@gotcos/glasses-server 6.55.0 → 6.56.0

package/work-task-runtime/task_write.py

@@ -0,0 +1,645 @@
+#!/usr/bin/env python3
+"""Canonical tasks.md writers. One file_lock, one _guarded_write (v19 §1b)."""
+
+from __future__ import annotations
+from runtime_io import safe_read_text
+
+import os
+import re
+import signal
+import time
+from contextlib import contextmanager
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Iterator, Optional
+from uuid import uuid4
+
+from task_work_metadata import WorkMetadataError, decode_source, write_source, reject_reserved, validate_meeting, phase_for, PHASES, LEGACY_STAGE, DEFAULT_PHASE
+from cos_atomic import atomic_write_text
+from task_checkout import LockStoreCorrupt, TaskCheckoutManager
+from task_rows import (
+    full_domains,
+    MARKER_ANYWHERE_RE,
+    SHORT_TO_FULL,
+    classify_heading,
+    domain_path,
+    parse_path,
+    split_source,
+    split_done_when,
+    STAGES,
+    strip_markers,
+)
+
+TASK_LOCK_TTL_MINUTES = 1
+TASK_LOCK_WRITE_BUDGET_S = 5
+_STALE_TMP_S = 5 * 60
+_CHECKOUT = TaskCheckoutManager()
+_first_lock = False
+_DAY_RE = re.compile(r"^today:(\d{4}-\d{2}-\d{2})$")
+_TODAY_HEADING = re.compile(r"^#{1,3}\s*TODAY\s+(\d{4}-\d{2}-\d{2})\s*$")
+SECTIONS = ("inbox", "today", "urgent", "this_week", "backlog")
+
+
+class LockLost(Exception):
+    """TTL remaining is below the write budget, or identity mismatch."""
+
+
+class TaskFileLocked(Exception):
+    """file:<domain> is held; fail fast, no wait."""
+
+
+class TaskRunning(Exception):
+    """Row shows [agent running]; refuse re-arm."""
+
+
+def _arm_sigterm() -> None:
+    signal.signal(signal.SIGTERM, lambda *_: os._exit(1))
+
+
+def _sweep_stale_tmps(directory: Path) -> None:
+    now = time.time()
+    for tmp in directory.glob("*.tmp.*"):
+        try:
+            if now - tmp.stat().st_mtime > _STALE_TMP_S:
+                tmp.unlink()
+        except OSError:
+            pass
+
+
+def _is_safe_domain(name: str) -> bool:
+    """Safe as a path component. A SAFETY check, not a naming policy."""
+    if not name or len(name) > 64 or name != name.strip():
+        return False
+    if name in (".", "..") or name.startswith("."):
+        return False
+    return not any(c in name for c in ("/", "\\", "\0")) and not any(ord(c) < 32 or ord(c) == 127 for c in name)
+
+
+def _full_domain(domain: str) -> str:
+    full = SHORT_TO_FULL.get(domain, domain)
+    # Writes must not require the domain to already exist: creating the first
+    # task in a new domain is exactly how its tasks.md comes into being, so
+    # gating on full_domains() here made a new domain uncreatable. Safety is
+    # still enforced, because this name becomes a path component.
+    if not _is_safe_domain(full):
+        raise ValueError(f"invalid domain: {domain}")
+    return full
+
+
+def domain_from_path(path: Path) -> str:
+    return _full_domain(path.parent.name)
+
+
+@contextmanager
+def file_lock(domain: str) -> Iterator[str]:
+    """Exclusive file:<domain> lock. Fail fast → TaskFileLocked. No nesting."""
