context-forge-cli 0.4.12__tar.gz → 0.4.14__tar.gz

{context_forge_cli-0.4.12 → context_forge_cli-0.4.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: context-forge-cli
-Version: 0.4.12
+Version: 0.4.14
 Summary: Session manager and context bridge for agentic CLI tools
 Project-URL: Homepage, https://github.com/emmver/contextforge
 Project-URL: Repository, https://github.com/emmver/contextforge
@@ -172,6 +172,20 @@ The method is chosen automatically based on token count and whether a target ses
 | Antigravity | `~/.gemini/antigravity-cli/history.jsonl` | `-i` / `--conversation` |
 | altimate-code | `~/.local/share/altimate-code/opencode.db` | `run -s` / `import` |
 | Cursor | `~/.config/Cursor/User/globalStorage/state.vscdb` | `cursor-agent -p` / `--resume` |
+| Pi agent | `~/.pi/agent/sessions/` JSONL | `--system-prompt` / `--session` + `--append-system-prompt` |
+| datatzis harness | `<repo>/.datatzis/` JSON (per-repo) | launch/resume only: `datatzis chat --resume` (delivered interactively) |
+
+Notes:
+
+- **Sub-agent tracking** — sessions spawned by sub-agents are tracked and cross-linked: Pi
+  (`pi-subagents`) child transcripts get `subagent` + `parent:<session-id>` tags naming the
+  parent session; the datatzis harness's orchestrator tasks (one per user prompt) get
+  `subagent` + `workspace:<session-id>` tags.
+- **datatzis discovery** — per-repo storage: ContextForge scans repository roots for
+  `*/.datatzis/workspaces/` (up to two levels deep). Configure roots via the
+  `DATATZIS_SEARCH_ROOTS` environment variable (colon-separated; default `~/Documents/Github`).
+  The datatzis CLI is interactive-only, so inject commands open or resume the session rather
+  than passing context as a flag.
 
 ## Session Summaries
 

{context_forge_cli-0.4.12 → context_forge_cli-0.4.14}/PLAN.md

@@ -89,6 +89,8 @@ Agents and contributors: update this file as you work.
 - [ ] `open-claw` adapter (if/when CLI interface is documented)
 - [x] Antigravity adapter (`adapters/antigravity.py`)
 - [x] Cursor adapter (`adapters/cursor.py`)
+- [x] Pi agent adapter (`adapters/pi.py`) — including sub-agent session tracking (`pi-subagents` runs nested under the parent session stem)
+- [x] datatzis harness adapter (`adapters/datatzis.py`) — per-repo `.datatzis/` workspaces + orchestrator-task sub-agent tracking (consultation/dossier runs intentionally out of scope)
 - [ ] Session search: `cf ls --search <query>`
 - [ ] Session archiving: `cf archive <id>`
 

{context_forge_cli-0.4.12 → context_forge_cli-0.4.14}/README.md

@@ -143,6 +143,20 @@ The method is chosen automatically based on token count and whether a target ses
 | Antigravity | `~/.gemini/antigravity-cli/history.jsonl` | `-i` / `--conversation` |
 | altimate-code | `~/.local/share/altimate-code/opencode.db` | `run -s` / `import` |
 | Cursor | `~/.config/Cursor/User/globalStorage/state.vscdb` | `cursor-agent -p` / `--resume` |
+| Pi agent | `~/.pi/agent/sessions/` JSONL | `--system-prompt` / `--session` + `--append-system-prompt` |
+| datatzis harness | `<repo>/.datatzis/` JSON (per-repo) | launch/resume only: `datatzis chat --resume` (delivered interactively) |
+
+Notes:
+
+- **Sub-agent tracking** — sessions spawned by sub-agents are tracked and cross-linked: Pi
+  (`pi-subagents`) child transcripts get `subagent` + `parent:<session-id>` tags naming the
+  parent session; the datatzis harness's orchestrator tasks (one per user prompt) get
+  `subagent` + `workspace:<session-id>` tags.
+- **datatzis discovery** — per-repo storage: ContextForge scans repository roots for
+  `*/.datatzis/workspaces/` (up to two levels deep). Configure roots via the
+  `DATATZIS_SEARCH_ROOTS` environment variable (colon-separated; default `~/Documents/Github`).
+  The datatzis CLI is interactive-only, so inject commands open or resume the session rather
+  than passing context as a flag.
 
