Skip to content
Python

Harden team accounting and shared status interfaces

Rohit C Prasad592eb7a66932
diff --git a/coworker/server/app.py b/coworker/server/app.py
index 1859a83d..7765e0a2 100644
--- a/coworker/server/app.py
+++ b/coworker/server/app.py
@@ -985,6 +985,12 @@ def create_app(manager: SessionManager) -> FastAPI:
return _board(request, run)
+ @app.post("/v1/board/items/status")
+ def board_set_status(request: Request, body: dict):
+ return _board(request, lambda actor: manager.team_store.set_status(
+ str(body.get("space", "")), actor, body.get("id"), body.get("text")
+ )) # Display-only: deliberately no team tick.
+
@app.post("/v1/board/items/claim")
def board_claim_item(request: Request, body: dict):
body = body or {}
diff --git a/coworker/teams/cli.py b/coworker/teams/cli.py
index 98d73082..0294778e 100644
--- a/coworker/teams/cli.py
+++ b/coworker/teams/cli.py
@@ -78,6 +78,10 @@ def _parser() -> argparse.ArgumentParser:
p = cmd("claim", _cmd_claim, "claim an open, unassigned item for yourself")
p.add_argument("id", type=int)
+ p = cmd("set-status", _cmd_set_status, "set a short progress line on your assigned item (worker only)")
+ p.add_argument("id", type=int)
+ p.add_argument("text")
+
p = cmd("move", _cmd_move, "transition an item")
p.add_argument("id", type=int)
p.add_argument("to", choices=_STATES[1:] + ("open",))
@@ -303,6 +307,12 @@ def _cmd_create(args) -> int:
return 0
+def _cmd_set_status(args) -> int:
+ item = _dialect(args).set_status(_space(args), args.id, args.text)
+ print(json.dumps(item) if args.json else item["status"])
+ return 0
+
+
def _cmd_claim(args) -> int:
item = _dialect(args).claim(_space(args), args.id)
print(
diff --git a/coworker/teams/dialect.py b/coworker/teams/dialect.py
index df66ac6e..71fcbd4e 100644
--- a/coworker/teams/dialect.py
+++ b/coworker/teams/dialect.py
@@ -77,6 +77,7 @@ class BoardDialect(Protocol):
) -> dict[str, Any]: ...
def assign(self, space: str, item_id: int, assignee: str) -> dict[str, Any]: ...
def claim(self, space: str, item_id: int) -> dict[str, Any]: ...
+ def set_status(self, space: str, item_id: int, text: str) -> dict[str, Any]: ...
def link(self, space: str, src: int, kind: str, dst: int) -> dict[str, Any]: ...
def attach(
self,
@@ -202,6 +203,9 @@ class LocalDialect:
def claim(self, space: str, item_id: int) -> dict[str, Any]:
return self.store.claim(space, self.actor, item_id)
+ def set_status(self, space: str, item_id: int, text: str) -> dict[str, Any]:
+ return self.store.set_status(space, self.actor, item_id, text)
+
def link(self, space: str, src: int, kind: str, dst: int) -> dict[str, Any]:
return self.store.link(space, self.actor, src, kind, dst)
@@ -434,6 +438,9 @@ class RemoteDialect:
def claim(self, space: str, item_id: int) -> dict[str, Any]:
return self._post("/v1/board/items/claim", {"space": space, "id": item_id})
+ def set_status(self, space: str, item_id: int, text: str) -> dict[str, Any]:
+ return self._post("/v1/board/items/status", {"space": space, "id": item_id, "text": text})
+
def link(self, space: str, src: int, kind: str, dst: int) -> dict[str, Any]:
return self._post(
"/v1/board/link", {"space": space, "src": src, "kind": kind, "dst": dst}
diff --git a/coworker/teams/mcp_server.py b/coworker/teams/mcp_server.py
index 1f9466a8..82f3c380 100644
--- a/coworker/teams/mcp_server.py
+++ b/coworker/teams/mcp_server.py
@@ -132,6 +132,14 @@ def build(dialect, *, space: str):
so they are not re-delivered."""
return _safe(lambda: (dialect.consume(space, upto_seq), {"ok": True})[1])
+ if role == "worker":
+
+ @mcp.tool()
+ def board_set_status(item: int, text: str) -> Any:
+ """Set one display-only progress line (at most 80 characters) on an item
+ currently assigned to you. Does not change state or wake the lead."""
+ return _safe(dialect.set_status, space, item, text)
+
if role in ("lead", "user"):
@mcp.tool()
diff --git a/coworker/teams/summary.py b/coworker/teams/summary.py
index 908f4e3d..262fd0f4 100644
--- a/coworker/teams/summary.py
+++ b/coworker/teams/summary.py
@@ -1,4 +1,5 @@
"""Read-only Team View projections. Never a wake source or an agent transcript tool."""