+    global _first_lock
+    full = _full_domain(domain)
+    if not _first_lock:
+        _arm_sigterm()
+        _sweep_stale_tmps(domain_path(full).parent)
+        _first_lock = True
+    identity = f"python:{os.getpid()}:{uuid4()}"
+    result = _CHECKOUT.checkout(
+        f"file:{full}",
+        identity,
+        exclusive=True,
+        timeout_minutes=TASK_LOCK_TTL_MINUTES,
+    )
+    if result.conflict:
+        raise TaskFileLocked(result.message)
+    try:
+        yield result.user
+    finally:
+        _CHECKOUT.release(f"file:{full}", identity)
+
+
+def _guarded_write(domain: str, identity: str, path: Path, text: str) -> None:
+    entry = _CHECKOUT.is_locked(f"file:{_full_domain(domain)}")
+    if entry is None or entry.user != identity:
+        raise LockLost("lock identity mismatch")
+    expires = datetime.fromisoformat(entry.expires_at)
+    remaining = (expires - datetime.now(timezone.utc)).total_seconds()
+    if remaining < TASK_LOCK_WRITE_BUDGET_S:
+        raise LockLost(f"write budget exhausted ({remaining:.1f}s left)")
+    atomic_write_text(path, text, encoding="utf-8", durable=True)
+
+
+def locked_replace(path: Path, text: str) -> None:
+    """Run-boundary helper for legacy writers."""
+    domain = domain_from_path(path)
+    with file_lock(domain) as identity:
+        _guarded_write(domain, identity, path, text)
+
+
+def _parse_section(token: str) -> tuple[str, Optional[str]]:
+    today = _DAY_RE.match(token)
+    if today:
+        return "today", today.group(1)
+    if token in SECTIONS and token != "today":
+        return token, None
+    raise ValueError(f"invalid section: {token}")
+
+
+def _heading_depth(line: str) -> Optional[int]:
+    stripped = line.lstrip()
+    if not stripped.startswith("#"):
+        return None
+    return len(stripped) - len(stripped.lstrip("#"))
+
+
+def _section_end(lines: list[str], start: int, depth: int) -> int:
+    idx = start + 1
+    while idx < len(lines):
+        stripped = lines[idx].strip()
+        if stripped.startswith("---"):
+            return idx
+        other = _heading_depth(lines[idx])
+        if other is not None and other <= depth:
+            return idx
+        idx += 1
+    return len(lines)
+
+
+def _find_heading(lines: list[str], want: str, day: Optional[str] = None) -> Optional[int]:
+    for idx, line in enumerate(lines):
+        kind, heading_day = classify_heading(line.strip())
+        if kind != want:
+            continue
+        if want == "today" and day and heading_day != day:
+            continue
+        return idx
+    return None
+
+
+def _insert_heading(lines: list[str], idx: int, heading: str) -> int:
+    block = ["", heading, ""]
+    lines[idx:idx] = block
+    return idx + 1
+
+
+def _locked_ensure_section(
+    lines: list[str], token: str, *, purge_before: Optional[str] = None
+) -> tuple[list[str], int, bool]:
+    """Return (lines, heading_index, fell_to_inbox)."""
+    kind, day = _parse_section(token)
+    if purge_before:
+        lines = _purge_empty_today(lines, purge_before)
+
+    if kind == "inbox":
+        found = _find_heading(lines, "inbox")
+        if found is not None:
+            return lines, found, False
+        first_h2 = next(
+            (i for i, line in enumerate(lines) if line.startswith("## ")),
+            len(lines),
+        )
+        return lines, _insert_heading(lines, first_h2, "## INBOX"), False
+
+    if kind == "today":
+        assert day
+        found = _find_heading(lines, "today", day)
+        if found is not None:
+            return lines, found, False
+        lines, inbox_idx, _ = _locked_ensure_section(lines, "inbox")
+        search_from = inbox_idx + 1
+        insert_at = len(lines)
+        for idx in range(search_from, len(lines)):
+            stripped = lines[idx].strip()
+            if stripped.startswith("## ") or stripped.startswith("---"):
+                insert_at = idx
+                break
+        return lines, _insert_heading(lines, insert_at, f"## TODAY {day}"), False
+
+    found = _find_heading(lines, kind)
+    if found is not None:
+        return lines, found, False
+    lines, inbox_idx, _ = _locked_ensure_section(lines, "inbox")
+    return lines, inbox_idx, True
+
+
+def _purge_empty_today(lines: list[str], purge_before: str) -> list[str]:
+    keep: list[str] = []
+    idx = 0
+    while idx < len(lines):
+        match = _TODAY_HEADING.match(lines[idx].strip())
+        if match and match.group(1) < purge_before:
+            depth = _heading_depth(lines[idx]) or 2
+            end = _section_end(lines, idx, depth)
+            body = [line for line in lines[idx + 1:end] if line.strip() and not line.strip().startswith("---")]
+            has_task = any(line.startswith("- [") for line in lines[idx + 1:end])
+            if not has_task and not any(classify_heading(b.strip())[0] for b in body):