 ## Session Summaries
 

context_forge_cli-0.4.14/contextforge/adapters/datatzis.py

@@ -0,0 +1,353 @@
+"""Adapter for the datatzis harness (per-repo writing harness).
+
+The datatzis harness stores its data inside each repository under
+``<repo>/.datatzis/``:
+
+* ``workspaces/<32-hex>.json`` — chat sessions.  A single JSON object with an
+  ``id`` (32-hex), ``name``, ``transcript`` (list of ``{role, text}`` rows with
+  roles ``user`` / ``assistant`` / ``notice``), the ids of the orchestrator
+  tasks it spawned, ``run_id``, ``agent`` spec and an ``updated_at`` stamp.
+* ``tasks/<32-hex>.json`` — one orchestrator execution per user prompt.  These
+  are the harness's sub-agent units: each records the ``prompt``, ``status``
+  (``running`` / ``completed`` / ``failed``), epoch float ``started`` /
+  ``ended`` stamps, tool ``events`` (tool name, ``arguments_sha256`` — actual
+  arguments are intentionally not stored — and results) and an ``info`` block
+  with token usage.  Workspaces link to their tasks via ``data["tasks"]``.
+* ``consultations/<id>.json`` and ``runs/<ts-slug>/`` are tied to research
+  dossiers rather than chat sessions and are deliberately not tracked here.
+
+Because ``notice`` rows carry stage/system notices, they are mapped to
+``system`` messages.  Per-message timestamps do not exist in workspaces (only
+a session-level ``updated_at``); tasks carry epoch stamps and are used for the
+sub-agent messages.
+
+The harness CLI is interactive-only: ``datatzis chat --resume <id-or-name>``
+reopens a session.  It has no non-interactive system-prompt surface, so
+``build_inject_command`` returns either a resume command or a plain ``datatzis
+chat`` launch.
+"""
+from __future__ import annotations
+
+import json
+import os
+import shlex
+from datetime import datetime, timezone
+from pathlib import Path
+
+from contextforge.adapters.base import ToolAdapter
+from contextforge.models.session import Message, Session
+from contextforge.utils.tokens import count_tokens
+
+DEFAULT_SEARCH_ROOTS: tuple[Path, ...] = (Path.home() / "Documents" / "Github",)
+
+_TITLE_SNIPPET_LEN = 80
+
+
+def _search_roots() -> list[Path]:
+    """Repository roots scanned for ``*/.datatzis/workspaces`` folders."""
+    env = os.environ.get("DATATZIS_SEARCH_ROOTS", "")
+    if env.strip():
+        return [Path(p).expanduser() for p in env.split(os.pathsep) if p.strip()]
+    return list(DEFAULT_SEARCH_ROOTS)
+
+
+def _parse_iso(ts) -> datetime | None:
+    if not isinstance(ts, str):
+        return None
+    try:
+        return datetime.fromisoformat(ts.replace("Z", "+00:00"))
+    except (ValueError, AttributeError):
+        return None
+
+
+def _parse_epoch(ts) -> datetime | None:
+    if isinstance(ts, (int, float)) and ts > 0:
+        try:
+            return datetime.fromtimestamp(ts, tz=timezone.utc)
+        except (OverflowError, OSError, ValueError):
+            return None
+    return None
+
+
+def _snippet(text: str, limit: int = _TITLE_SNIPPET_LEN) -> str:
+    return " ".join(str(text).split())[:limit]
+
+
+def _json_text(value) -> str:
+    if isinstance(value, str):
+        return value
+    try:
+        return json.dumps(value, ensure_ascii=False)
+    except (TypeError, ValueError):
+        return ""
+
+
+def _load_json(path: Path) -> dict | None:
+    try:
+        data = json.loads(path.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError, UnicodeDecodeError):
+        return None
+    return data if isinstance(data, dict) else None
+
+
+def _compute_message_tokens(msg: Message) -> int:
+    total = count_tokens(msg.content)
+    for tc in msg.tool_calls:
+        total += count_tokens(tc.get("input", ""))
+    for tr in msg.tool_results:
+        total += count_tokens(tr.get("output", ""))
+    return total
+
+
+class DatatzisAdapter(ToolAdapter):
+    tool_name = "datatzis"
+    default_paths = list(DEFAULT_SEARCH_ROOTS)
+
+    def is_available(self) -> bool:
+        return any(root.exists() for root in _search_roots())
+
+    # ------------------------------------------------------------------
+    # Storage layout
+    # ------------------------------------------------------------------
+
+    def _iter_workspaces(self) -> list[tuple[Path, Path]]:
+        """Yield (workspace_file, repo_root) pairs for every known repo."""
+        results: list[tuple[Path, Path]] = []
+        seen: set[Path] = set()
+        for root in _search_roots():
+            if not root.is_dir():
+                continue
+            candidates: list[Path] = []
+            for depth in (1, 2):  # <root>/<repo> and nested installs
+                star = "/".join(["*"] * depth)
+                candidates.extend(root.glob(f"{star}/.datatzis/workspaces"))
+            candidates.append(root / ".datatzis" / "workspaces")
+            for ws_dir in sorted(set(candidates)):
+                if not ws_dir.is_dir():
+                    continue