+
from __future__ import annotations
import json
@@ -13,7 +14,11 @@ KINDS = ("input", "cache_read", "output", "cache_write")
def stamp(value) -> float:
try:
- number = float(value) if isinstance(value, (float, int)) else datetime.fromisoformat(str(value).replace("Z", "+00:00")).timestamp()
+ number = (
+ float(value)
+ if isinstance(value, (float, int))
+ else datetime.fromisoformat(str(value).replace("Z", "+00:00")).timestamp()
+ )
return number if math.isfinite(number) else 0.0
except (ValueError, TypeError, OverflowError):
return 0.0
@@ -33,6 +38,29 @@ def add_tokens(rows) -> dict:
return {key: sum(row.get(key, 0) for row in rows) for key in KINDS}
+def compact_usage_points(points, buckets=120):
+ """Bound the optional chart payload, preserving every recorded token by role."""
+ points = sorted(points, key=lambda p: p["ts"])
+ if len(points) <= buckets:
+ return points
+ start, end = points[0]["ts"], points[-1]["ts"]
+ width = max((end - start) / buckets, 0.001)
+ grouped = {}
+ for point in points:
+ bucket = min(buckets - 1, int((point["ts"] - start) / width))
+ key = (point["role"], bucket)
+ row = grouped.setdefault(
+ key,
+ {
+ "role": point["role"],
+ "ts": min(end, start + (bucket + 1) * width),
+ "tokens": 0,
+ },
+ )
+ row["tokens"] += point["tokens"]
+ return sorted(grouped.values(), key=lambda p: p["ts"])
+
+
def all_events(store, space):
"""Read every page, not just the first 500 events of a long-running team."""
cursor = 0
@@ -53,23 +81,40 @@ def waiting_items(events, inbox, items):
p = event.get("payload") or {}
item = current.get(event.get("item_id"))
prompt = inbox.get(p.get("prompt_id", ""))
- if (item and prompt and prompt.state == "pending"
- and item["state"] not in ("done", "canceled")
- and item["assignee"] == event["actor"]
- and prompt.session_id == p.get("session_id")):
- result[item["id"]] = {"prompt_id": prompt.id, "tool": p.get("tool", ""),
- "preview": p.get("preview", ""), "session_id": prompt.session_id}
+ if (
+ item
+ and prompt
+ and prompt.state == "pending"
+ and item["state"] not in ("done", "canceled")
+ and item["assignee"] == event["actor"]
+ and prompt.session_id == p.get("session_id")
+ ):
+ result[item["id"]] = {
+ "prompt_id": prompt.id,
+ "tool": p.get("tool", ""),
+ "preview": p.get("preview", ""),
+ "session_id": prompt.session_id,
+ }
return result
def breakdown(intervals, start, end):
"""Partition wall time; overlapping parallel calls count once, not N times."""
- clipped = [(max(start, a), min(end, b), kind) for a, b, kind in intervals if b > start and a < end and b > a]
- edges = sorted({start, end, *(a for a, _, _ in clipped), *(b for _, b, _ in clipped)})
+ clipped = [
+ (max(start, a), min(end, b), kind)
+ for a, b, kind in intervals
+ if b > start and a < end and b > a
+ ]
+ edges = sorted(
+ {start, end, *(a for a, _, _ in clipped), *(b for _, b, _ in clipped)}
+ )
out = dict(model_ms=0, tool_ms=0, waited_ms=0, queued_ms=0)
for a, b in zip(edges, edges[1:]):
active = {kind for x, y, kind in clipped if x < b and y > a}
- kind = next((k for k in ("waited_ms", "tool_ms", "model_ms") if k in active), "queued_ms")
+ kind = next(
+ (k for k in ("waited_ms", "tool_ms", "model_ms") if k in active),
+ "queued_ms",
+ )
out[kind] += max(0, (b - a) * 1000)
return {k: round(v) for k, v in out.items()}
@@ -101,88 +146,206 @@ def make_summary(manager, team, now):