+                idx = end
+                continue
+        keep.append(lines[idx])
+        idx += 1
+    return keep
+
+
+def _top_insert(lines: list[str], heading_idx: int, task_line: str) -> None:
+    depth = _heading_depth(lines[heading_idx]) or 2
+    insert_at = heading_idx + 1
+    while insert_at < len(lines) and not lines[insert_at].strip():
+        insert_at += 1
+    end = _section_end(lines, heading_idx, depth)
+    insert_at = min(insert_at, end)
+    lines.insert(insert_at, task_line)
+
+
+def _rebuild_line(description: str, source: Optional[str], run_at, agent_state, agent_no, checked: bool, archived: bool, stage=None, done_when=None) -> str:
+    mark = "~" if archived else ("x" if checked else " ")
+    body = description
+    if source:
+        body = f"{body} — **Source:** {source}"
+    if done_when:
+        body = f"{body} — **Done when:** {done_when}"
+    parts = []
+    if run_at:
+        parts.append(f"[run {run_at}]")
+    if agent_state == "running":
+        parts.append("[agent running]")
+    elif agent_state == "done" and agent_no is not None:
+        parts.append(f"[agent #{agent_no} done]")
+    elif agent_state == "failed":
+        if agent_no is not None:
+            parts.append(f"[agent #{agent_no} failed]")
+        else:
+            parts.append("[agent failed]")
+    if stage:
+        parts.append(f"[stage {stage}]")
+    if parts:
+        body = f"{body} {' '.join(parts)}"
+    return f"- [{mark}] {body}"
+
+
+def _match_row(row, task_id: str, domain: str) -> bool:
+    if row.id == task_id or row.ref == task_id:
+        return True
+    short = {"sprocket_rocket": "sr", "hermit_crabs": "hc"}.get(domain, domain)
+    ordinal = row.ref.rsplit("-", 1)[-1]
+    return task_id == f"{short}-{ordinal}"
+
+
+def _rewrite_row(path: Path, domain: str, task_id: str, mutator):
+    rows = parse_path(path, domain)
+    target = next((row for row in rows if _match_row(row, task_id, domain)), None)
+    if target is None:
+        return False
+    lines = safe_read_text(path).splitlines()
+    idx = target.line_number - 1
+    raw = lines[idx]
+    prefix = raw[: raw.index("- [")] if "- [" in raw else ""
+    desc, source = split_source(raw.split("]", 1)[-1].lstrip())
+    desc, markers = strip_markers(desc)
+    if source:
+        source, src_m = strip_markers(source)
+        if src_m.get("run_at"):
+            markers["run_at"] = src_m["run_at"]
+        if src_m.get("agent_state"):
+            markers["agent_state"] = src_m["agent_state"]
+            markers["agent_no"] = src_m.get("agent_no")
+    desc, dw_desc = split_done_when(desc)
+    if source:
+        source, dw_src = split_done_when(source)
+    else:
+        dw_src = None
+    state = {
+        "description": desc,
+        "source": source,
+        "stage": markers.get("stage") or target.stage,
+        "done_when": dw_desc or dw_src or target.done_when,
+        "run_at": markers.get("run_at") or target.run_at,
+        "agent_state": markers.get("agent_state") or target.agent_state,
+        "agent_no": markers.get("agent_no") if markers.get("agent_no") is not None else target.agent_no,
+        "checked": target.is_checked,
+        "archived": target.archived,
+    }
+    mutator(state)
+    lines[idx] = prefix + _rebuild_line(**state)
+    return True, "\n".join(lines) + "\n"
+
+
+def capture(domain: str, text: str, *, section: str, run_at=None, purge_before=None) -> dict:
+    reject_marker_text(text, "task text")
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        return _locked_capture(
+            full, path, identity, text, section=section,
+            run_at=run_at, purge_before=purge_before,
+        )
+
+
+def _locked_capture(domain, path, identity, text, *, section, run_at, purge_before) -> dict:
+    lines = safe_read_text(path).splitlines() if path.exists() else []
+    lines, heading_idx, fell = _locked_ensure_section(
+        lines, section, purge_before=purge_before
+    )
+    stamp = datetime.now().strftime("%Y-%m-%d")
+    line = _rebuild_line(text, f"Manual entry {stamp}", run_at, None, None, False, False)
+    _top_insert(lines, heading_idx, line)
+    _guarded_write(domain, identity, path, "\n".join(lines) + "\n")
+    landed = "inbox" if fell else section
+    return {"ok": True, "fell_to_inbox": fell, "section": landed}
+
+
+def ensure_section(domain: str, token: str, *, purge_before=None) -> None:
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        lines = safe_read_text(path).splitlines() if path.exists() else []
+        lines, _, _ = _locked_ensure_section(lines, token, purge_before=purge_before)
+        _guarded_write(full, identity, path, "\n".join(lines) + "\n")
+
+
+def reject_marker_text(value: str, field: str) -> None:
+    """User text may never contain a marker. This is the only gate.