+                repo_root = ws_dir.parent.parent
+                for ws_file in sorted(ws_dir.glob("*.json")):
+                    if ws_file in seen:
+                        continue
+                    seen.add(ws_file)
+                    results.append((ws_file, repo_root))
+        return results
+
+    def _task_file(self, repo_root: Path, task_id: str) -> Path:
+        return repo_root / ".datatzis" / "tasks" / f"{task_id}.json"
+
+    def _session_path(self, session_id: str) -> tuple[Path, Path] | None:
+        for ws_file, repo_root in self._iter_workspaces():
+            if ws_file.stem == session_id:
+                return ws_file, repo_root
+            data = _load_json(ws_file)
+            if data and data.get("id") == session_id:
+                return ws_file, repo_root
+        for ws_file, repo_root in self._iter_workspaces():
+            task_file = self._task_file(repo_root, session_id)
+            if task_file.exists():
+                return task_file, repo_root
+        # Fallback for ids found in the header when workspaces live outside
+        # the configured roots is not needed: files are named after ids.
+        return None
+
+    # ------------------------------------------------------------------
+    # Discovery
+    # ------------------------------------------------------------------
+
+    def discover_sessions(self) -> list[Session]:
+        """Discover workspace sessions and their orchestrator-task sub-agents."""
+        sessions: list[Session] = []
+        for ws_file, repo_root in self._iter_workspaces():
+            data = _load_json(ws_file)
+            if not data or not data.get("id"):
+                continue
+            sessions.append(self._workspace_session(ws_file, repo_root, data))
+            sessions.extend(self._task_sessions(repo_root, data))
+        return sessions
+
+    def _workspace_session(self, ws_file: Path, repo_root: Path, data: dict) -> Session:
+        session_id = str(data["id"])
+        tasks = data.get("tasks") or []
+        created: datetime | None = None
+        updated = _parse_iso(data.get("updated_at"))
+        if updated is None:
+            updated = datetime.fromtimestamp(ws_file.stat().st_mtime, tz=timezone.utc)
+        if tasks:
+            for task_id in tasks:
+                task = _load_json(self._task_file(repo_root, str(task_id)))
+                started = _parse_epoch(task.get("started")) if task else None
+                if started is not None:
+                    created = started if created is None else min(created, started)
+        if created is None:
+            created = updated
+
+        # Token count: the displayed transcript (not the agent restore-thread)
+        tokens = 0
+        for row in data.get("transcript") or []:
+            tokens += count_tokens(str(row.get("text", "")))
+        tokens += count_tokens("\n".join(str(u) for u in data.get("unresolved") or []))
+
+        name = str(data.get("name") or "")
+        title = name if name and name != "Untitled" else None
+        if title is None:
+            for row in data.get("transcript") or []:
+                if row.get("role") == "user" and row.get("text"):
+                    title = _snippet(row["text"])
+                    break
+
+        return Session(
+            id=session_id,
+            tool=self.tool_name,
+            title=title,
+            cwd=str(repo_root),
+            created_at=created,
+            updated_at=updated,
+            status="unknown",
+            token_count=tokens or None,
+            raw_path=str(ws_file),
+        )
+
+    def _task_sessions(self, repo_root: Path, workspace: dict) -> list[Session]:
+        """One Session per orchestrator task (sub-agent execution)."""
+        session: list[Session] = []
+        workspace_id = str(workspace.get("id"))
+        for task_id in workspace.get("tasks") or []:
+            task_file = self._task_file(repo_root, str(task_id))
+            data = _load_json(task_file)
+            if not data:
+                continue
+            created = _parse_epoch(data.get("started"))
+            updated = _parse_epoch(data.get("ended"))
+            events = data.get("events") or []
+            if updated is None and events:
+                updated = _parse_epoch(events[-1].get("ended")) or created
+            updated = updated or created
+            tokens = count_tokens(str(data.get("prompt", "")))
+            summary = None
+            for event in events:
+                tokens += count_tokens(event.get("tool", ""))
+                result = event.get("result")
+                if result is not None:
+                    tokens += count_tokens(_json_text(result))
+                if event.get("error"):
+                    tokens += count_tokens(str(event["error"]))
+                    summary = str(event["error"])
+            if data.get("error"):
+                summary = str(data["error"])