actors = {team.lead_actor, *(w.actor for w in team.workers)}
# A project board can be shared. A team view includes its filings/assignments,
# not another lead's unrelated backlog (the board itself remains unchanged).
- items = [i for i in board["items"] if i["creator"] in actors or i["assignee"] in actors]
+ items = [
+ i for i in board["items"] if i["creator"] in actors or i["assignee"] in actors
+ ]
by_item = {i["id"]: [] for i in items}
for e in events:
if e.get("item_id") in by_item:
by_item[e["item_id"]].append(e)
waits = waiting_items(events, manager.inbox, items)
session_ids = {team.lead_session, *(w.session_id for w in team.workers)}
- prompts = [p for p in manager.inbox.list() if p.session_id in session_ids and p.kind != "notification"]
+ prompts = [
+ p
+ for p in manager.inbox.list()
+ if p.session_id in session_ids and p.kind != "notification"
+ ]
# The lead decision and the original worker approval are one human interruption.
- forwarded = {(p.data.get("arguments") or {}).get("call_id") for p in prompts
- if p.data.get("tool") == "decide_worker_call"}
+ forwarded = {
+ (p.data.get("arguments") or {}).get("call_id")
+ for p in prompts
+ if p.data.get("tool") == "decide_worker_call"
+ }
asks = []
for p in prompts:
if p.id in forwarded:
continue
- kind = ("staffing" if p.data.get("members") else "reviews" if p.kind == "plan"
- else "questions" if p.kind == "question" else "grants" if p.kind in ("connector", "directory", "tool")
- else "approvals")
- asks.append({"id": p.id, "kind": kind, "state": p.state,
- "waited_s": max(0, (stamp(p.resolved_at) if p.resolved_at else now) - stamp(p.created_at))})
- ask_groups = {k: {"answered": sum(a["kind"] == k and a["state"] == "resolved" for a in asks),
- "waiting": sum(a["kind"] == k and a["state"] == "pending" for a in asks)}
- for k in ("approvals", "questions", "reviews", "staffing", "grants")}
+ kind = (
+ "staffing"
+ if p.data.get("members")
+ else (
+ "reviews"
+ if p.kind == "plan"
+ else (
+ "questions"
+ if p.kind == "question"
+ else (
+ "grants"
+ if p.kind in ("connector", "directory", "tool")
+ else "approvals"
+ )
+ )
+ )
+ )
+ asks.append(
+ {
+ "id": p.id,
+ "kind": kind,
+ "state": p.state,
+ "waited_s": max(
+ 0,
+ (stamp(p.resolved_at) if p.resolved_at else now)
+ - stamp(p.created_at),
+ ),
+ }
+ )
+ ask_groups = {
+ k: {
+ "answered": sum(a["kind"] == k and a["state"] == "resolved" for a in asks),
+ "waiting": sum(a["kind"] == k and a["state"] == "pending" for a in asks),
+ }
+ for k in ("approvals", "questions", "reviews", "staffing", "grants")
+ }
answered = [a["waited_s"] for a in asks if a["state"] == "resolved"]
- members = [(team.lead_actor, "lead", team.lead_session)] + [(w.actor, w.persona, w.session_id) for w in team.workers]
+ members = [(team.lead_actor, "lead", team.lead_session)] + [
+ (w.actor, w.persona, w.session_id) for w in team.workers
+ ]
workers, usage_points, records = [], [], {}
for actor, role, sid in members:
record = manager.session_store.load(sid)
engine = manager._engines.get(sid)
- messages = list(engine.messages) if engine is not None else (record.messages if record else [])
+ messages = (
+ list(engine.messages)
+ if engine is not None
+ else (record.messages if record else [])
+ )
records[actor] = messages
- usage = [tokens(m["usage"]) for m in messages if m.get("role") == "assistant" and isinstance(m.get("usage"), dict)]
+ usage = [
+ tokens(m["usage"])
+ for m in messages
+ if m.get("role") == "assistant" and isinstance(m.get("usage"), dict)
+ ]