+
+    _rebuild_line appends the marker blob AFTER the description and the
+    **Done when:** segment, so any marker inside user text becomes part of the
+    line's trailing syntax and is claimed by MARKER_RE on the next read. The
+    typed words vanish and the row changes state silently.
+
+    Reproduced before this guard existed: set_done_when(id, "[run 2026-12-25
+    09:00]") cleared a live finish line AND armed a scheduled run, on the field
+    labelled "What does finished look like?". set_text(id, "... [stage
+    planning]") moved a row out of ACTIVE and dropped the words.
+
+    Both writers pass through here because it is the one layer every client
+    (phone, lens, COS Control, CLI) reaches.
+    """
+    reject_reserved(value)
+    if "**source:**" in value.lower():
+        raise WorkMetadataError("reserved_source_delimiter", "Source fields may only be written through source-link controls.")
+    if "**done when:**" in value.lower():
+        raise WorkMetadataError("reserved_done_when_delimiter", "Done when fields may only be written through finish-line controls.")
+    hit = MARKER_ANYWHERE_RE.search(value)
+    if hit:
+        raise ValueError(
+            f"{field} may not contain {hit.group(0)}: square-bracket run, agent "
+            "and stage markers are reserved for the task line itself"
+        )
+
+
+def set_run_at(domain: str, task_id: str, run_at: Optional[str]) -> bool:
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        rows = parse_path(path, full) if path.exists() else []
+        target = next((row for row in rows if _match_row(row, task_id, full)), None)
+        if target is None:
+            return False
+        if target.agent_state == "running":
+            raise TaskRunning(f"task {task_id} is already running")
+        result = _rewrite_row(path, full, task_id, lambda s: s.update(run_at=run_at))
+        if result is False:
+            return False
+        _, text = result
+        _guarded_write(full, identity, path, text)
+        return True
+
+
+def set_text(domain: str, task_id: str, text: str) -> bool:
+    """Rewrite a task's description, preserving its source block and markers.
+
+    Only the description changes. The `**Source:**` block, the `[run ...]`
+    schedule and any agent marker are rebuilt from the parsed state, so editing
+    the words of a task never silently drops its provenance or its schedule.
+    """
+    clean = " ".join(text.split()).strip()
+    if not clean:
+        raise ValueError("task text required")
+    if "\n" in text or "\r" in text:
+        raise ValueError("task text must be a single line")
+    reject_marker_text(clean, "task text")
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        rows = parse_path(path, full) if path.exists() else []
+        target = next((row for row in rows if _match_row(row, task_id, full)), None)
+        if target is None:
+            return False
+        # A running agent owns the line; rewriting it underneath would change the
+        # task the agent was dispatched against.