+
+            title = _snippet(str(data.get("prompt", ""))) or None
+            session.append(
+                Session(
+                    id=str(data.get("id") or task_id),
+                    tool=self.tool_name,
+                    title=title,
+                    cwd=str(repo_root),
+                    created_at=created,
+                    updated_at=updated,
+                    status=str(data.get("status") or "unknown"),
+                    token_count=tokens or None,
+                    raw_path=str(task_file),
+                    summary=summary,
+                    tags=["subagent", f"workspace:{workspace_id}"],
+                )
+            )
+        return session
+
+    # ------------------------------------------------------------------
+    # Message loading
+    # ------------------------------------------------------------------
+
+    def load_messages(self, session_id: str) -> list[Message]:
+        resolved = self._session_path(session_id)
+        if resolved is None:
+            return []
+        path, _repo_root = resolved
+        data = _load_json(path)
+        if data is None:
+            return []
+        if data.get("id") == session_id and (data.get("transcript") is not None or "tasks" in data):
+            return self._workspace_messages(data)  # main workspace session
+        if "events" in data or "prompt" in data:
+            return self._task_messages(data)  # orchestrator task session
+        return []
+
+    def _workspace_messages(self, data: dict) -> list[Message]:
+        messages: list[Message] = []
+        for row in data.get("transcript") or []:
+            text = str(row.get("text", "") or "")
+            if not text:
+                continue
+            role = row.get("role")
+            if role in ("user", "assistant"):
+                mapped = role
+            elif role == "notice":
+                mapped = "system"
+            else:
+                continue
+            msg = Message(role=mapped, content=text)
+            msg.token_count = count_tokens(text)
+            messages.append(msg)
+        return messages
+
+    def _task_messages(self, data: dict) -> list[Message]:
+        """User prompt + one aggregate assistant turn holding all tool events.
+
+        Task events record ``arguments_sha256`` only (the harness does not
+        persist raw arguments), so tool-call ``input`` is empty by design.
+        """
+        messages: list[Message] = []
+        prompt = str(data.get("prompt", "") or "")
+        if prompt:
+            msg = Message(role="user", content=prompt, timestamp=_parse_epoch(data.get("started")))
+            msg.token_count = count_tokens(prompt)
+            messages.append(msg)
+
+        tool_calls: list[dict] = []
+        tool_results: list[dict] = []
+        last_ended = _parse_epoch(data.get("ended"))
+        for event in data.get("events") or []:
+            tool_name = event.get("tool", "?")
+            tool_calls.append({"name": tool_name, "input": ""})
+            result = event.get("result")
+            if result is not None:
+                tool_results.append({"tool": tool_name, "output": _json_text(result)})
+            if event.get("error"):
+                tool_results.append({"tool": tool_name, "output": f"error: {event['error']}"})
+            ended = _parse_epoch(event.get("ended"))
+            if ended is not None:
+                last_ended = ended if last_ended is None else max(last_ended, ended)
+        if tool_calls:
+            msg = Message(
+                role="assistant",
+                content=f"{len(tool_calls)} tool calls ({data.get('status') or 'unknown'})",
+                timestamp=last_ended,
+                tool_calls=tool_calls,
+                tool_results=tool_results,
+            )
+            msg.token_count = _compute_message_tokens(msg)
+            messages.append(msg)
+        return messages
+
+    # ------------------------------------------------------------------
+    # Injection
+    # ------------------------------------------------------------------
+
+    def build_inject_command(
+        self,
+        context: str,
+        target_session_id: str | None = None,
+        cwd: str | None = None,
+        method: str = "system_prompt",
+    ) -> str:
+        # datatzis chat is interactive-only: there is no system-prompt or
+        # one-shot flag, so injection opens (or resumes) the session and the
+        # harness's own handoff/line-mode delivers the context.
+        if method == "resume" and target_session_id:
+            if cwd:
+                return f"cd {shlex.quote(cwd)} && datatzis chat --resume {shlex.quote(target_session_id)}"
+            return f"datatzis chat --resume {shlex.quote(target_session_id)}"
+        cmd = "datatzis chat"
+        if cwd:
+            cmd = f"cd {shlex.quote(cwd)} && {cmd}"
+        return cmd