# A persisted session's usage may be available before its full transcript.
- total = add_tokens(usage or [tokens(u) for u in (record.usage if record else {}).values()])
+ total = add_tokens(
+ usage or [tokens(u) for u in (record.usage if record else {}).values()]
+ )
step = None
for m in messages:
if m.get("role") == "assistant" and isinstance(m.get("usage"), dict):
ts = stamp((m.get("timing") or {}).get("model_started") or m.get("ts"))
if ts:
- usage_points.append({"ts": ts, "role": role, "tokens": sum(tokens(m["usage"]).values())})
+ usage_points.append(
+ {
+ "ts": ts,
+ "role": role,
+ "tokens": sum(tokens(m["usage"]).values()),
+ }
+ )
for call in m.get("tool_calls") or []:
if (call.get("function") or {}).get("name") == "todo_write":
try:
a = json.loads(call["function"].get("arguments") or "{}")
todos = a.get("todos", a.get("items", []))
if isinstance(todos, list) and todos:
- step = {"done": sum(x.get("status") == "done" for x in todos if isinstance(x, dict)), "total": len(todos)}
+ step = {
+ "done": sum(
+ x.get("status") == "done"
+ for x in todos
+ if isinstance(x, dict)
+ ),
+ "total": len(todos),
+ }
except (TypeError, ValueError):
pass
- workers.append({"actor": actor, "role": role, "session_id": sid, "model": record.model if record else "",
- "workspace": record.workspace if record else "", "running": manager.is_running(sid),
- "tokens": total, "step": step, "items": [i["id"] for i in items if i["assignee"] == actor]})
+ workers.append(
+ {
+ "actor": actor,
+ "role": role,
+ "session_id": sid,
+ "model": record.model if record else "",
+ "workspace": record.workspace if record else "",
+ "running": manager.is_running(sid),
+ "tokens": total,
+ "step": step,
+ "items": [i["id"] for i in items if i["assignee"] == actor],
+ }
+ )
windows = {i["id"]: active_windows(by_item[i["id"]], now) for i in items}
worker_by_actor = {w["actor"]: w for w in workers}
all_item_times = []
for item in items:
story = by_item[item["id"]]
activity = [e for e in story if e["kind"] != ITEM_STATUS]
- item["updated"] = stamp(activity[-1]["ts"]) if activity else stamp(item["created_ts"])
+ item["updated"] = (
+ stamp(activity[-1]["ts"]) if activity else stamp(item["created_ts"])
+ )
item["last_event"] = activity[-1]["kind"] if activity else "item_created"
item["waiting"] = waits.get(item["id"])
blockers = [l for l in item.get("links", []) if l["kind"] == "blocked_by"]
- item["group"] = ("waiting" if item["waiting"] or (item["state"] == "blocked" and not blockers)
- else "review" if item["state"] == "review" else "working" if item["state"] == "in_progress"
- else "done" if item["state"] == "done" else "canceled" if item["state"] == "canceled" else "queued")
+ item["group"] = (
+ "waiting"
+ if item["waiting"] or (item["state"] == "blocked" and not blockers)
+ else (
+ "review"
+ if item["state"] == "review"
+ else (
+ "working"
+ if item["state"] == "in_progress"
+ else (
+ "done"
+ if item["state"] == "done"
+ else "canceled" if item["state"] == "canceled" else "queued"
+ )
+ )
+ )
+ )
worker = worker_by_actor.get(item["assignee"])
item["worker_session"] = worker["session_id"] if worker else None
- active_assigned = [i for i in items if i["assignee"] == item["assignee"] and i["state"] not in ("review", "done", "canceled")]
- item["step"] = worker["step"] if worker and len(active_assigned) == 1 and item in active_assigned else None
- item["pr"] = next((ref for ref in item["refs"] if ref.startswith(("https://", "http://")) and "/pull/" in ref), None)
+ active_assigned = [
+ i
+ for i in items
+ if i["assignee"] == item["assignee"]
+ and i["state"] not in ("review", "done", "canceled")
+ ]
+ item["step"] = (
+ worker["step"]
+ if worker and len(active_assigned) == 1 and item in active_assigned
+ else None
+ )
+ item["pr"] = next(
+ (
+ ref
+ for ref in item["refs"]
+ if ref.startswith(("https://", "http://")) and "/pull/" in ref
+ ),
+ None,
+ )
started = stamp(item["created_ts"])
- closed = next((stamp(e["ts"]) for e in reversed(story) if e["kind"] == "item_transitioned"
- and e["payload"].get("to") in ("done", "canceled")), now)
+ closed = next(
+ (
+ stamp(e["ts"])
+ for e in reversed(story)
+ if e["kind"] == "item_transitioned"
+ and e["payload"].get("to") in ("done", "canceled")
+ ),
+ now,
+ )
+ if item["state"] not in ("done", "canceled"):
+ closed = now
item["elapsed_s"] = max(0, closed - started)
intervals, attributed_usage = [], []
ambiguous = False
for a, b, actor in windows[item["id"]]:
+ overlaps = [
+ (x, y)
+ for other, ws in windows.items()
+ if other != item["id"]
+ for x, y, who in ws
+ if who == actor and x < b and y > a
+ ]
for m in records.get(actor, []):
ts = stamp((m.get("timing") or {}).get("model_started") or m.get("ts"))