+        if target.agent_state == "running":
+            raise TaskRunning(f"task {task_id} is already running")
+        if target.work_metadata_error:
+            raise WorkMetadataError("work_metadata_invalid", target.work_metadata_error)
+        def rename(state):
+            _, metadata, error = decode_source(state["source"])
+            if error:
+                raise WorkMetadataError("work_metadata_invalid", error)
+            metadata.setdefault("workIdentity", target.work_identity or target.id)
+            state.update(description=clean, source=write_source(state["source"], metadata))
+        result = _rewrite_row(path, full, task_id, rename)
+        if result is False:
+            return False
+        _, rebuilt = result
+        _guarded_write(full, identity, path, rebuilt)
+        return True
+
+
+def set_marker(domain: str, task_id: str, marker: Optional[str]) -> bool:
+    def mutate(state):
+        if marker is None or marker == "--clear":
+            state["agent_state"] = None
+            state["agent_no"] = None
+            return
+        if marker == "running":
+            state["agent_state"] = "running"
+            state["agent_no"] = None
+            return
+        if marker == "failed":
+            state["agent_state"] = "failed"
+            state["agent_no"] = None
+            return
+        kind, _, number = marker.partition(":")
+        state["agent_state"] = kind
+        state["agent_no"] = int(number) if number else None
+
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        result = _rewrite_row(path, full, task_id, mutate)
+        if result is False:
+            return False
+        _, text = result
+        _guarded_write(full, identity, path, text)
+        return True
+
+
+def set_stage(domain: str, task_id: str, stage: Optional[str]) -> bool:
+    """Write the workflow stage. None or --clear returns the row to planning,
+    which is the unmarked default, so clearing removes the marker entirely."""
+    if stage in (None, "", "--clear", "planning"):
+        wanted = None
+    elif stage in STAGES:
+        wanted = stage
+    else:
+        raise ValueError(f"invalid stage: {stage}")
+
+    def mutate(state):
+        _, metadata, error = decode_source(state["source"])
+        if error:
+            raise WorkMetadataError("work_metadata_invalid", error)
+        if metadata:
+            metadata["workflowPhase"] = DEFAULT_PHASE[wanted or 'planning']
+            state["source"] = write_source(state["source"], metadata)
+        state["stage"] = wanted
+
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        result = _rewrite_row(path, full, task_id, mutate)
+        if result is False:
+            return False
+        _, text = result
+        _guarded_write(full, identity, path, text)
+        return True
+
+
+def set_done_when(domain: str, task_id: str, text: Optional[str]) -> bool:
+    """Set or clear the finish line. A run is refused until this is non-empty."""
+    wanted = (text or "").strip() or None
+    if wanted and ("\n" in wanted or "**" in wanted):
+        raise ValueError("done_when must be a single line without ** markup")
+    if wanted:
+        reject_marker_text(wanted, "done_when")
+
+    def mutate(state):
+        state["done_when"] = wanted
+
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        result = _rewrite_row(path, full, task_id, mutate)
+        if result is False:
+            return False
+        _, text2 = result
+        _guarded_write(full, identity, path, text2)
+        return True
+
+
+def move(domain: str, task_id: str, section: str, *, purge_before=None) -> bool:
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        rows = parse_path(path, full)
+        target = next((row for row in rows if _match_row(row, task_id, full)), None)
+        if target is None:
+            return False
+        lines = safe_read_text(path).splitlines()
+        idx = target.line_number - 1
+        task_line = lines.pop(idx)
+        lines, heading_idx, _ = _locked_ensure_section(
+            lines, section, purge_before=purge_before
+        )
+        _top_insert(lines, heading_idx, task_line)
+        _guarded_write(full, identity, path, "\n".join(lines) + "\n")
+        return True
+
+
+def check(domain: str, task_id: Optional[str] = None, *, text: Optional[str] = None, uncheck: bool = False) -> str:
+    """Returns 'ok' | 'task_not_found' | 'ambiguous_text'."""
