAI/bge-base-en-v1.5", pca_dims=256)
+
+# Save the model
+m2v_model.save_pretrained("m2v_model")
+```
+
+## How it works
+
+Model2vec creates a small, fast, and powerful model that outperforms other static embedding models by a large margin on all tasks we could find, while being much faster to create than traditional static embedding models such as GloVe. Best of all, you don't need any data to distill a model using Model2Vec.
+
+It works by passing a vocabulary through a sentence transformer model, then reducing the dimensionality of the resulting embeddings using PCA, and finally weighting the embeddings using [SIF weighting](https://openreview.net/pdf?id=SyK00v5xx). During inference, we simply take the mean of all token embeddings occurring in a sentence.
+
+## Additional Resources
+
+- [Model2Vec Repo](https://github.com/MinishLab/model2vec)
+- [Model2Vec Base Models](https://huggingface.co/collections/minishlab/model2vec-base-models-66fd9dd9b7c3b3c0f25ca90e)
+- [Model2Vec Results](https://github.com/MinishLab/model2vec/tree/main/results)
+- [Model2Vec Docs](https://minish.ai/packages/model2vec/introduction)
+
+
+## Library Authors
+
+Model2Vec was developed by the [Minish Lab](https://github.com/MinishLab) team consisting of [Stephan Tulkens](https://github.com/stephantul) and [Thomas van Dongen](https://github.com/Pringled).
+
+## Citation
+
+Please cite the [Model2Vec repository](https://github.com/MinishLab/model2vec) if you use this model in your work.
+```
+@software{minishlab2024model2vec,
+  author       = {Stephan Tulkens and {van Dongen}, Thomas},
+  title        = {Model2Vec: Fast State-of-the-Art Static Embeddings},
+  year         = {2024},
+  publisher    = {Zenodo},
+  doi          = {10.5281/zenodo.17270888},
+  url          = {https://github.com/MinishLab/model2vec},
+  license      = {MIT}
+}
+```

traceforge/phase/data/potion-base-8M/config.json

@@ -0,0 +1,13 @@
+{
+    "model_type": "model2vec",
+    "architectures": [
+        "StaticModel"
+    ],
+    "tokenizer_name": "baai/bge-base-en-v1.5",
+    "apply_pca": 256,
+    "apply_zipf": true,
+    "hidden_dim": 256,
+    "seq_length": 1000000,
+    "normalize": true,
+    "embedding_dtype": "float32"
+}

traceforge/phase/data/potion-base-8M/modules.json

@@ -0,0 +1,14 @@
+[
+    {
+        "idx": 0,
+        "name": "0",
+        "path": ".",
+        "type": "sentence_transformers.models.StaticEmbedding"
+    },
+    {
+        "idx": 1,
+        "name": "1",
+        "path": "1_Normalize",
+        "type": "sentence_transformers.models.Normalize"
+    }
+]