# Multiple assignments: don't attribute a single call to two tasks.
- overlaps = [(x, y) for other, ws in windows.items() if other != item["id"]
- for x, y, who in ws if who == actor and x < b and y > a]
if any(x <= ts < y for x, y in overlaps):
ambiguous = True
continue
@@ -196,37 +359,96 @@ def make_summary(manager, team, now):
stop = start + duration / 1000
if stop <= a or start >= b:
continue
- cuts = sorted({max(a, start), min(b, stop), *(max(a, start, x) for x, y in overlaps if x < min(b, stop) and y > max(a, start)), *(min(b, stop, y) for x, y in overlaps if x < min(b, stop) and y > max(a, start))})
+ cuts = sorted(
+ {
+ max(a, start),
+ min(b, stop),
+ *(
+ max(a, start, x)
+ for x, y in overlaps
+ if x < min(b, stop) and y > max(a, start)
+ ),
+ *(
+ min(b, stop, y)
+ for x, y in overlaps
+ if x < min(b, stop) and y > max(a, start)
+ ),
+ }
+ )
for x, y in zip(cuts, cuts[1:]):
- if x < y and not any(ox < y and oy > x for ox, oy in overlaps):
+ if x < y and not any(
+ ox < y and oy > x for ox, oy in overlaps
+ ):
intervals.append((x, y, kind))
for event in story:
if event["kind"] == WORKER_WAITING:
p = manager.inbox.get(event["payload"].get("prompt_id", ""))
if p:
- intervals.append((stamp(p.created_at), stamp(p.resolved_at) if p.resolved_at else now, "waited_ms"))
+ intervals.append(
+ (
+ stamp(p.created_at),
+ stamp(p.resolved_at) if p.resolved_at else now,
+ "waited_ms",
+ )
+ )
item["tokens"] = add_tokens(attributed_usage)
- item["usage_partial"] = ambiguous or any(m.get("role") == "assistant" and not m.get("usage")
- for _, _, actor in windows[item["id"]] for m in records.get(actor, []))
- item["timing_partial"] = ambiguous or any(m.get("role") == "assistant" and not m.get("timing")
- for _, _, actor in windows[item["id"]] for m in records.get(actor, []))
+ item["usage_partial"] = ambiguous or any(
+ m.get("role") == "assistant" and not m.get("usage")
+ for _, _, actor in windows[item["id"]]
+ for m in records.get(actor, [])
+ )
+ item["timing_partial"] = ambiguous or any(
+ m.get("role") == "assistant" and not m.get("timing")
+ for _, _, actor in windows[item["id"]]
+ for m in records.get(actor, [])
+ )
item["timing"] = breakdown(intervals, started, closed) if intervals else None
if item["timing"]:
all_item_times.append(item["timing"])
lead = workers[0]
workers = workers[1:]
by_kind = add_tokens([lead["tokens"], *(w["tokens"] for w in workers)])
- counts = {g: sum(i["group"] == g for i in items) for g in ("waiting", "review", "working", "queued", "done")}
+ counts = {
+ g: sum(i["group"] == g for i in items)
+ for g in ("waiting", "review", "working", "queued", "done")
+ }
active = [i for i in items if i["group"] != "canceled"]
- end = max((stamp(e["ts"]) for e in events if e.get("item_id") in by_item and e["kind"] != ITEM_STATUS), default=now) if active and all(i["group"] == "done" for i in active) else now
+ end = (
+ max((stamp(i["created_ts"]) + i["elapsed_s"] for i in active), default=now)
+ if active and all(i["group"] == "done" for i in active)
+ else now
+ )
elapsed = max(0, end - stamp(team.created_at))
- timing = {k: sum(v[k] for v in all_item_times) for k in ("model_ms", "tool_ms", "waited_ms", "queued_ms")}
- return {"team_id": team.team_id, "space": team.space, "lead_session": team.lead_session,
- "items": items, "workers": workers, "lead": lead, "counts": counts,
- "totals": {"done": counts["done"], "total": len(active), "elapsed_s": elapsed,
- "tokens": sum(by_kind.values()), "tokens_by_kind": by_kind,
- "asks_answered": len(answered), "asks_waiting": sum(a["state"] == "pending" for a in asks)},
- "timing": timing if all_item_times else None, "timing_partial": any(i["timing_partial"] for i in items),
- "usage_partial": any(m.get("role") == "assistant" and not m.get("usage") for messages in records.values() for m in messages),
- "asks": ask_groups, "median_answer_s": median(answered) if answered else None,
- "usage_points": sorted(usage_points, key=lambda p: p["ts"]), "generated_at": now}
