agent-usage-tracker 0.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- agent_usage_tracker/__init__.py +8 -0
- agent_usage_tracker/backup.py +225 -0
- agent_usage_tracker/cli.py +479 -0
- agent_usage_tracker/collectors/__init__.py +115 -0
- agent_usage_tracker/collectors/_util.py +310 -0
- agent_usage_tracker/collectors/activity_common.py +294 -0
- agent_usage_tracker/collectors/activitywatch.py +394 -0
- agent_usage_tracker/collectors/claude_code.py +484 -0
- agent_usage_tracker/collectors/claude_code_activity.py +317 -0
- agent_usage_tracker/collectors/codex.py +612 -0
- agent_usage_tracker/collectors/commits.py +154 -0
- agent_usage_tracker/collectors/cursor.py +193 -0
- agent_usage_tracker/collectors/cursor_api.py +247 -0
- agent_usage_tracker/collectors/cursor_cli.py +592 -0
- agent_usage_tracker/collectors/cursor_sdk.py +95 -0
- agent_usage_tracker/collectors/generic_sdk.py +100 -0
- agent_usage_tracker/collectors/pi.py +397 -0
- agent_usage_tracker/collectors/session_digest.py +379 -0
- agent_usage_tracker/collectors/t3.py +490 -0
- agent_usage_tracker/config.py +392 -0
- agent_usage_tracker/coverage.py +615 -0
- agent_usage_tracker/coverage_catalog.json +588 -0
- agent_usage_tracker/cursor_run.py +62 -0
- agent_usage_tracker/diagnostics.py +74 -0
- agent_usage_tracker/episode_store.py +78 -0
- agent_usage_tracker/event.py +132 -0
- agent_usage_tracker/git_day_cache.py +128 -0
- agent_usage_tracker/git_days.py +499 -0
- agent_usage_tracker/git_safe.py +58 -0
- agent_usage_tracker/harness_inventory.py +208 -0
- agent_usage_tracker/history_steers.py +88 -0
- agent_usage_tracker/improve/__init__.py +1 -0
- agent_usage_tracker/improve/case_view.py +112 -0
- agent_usage_tracker/improve/cli.py +138 -0
- agent_usage_tracker/improve/desktop.py +347 -0
- agent_usage_tracker/improve/evidence.py +322 -0
- agent_usage_tracker/improve/judge.py +671 -0
- agent_usage_tracker/improve/proposals.py +72 -0
- agent_usage_tracker/improve/rubric.py +360 -0
- agent_usage_tracker/improve/rubrics/avoidable_recovery.md +20 -0
- agent_usage_tracker/improve/rubrics/code_quality.md +22 -0
- agent_usage_tracker/improve/rubrics/completion_supported.md +20 -0
- agent_usage_tracker/improve/rubrics/redundant_tests.md +21 -0
- agent_usage_tracker/improve/rubrics/verbosity.md +21 -0
- agent_usage_tracker/improve/runtime.py +117 -0
- agent_usage_tracker/improve/schedule.py +291 -0
- agent_usage_tracker/improve/search_check.py +563 -0
- agent_usage_tracker/improve/store.py +244 -0
- agent_usage_tracker/improve/studies.py +285 -0
- agent_usage_tracker/improve/study_cli.py +46 -0
- agent_usage_tracker/improve/study_decision.py +121 -0
- agent_usage_tracker/improve/study_followup.py +174 -0
- agent_usage_tracker/improve/t3.py +269 -0
- agent_usage_tracker/improve/trial.py +318 -0
- agent_usage_tracker/improve/validation.py +148 -0
- agent_usage_tracker/inspector.py +350 -0
- agent_usage_tracker/label/__init__.py +22 -0
- agent_usage_tracker/label/backends.py +742 -0
- agent_usage_tracker/label/bakeoff.py +661 -0
- agent_usage_tracker/label/baseline.py +173 -0
- agent_usage_tracker/label/clef.py +258 -0
- agent_usage_tracker/label/context.py +382 -0
- agent_usage_tracker/label/core.py +355 -0
- agent_usage_tracker/label/covered_run.py +92 -0
- agent_usage_tracker/label/evalset.py +154 -0
- agent_usage_tracker/label/gold.py +209 -0
- agent_usage_tracker/label/local.py +326 -0
- agent_usage_tracker/label/local_parallel.py +232 -0