+    full = _full_domain(domain)
+    path = domain_path(full)
+    rows = parse_path(path, full)
+    if task_id:
+        matches = [row for row in rows if _match_row(row, task_id, full)]
+    else:
+        needle = (text or "").lower()
+        matches = [
+            row for row in rows
+            if needle and (
+                needle[:50] in row.description.lower()
+                or row.description.lower()[:50] in needle
+            )
+        ]
+    if not matches:
+        return "task_not_found"
+    if len(matches) > 1 and text and not task_id:
+        return "ambiguous_text"
+    target = matches[0]
+    with file_lock(full) as identity:
+        result = _rewrite_row(
+            path, full, target.id,
+            lambda s: s.update(checked=not uncheck),
+        )
+        if result is False:
+            return "task_not_found"
+        _, body = result
+        _guarded_write(full, identity, path, body)
+        return "ok"
+
+
+# Re-export for tests / writers
+LockStoreCorrupt = LockStoreCorrupt
+
+
+def _work_mutation(domain: str, task_id: str, expected_text: str, expected_revision: str, change) -> bool:
+    if not isinstance(expected_text, str) or not expected_text or len(expected_text) > 20_000:
+        raise WorkMetadataError("expected_text_required", "Full current task text is required.")
+    if not isinstance(task_id, str) or not re.fullmatch(r"[a-f0-9]{12}", task_id):
+        raise WorkMetadataError("invalid_task_id", "An exact canonical task ID is required.")
+    if not isinstance(expected_revision, str) or not re.fullmatch(r"[a-f0-9]{64}", expected_revision):
+        raise WorkMetadataError("expected_revision_required", "The current Work revision is required.")
+    full = _full_domain(domain)
+    path = domain_path(full)
+    with file_lock(full) as identity:
+        matches = [r for r in parse_path(path, full) if r.id == task_id]
+        if not matches:
+            return False
+        if len(matches) != 1:
+            raise WorkMetadataError("ambiguous_task_id", "Task ID is not unique in this domain.")
+        target = matches[0]
+        if target.archived:
+            raise WorkMetadataError("task_archived", "Restore the archived task before changing its Work details.")
+        if target.work_revision != expected_revision:
+            raise WorkMetadataError("task_revision_changed", "Task file changed. Refresh before editing Work details.")
+        if target.description != expected_text:
+            raise WorkMetadataError("task_text_changed", "Task text changed. Refresh before editing its Work details.")
+        if target.work_metadata_error:
+            raise WorkMetadataError("work_metadata_invalid", target.work_metadata_error)
+        def mutate(state):
+            _, metadata, error = decode_source(state["source"])
+            if error:
+                raise WorkMetadataError("work_metadata_invalid", error)
+            metadata.setdefault("workIdentity", target.work_identity or target.id)
+            change(state, metadata)
+            state["source"] = write_source(state["source"], metadata)
+        result = _rewrite_row(path, full, task_id, mutate)
+        if result is False:
+            return False
+        _, contents = result
+        _guarded_write(full, identity, path, contents)
+        return True
+
+
+def set_work_stage(domain: str, task_id: str, phase: str, *, expected_text: str, expected_revision: str) -> bool:
+    if phase not in (*PHASES, 'complete'):
+        raise WorkMetadataError("invalid_work_stage", "Unknown Work phase.")
+    def change(state, metadata):
+        if phase == 'complete':
+            metadata['workflowPhase'] = phase_for(state['stage'], False, metadata)
+            state['checked'] = True
+        else:
+            metadata['workflowPhase'] = phase
+            state['stage'] = None if LEGACY_STAGE[phase] == 'planning' else LEGACY_STAGE[phase]
+            state['checked'] = False
+    return _work_mutation(domain, task_id, expected_text, expected_revision, change)
+
+
+def link_meeting(domain: str, task_id: str, *, expected_text: str, expected_revision: str, meeting=None, remove_record_id=None) -> bool:
+    if (meeting is None) == (remove_record_id is None):
+        raise WorkMetadataError("invalid_meeting_ref", "Provide one meeting reference or one record ID to unlink.")
+    if meeting is not None:
+        meeting = validate_meeting(meeting)
+    elif not isinstance(remove_record_id, str) or not remove_record_id or len(remove_record_id) > 512:
+        raise WorkMetadataError("invalid_meeting_ref", "Invalid record ID to unlink.")
+    def change(state, metadata):
+        key = meeting['recordId'] if meeting is not None else remove_record_id
+        refs = [ref for ref in metadata.get('meetingRefs', []) if ref['recordId'] != key]
+        if meeting is not None:
+            refs.append(meeting)
+        metadata['meetingRefs'] = refs
+    return _work_mutation(domain, task_id, expected_text, expected_revision, change)