+ timing = {
+ k: sum(v[k] for v in all_item_times)
+ for k in ("model_ms", "tool_ms", "waited_ms", "queued_ms")
+ }
+ return {
+ "team_id": team.team_id,
+ "space": team.space,
+ "lead_session": team.lead_session,
+ "items": items,
+ "workers": workers,
+ "lead": lead,
+ "counts": counts,
+ "totals": {
+ "done": counts["done"],
+ "total": len(active),
+ "elapsed_s": elapsed,
+ "tokens": sum(by_kind.values()),
+ "tokens_by_kind": by_kind,
+ "asks_answered": len(answered),
+ "asks_waiting": sum(a["state"] == "pending" for a in asks),
+ },
+ "timing": timing if all_item_times else None,
+ "timing_partial": any(i["timing_partial"] for i in items),
+ "usage_partial": any(
+ m.get("role") == "assistant" and not m.get("usage")
+ for messages in records.values()
+ for m in messages
+ ),
+ "asks": ask_groups,
+ "median_answer_s": median(answered) if answered else None,
+ "usage_points": compact_usage_points(usage_points),
+ "generated_at": now,
+ }
diff --git a/tests/test_team_open_surface.py b/tests/test_team_open_surface.py
index 6cde40d5..2f70e664 100644
--- a/tests/test_team_open_surface.py
+++ b/tests/test_team_open_surface.py
@@ -189,6 +189,23 @@ def test_board_api_requires_a_token(api):
assert "board token" in response.json()["error"]
+def test_status_over_http_is_owned_and_never_kicks_a_wake(api, monkeypatch):
+ client, manager, app = api
+ item = manager.team_store.create_item("proj", LEAD, title="Invoice PDF", criteria="verified")
+ manager.team_store.assign("proj", LEAD, item["id"], "sam")
+ sam = manager.board_tokens.mint("sam", "worker")
+ maya = manager.board_tokens.mint("maya", "worker")
+ kicks = []
+ monkeypatch.setattr(manager, "kick_team_tick", lambda: kicks.append(True))
+ body = {"space": "proj", "id": item["id"], "text": "Running tests"}
+ allowed = client.post("/v1/board/items/status", json=body, headers={"Authorization": f"Bearer {sam}"})
+ assert allowed.status_code == 200
+ assert allowed.json()["status"] == "Running tests"
+ denied = client.post("/v1/board/items/status", json=body, headers={"Authorization": f"Bearer {maya}"})
+ assert denied.status_code >= 400
+ assert kicks == []
+
+
def test_board_api_rejects_the_sidecar_token_as_a_board_token(api):
client, _, _ = api
response = client.get(
diff --git a/tests/test_team_view.py b/tests/test_team_view.py
index c425b486..afe201ce 100644
--- a/tests/test_team_view.py
+++ b/tests/test_team_view.py
@@ -1,4 +1,5 @@
"""Team View contracts: status ownership, quiet delivery, projections and timing."""
+
import asyncio
import json
from types import SimpleNamespace
@@ -6,7 +7,13 @@ from types import SimpleNamespace
import pytest
from coworker.teams import Actor, AuthorityError, BoardError, Role, TeamStore
from coworker.teams.store import ITEM_STATUS, WORKER_WAITING
-from coworker.teams.summary import all_events, breakdown, waiting_items, tokens, make_summary
+from coworker.teams.summary import (
+ all_events,
+ breakdown,
+ waiting_items,
+ tokens,
+ make_summary,
+)
from coworker.teams.tools import board_tools, with_mention
from coworker.teams.registry import TeamRegistry, TeamWorker
from coworker.inbox import InboxStore
@@ -26,7 +33,9 @@ def store(tmp_path):
def assigned(store):
- item = store.create_item("acme", LEAD, title="Invoice PDF", criteria="Download verified")
+ item = store.create_item(
+ "acme", LEAD, title="Invoice PDF", criteria="Download verified"
+ )
return store.assign("acme", LEAD, item["id"], WORKER.id)
@@ -96,14 +105,29 @@ def test_mention_escapes_markdown_and_does_not_mutate():
def test_role_filtered_tool_surface(store):
- assert "set_status" in {t.__name__ for t in board_tools(store, space="acme", actor=WORKER)}
- assert "set_status" not in {t.__name__ for t in board_tools(store, space="acme", actor=LEAD)}
+ assert "set_status" in {
+ t.__name__ for t in board_tools(store, space="acme", actor=WORKER)
+ }
+ assert "set_status" not in {