- agent_usage_tracker/label/merge.py +57 -0
- agent_usage_tracker/label/outcome_coverage.py +57 -0
- agent_usage_tracker/label/replay.py +194 -0
- agent_usage_tracker/label/run.py +339 -0
- agent_usage_tracker/label/tasks.py +579 -0
- agent_usage_tracker/label/teach.py +145 -0
- agent_usage_tracker/local_page.py +128 -0
- agent_usage_tracker/manual_time.py +69 -0
- agent_usage_tracker/net.py +23 -0
- agent_usage_tracker/observation_cli.py +65 -0
- agent_usage_tracker/observation_report.py +116 -0
- agent_usage_tracker/observation_server.py +110 -0
- agent_usage_tracker/observations.py +365 -0
- agent_usage_tracker/outcome_git.py +383 -0
- agent_usage_tracker/outcomes.py +464 -0
- agent_usage_tracker/pricing.py +248 -0
- agent_usage_tracker/push.py +530 -0
- agent_usage_tracker/redact.py +45 -0
- agent_usage_tracker/repo_keys.py +165 -0
- agent_usage_tracker/review.py +252 -0
- agent_usage_tracker/scan.py +472 -0
- agent_usage_tracker/scheduled.py +164 -0
- agent_usage_tracker/sdk.py +79 -0
- agent_usage_tracker/sink.py +82 -0
- agent_usage_tracker/sink_stack/README.md +71 -0
- agent_usage_tracker/sink_stack/apply.sh +63 -0
- agent_usage_tracker/sink_stack/compose.yaml +94 -0
- agent_usage_tracker/sink_stack/env.example +20 -0
- agent_usage_tracker/sink_stack/grafana/dashboards/agents-effort.json +634 -0
- agent_usage_tracker/sink_stack/grafana/dashboards/agents-usage.json +1279 -0
- agent_usage_tracker/sink_stack/grafana/provisioning/dashboards/aut.yaml +9 -0
- agent_usage_tracker/sink_stack/grafana/provisioning/datasources/agents.yaml +16 -0
- agent_usage_tracker/sink_stack/sql/01_schema.sql +1186 -0
- agent_usage_tracker/sink_stack/sql/02_price.sql +60 -0
- agent_usage_tracker/sink_stack/sql/03_effort.sql +1180 -0
- agent_usage_tracker/sink_stack/sql/04_grants.sql +149 -0
- agent_usage_tracker/sink_stack/sql/refresh.sql +66 -0
- agent_usage_tracker/sources.py +180 -0
- agent_usage_tracker/state.py +191 -0
- agent_usage_tracker/telemetry.py +165 -0
- agent_usage_tracker-0.1.0.dist-info/METADATA +366 -0
- agent_usage_tracker-0.1.0.dist-info/RECORD +113 -0
- agent_usage_tracker-0.1.0.dist-info/WHEEL +4 -0
- agent_usage_tracker-0.1.0.dist-info/entry_points.txt +3 -0
- agent_usage_tracker-0.1.0.dist-info/licenses/LICENSE +21 -0
|
@@ -0,0 +1,382 @@
|
|
|
1
|
+
"""Read local agent text against collector identities, without advancing ingestion state."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
import sqlite3
|
|
6
|
+
from collections import Counter
|
|
7
|
+
from dataclasses import dataclass
|
|
8
|
+
from datetime import datetime
|
|
9
|
+
from itertools import islice
|
|
10
|
+
from pathlib import Path
|
|
11
|
+
from typing import Any, Iterable
|
|
12
|
+
|
|
13
|
+
from ..collectors import claude_code, codex, cursor_cli, pi, t3
|
|
14
|
+
from ..collectors.activity_common import first_record, text_of
|
|
15
|
+
from ..collectors.cursor import connect_ro
|
|
16
|
+
from ..cursor_run import automated_sessions
|
|
17
|
+
from ..event import CollectorUnavailable, make_event_id
|
|
18
|
+
from ..repo_keys import RepoKeys
|
|
19
|
+
from ..state import parse_record
|
|
20
|
+
|
|
21
|
+
AGENTS = frozenset({"claude_code", "codex", "cursor_cli", "pi", "t3"})
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
def session_paths(home: Path, agent: str) -> dict[str, list[tuple[int, str]]]:
|
|
25
|
+
"""Canonical session identities and paths in discovery order, without retaining text."""