+ t.__name__ for t in board_tools(store, space="acme", actor=LEAD)
+ }
def test_waiting_requires_a_current_pending_prompt():
inbox = InboxStore()
p = inbox.add_approval("worker-session", "Run tests")
- events = [{"kind": WORKER_WAITING, "item_id": 2, "actor": "sam", "payload": {"prompt_id": p.id, "session_id": "worker-session", "tool": "run_shell"}}]
+ events = [
+ {
+ "kind": WORKER_WAITING,
+ "item_id": 2,
+ "actor": "sam",
+ "payload": {
+ "prompt_id": p.id,
+ "session_id": "worker-session",
+ "tool": "run_shell",
+ },
+ }
+ ]
items = [{"id": 2, "assignee": "sam", "state": "in_progress"}]
assert 2 in waiting_items(events, inbox, items)
inbox.resolve(p.id, "allow")
@@ -111,18 +135,24 @@ def test_waiting_requires_a_current_pending_prompt():
def test_parallel_time_is_wall_clock_union():
- result = breakdown([(0, 5, "tool_ms"), (2, 8, "tool_ms"), (8, 9, "model_ms")], 0, 10)
+ result = breakdown(
+ [(0, 5, "tool_ms"), (2, 8, "tool_ms"), (8, 9, "model_ms")], 0, 10
+ )
assert result == dict(model_ms=1000, tool_ms=8000, waited_ms=0, queued_ms=1000)
assert sum(result.values()) == 10000
def test_waiting_has_priority_and_intervals_are_clipped():
- result = breakdown([(-5, 20, "model_ms"), (3, 6, "waited_ms"), (2, 7, "tool_ms")], 0, 10)
+ result = breakdown(
+ [(-5, 20, "model_ms"), (3, 6, "waited_ms"), (2, 7, "tool_ms")], 0, 10
+ )
assert result == dict(model_ms=5000, tool_ms=2000, waited_ms=3000, queued_ms=0)
def test_usage_is_normalized_defensively():
- assert tokens(dict(input=-1, output="12", cache_read=None, cache_write="bad")) == dict(input=0, output=12, cache_read=0, cache_write=0)
+ assert tokens(
+ dict(input=-1, output="12", cache_read=None, cache_write="bad")
+ ) == dict(input=0, output=12, cache_read=0, cache_write=0)
def test_event_reader_paginates(store):
@@ -134,13 +164,51 @@ def test_event_reader_paginates(store):
def test_real_manager_summary_and_current_wait(tmp_path, monkeypatch):
monkeypatch.setenv("COWORKER_STATE_DIR", str(tmp_path / "state"))
manager = SessionManager(workspace=tmp_path / "repo", data_dir=tmp_path / "data")
- manager.session_store.save(SessionRecord(session_id="lead-session", workspace=str(tmp_path / "repo"), model="sample", mode="interactive", agent="swe-lead"))
- team = manager.teams.create(space=manager._board_space("lead-session"), lead_session="lead-session", lead_actor="lead", workers=[TeamWorker(actor="sam", persona="swe-worker", session_id="worker-session")])
- item = manager.team_store.create_item(team.space, LEAD, title="Invoice PDF", criteria="Verified")
+ manager.session_store.save(
+ SessionRecord(
+ session_id="lead-session",
+ workspace=str(tmp_path / "repo"),
+ model="sample",
+ mode="interactive",
+ agent="swe-lead",
+ )
+ )
+ team = manager.teams.create(
+ space=manager._board_space("lead-session"),
+ lead_session="lead-session",
+ lead_actor="lead",
+ workers=[
+ TeamWorker(actor="sam", persona="swe-worker", session_id="worker-session")
+ ],
+ )
+ item = manager.team_store.create_item(
+ team.space, LEAD, title="Invoice PDF", criteria="Verified"
+ )
manager.team_store.assign(team.space, LEAD, item["id"], WORKER.id)
manager.team_store.transition(team.space, WORKER, item["id"], "in_progress")
manager.team_store.set_status(team.space, WORKER, item["id"], "Running tests")
- manager.session_store.save(SessionRecord(session_id="worker-session", workspace=str(tmp_path / "repo"), model="sample", mode="interactive", agent="swe-worker", messages=[{"role": "assistant", "content": "Testing", "ts": 10, "usage": {"input": 10, "cache_read": 20, "output": 5, "cache_write": 2}}]))
+ manager.session_store.save(
+ SessionRecord(
+ session_id="worker-session",
+ workspace=str(tmp_path / "repo"),
+ model="sample",
+ mode="interactive",
+ agent="swe-worker",
+ messages=[
+ {
+ "role": "assistant",
+ "content": "Testing",
+ "ts": 10,
+ "usage": {
+ "input": 10,
+ "cache_read": 20,
+ "output": 5,
+ "cache_write": 2,
+ },
+ }
+ ],
+ )
+ )
summary = manager.team_summary(team.team_id)
assert summary["items"][0]["status"] == "Running tests"
assert summary["totals"]["tokens"] == 37
@@ -152,8 +220,11 @@ def test_real_manager_summary_and_current_wait(tmp_path, monkeypatch):
def test_engine_records_and_strips_timing(tmp_path):
from test_engine import _engine, _tool_turn, _text_turn, _collect
+
(tmp_path / "sample.txt").write_text("sample")
- engine, _ = _engine(tmp_path, [_tool_turn("read_file", {"path": "sample.txt"}), _text_turn("Read")])
+ engine, _ = _engine(
+ tmp_path, [_tool_turn("read_file", {"path": "sample.txt"}), _text_turn("Read")]
+ )
_collect(engine, "read")
model = next(m for m in engine.messages if m["role"] == "assistant")
tool = next(m for m in engine.messages if m["role"] == "tool")
@@ -165,23 +236,88 @@ def test_engine_records_and_strips_timing(tmp_path):
def test_call_after_review_is_not_attributed_to_the_previous_item():
events = [
{"seq": 1, "ts": 1, "kind": "item_created", "item_id": 1, "payload": {}},
- {"seq": 2, "ts": 10, "kind": "item_assigned", "item_id": 1, "payload": {"assignee": "sam"}},
- {"seq": 3, "ts": 20, "kind": "item_transitioned", "item_id": 1, "payload": {"to": "review"}},
- {"seq": 4, "ts": 30, "kind": "item_transitioned", "item_id": 1, "payload": {"to": "done"}},
+ {
+ "seq": 2,
+ "ts": 10,
+ "kind": "item_assigned",
+ "item_id": 1,
+ "payload": {"assignee": "sam"},
+ },
+ {
+ "seq": 3,
+ "ts": 20,
+ "kind": "item_transitioned",
+ "item_id": 1,
+ "payload": {"to": "review"},
+ },
+ {
+ "seq": 4,
+ "ts": 30,
+ "kind": "item_transitioned",
+ "item_id": 1,
+ "payload": {"to": "done"},
+ },
]
- item = {"id": 1, "title": "Task", "created_ts": 1, "creator": "lead", "assignee": "sam", "state": "done", "refs": [], "links": []}
- worker = SessionRecord(session_id="worker", workspace="acme", model="sample", mode="interactive", messages=[
- {"role": "assistant", "ts": 29, "usage": {"input": 100}, "timing": {"model_started": 25, "model_ms": 10000}}
- ])
+ item = {
+ "id": 1,
+ "title": "Task",
+ "created_ts": 1,
+ "creator": "lead",
+ "assignee": "sam",
+ "state": "done",
+ "refs": [],
+ "links": [],
+ }
+ worker = SessionRecord(
+ session_id="worker",
+ workspace="acme",
+ model="sample",
+ mode="interactive",
+ messages=[
+ {
+ "role": "assistant",
+ "ts": 29,
+ "usage": {"input": 100},
+ "timing": {"model_started": 25, "model_ms": 10000},
+ }
+ ],
+ )
manager = SimpleNamespace(
session_board=lambda sid: {"items": [item]},
- team_store=SimpleNamespace(events=lambda space, since_seq, limit: [e for e in events if e["seq"] > since_seq]),
- inbox=InboxStore(), _engines={}, is_running=lambda sid: False,
- session_store=SimpleNamespace(load=lambda sid: worker if sid == "worker" else None),
+ team_store=SimpleNamespace(
+ events=lambda space, since_seq, limit: [
+ e for e in events if e["seq"] > since_seq
+ ]
+ ),
+ inbox=InboxStore(),
+ _engines={},
+ is_running=lambda sid: False,
+ session_store=SimpleNamespace(
+ load=lambda sid: worker if sid == "worker" else None
+ ),
+ )
+ team = SimpleNamespace(
+ lead_session="lead",
+ lead_actor="lead",
+ team_id="t",
+ space="acme",
+ created_at=1,
+ workers=[TeamWorker(actor="sam", persona="swe-worker", session_id="worker")],
)
- team = SimpleNamespace(lead_session="lead", lead_actor="lead", team_id="t", space="acme", created_at=1,
- workers=[TeamWorker(actor="sam", persona="swe-worker", session_id="worker")])
result = make_summary(manager, team, 40)
assert result["items"][0]["timing"] is None
assert sum(result["items"][0]["tokens"].values()) == 0
assert result["totals"]["tokens"] == 100
+
+
+def test_chart_bucketing_preserves_all_tokens_by_role():
+ from coworker.teams.summary import compact_usage_points
+
+ points = [
+ {"ts": n, "role": "lead" if n % 2 else "swe-worker", "tokens": 3}
+ for n in range(10000)
+ ]
+ result = compact_usage_points(points)
+ assert len(result) <= 240
+ assert sum(p["tokens"] for p in result) == 30000
+ assert sum(p["tokens"] for p in result if p["role"] == "lead") == 15000