|
|
26
|
+
if agent == "t3":
|
|
27
|
+
try:
|
|
28
|
+
with t3.connection(home) as con:
|
|
29
|
+
return {"t3:" + tid: [(order, str(t3.database_path(home)))]
|
|
30
|
+
for order, (tid,) in enumerate(t3._read(con,
|
|
31
|
+
f"SELECT thread_id FROM {t3._PREFIX}threads"))}
|
|
32
|
+
except (OSError, sqlite3.Error, ValueError, TypeError) as exc:
|
|
33
|
+
raise CollectorUnavailable("t3: local context index unavailable") from exc
|
|
34
|
+
discover = {"claude_code": claude_code.main_session_files, "codex": codex.session_files,
|
|
35
|
+
"cursor_cli": cursor_cli.chat_stores, "pi": pi.session_files}[agent]
|
|
36
|
+
index: dict[str, list[tuple[int, str]]] = {}
|
|
37
|
+
for order, path in enumerate(discover(home)):
|
|
38
|
+
sid: Any = None
|
|
39
|
+
if agent == "claude_code":
|
|
40
|
+
sid = Path(path).stem
|
|
41
|
+
elif agent == "cursor_cli":
|
|
42
|
+
sid = Path(path).parent.name
|
|
43
|
+
else:
|
|
44
|
+
head = first_record(path)
|
|
45
|
+
if agent == "codex":
|
|
46
|
+
payload = head.get("payload") if head else None
|
|
47
|
+
if isinstance(payload, dict):
|
|
48
|
+
sid = payload.get("id") or payload.get("session_id")
|
|
49
|
+
elif head and head.get("type") == "session":
|
|
50
|
+
sid = pi.header(head, path)["id"]
|
|
51
|
+
if isinstance(sid, str):
|
|
52
|
+
index.setdefault(sid, []).append((order, path))
|
|
53
|
+
return index
|
|
54
|
+
|
|
55
|
+
|
|
56
|
+
@dataclass(frozen=True)
|
|
57
|
+
class Context:
|
|
58
|
+
text: str
|
|
59
|
+
before: str = ""
|
|
60
|
+
after: str = ""
|
|
61
|
+
prompt_ordinal: int | None = None
|
|
62
|
+
|
|
63
|
+
|
|
64
|
+
def _records(path: str) -> Iterable[dict[str, Any]]:
|
|
65
|
+
with open(path, encoding="utf-8", errors="replace") as stream:
|
|
66
|
+
for line in stream:
|
|
67
|
+
record = parse_record(line)
|
|
68
|
+
if record is not None:
|
|
69
|
+
yield record
|
|
70
|
+
|
|
71
|
+
|
|
72
|
+
def _in_time(record: dict[str, Any], until: datetime | None) -> bool:
|
|
73
|
+
if until is None:
|
|
74
|
+
return True
|
|
75
|
+
try:
|
|
76
|
+
return datetime.fromisoformat(str(record.get("timestamp")).replace("Z", "+00:00")) < until
|
|
77
|
+
except (ValueError, TypeError):
|
|
78
|
+
return False
|
|
79
|
+
|
|
80
|
+
|
|
81
|
+
def _join(events: Iterable[tuple[str | None, str, int | None]], want: set[str]) -> dict[str, Context]:
|
|
82
|
+
out: dict[str, Context] = {}
|
|
83
|
+
last, current = "", None
|
|
84
|
+
for eid, text, ordinal in events:
|
|
85
|
+
if eid is not None:
|
|
86
|
+
current = eid if eid in want else None
|
|
87
|
+
if current:
|
|
88
|
+
out[current] = Context(text, last, prompt_ordinal=ordinal)
|
|
89
|
+
elif text.strip():
|
|
90
|
+
last = text
|
|
91
|
+
if current:
|
|
92
|
+
previous = out[current]
|
|
93
|
+
out[current] = Context(previous.text, previous.before, text, previous.prompt_ordinal)
|
|
94
|
+
return out
|
|
95
|
+
|
|
96
|
+
|
|
97
|
+
def _codex_records(path: str, cur: codex.FileState) -> Iterable[codex.ParsedRecord]:
|
|
98
|
+
with open(path, "rb") as stream:
|
|
99
|
+
offset = 0
|
|
100
|
+
while lines := list(islice(stream, 1000)):
|
|
101
|
+
yield from codex.parsed_records(lines, offset, cur, Counter())
|
|
102
|
+
offset += len(lines)
|
|
103
|
+
|
|
104
|
+
|
|
105
|
+
def _codex_prefixes(path: str, keys: RepoKeys, want: set[str]) -> list[int]:
|
|
106
|
+
# An incrementally collected old file can acquire user_message records later. Resolve
|
|
107
|
+
# the retained legacy prefix against existing IDs, rather than renumber those inputs.
|
|
108
|
+
cur = codex._fresh_state(Path(path).stat())
|
|
109
|
+
legacy: list[str] = []
|
|
110
|
+
seen_modern = False
|
|
111
|
+
modern: list[tuple[str, str, int]] = []
|
|
112
|
+
for record, kind, payload, copied in _codex_records(path, cur):
|
|
113
|
+
codex._note_context(kind, payload, copied, cur, keys)
|
|
114
|
+
if copied:
|
|
115
|
+
continue
|
|
116
|
+
is_modern = kind == "event_msg" and payload.get("type") == "user_message"
|
|
117
|
+
seen_modern |= is_modern
|
|
118
|
+
prompt = codex.prompt_of(kind, payload, seen_modern)
|
|
119
|
+
if prompt is None or codex.is_scheduled_automation(prompt[0]):
|
|
120
|
+
continue
|
|
121
|
+
seed = cur["session_id"] or Path(path).name
|
|
122
|
+
stamp = record.get("timestamp") or ""
|
|
123
|
+
if is_modern:
|
|
124
|
+
modern.append((seed, stamp, len(modern) + 1))
|
|
125
|
+
else:
|
|
126
|
+
legacy.append(make_event_id("codex", "steer", seed, stamp, len(legacy) + 1))
|
|
127
|
+
if not modern:
|
|
128
|
+
return [len(legacy)] if set(legacy) & want else []
|
|
129
|
+
profiles: dict[tuple[tuple[str, str, int], ...], tuple[int, int]] = {}
|
|
130
|
+
for offset in range(len(legacy) + 1):
|
|
131
|
+
matched = [(eid, "legacy", i) for i, eid in enumerate(legacy[:offset]) if eid in want]
|
|
132
|
+
matched.extend((eid, "modern", ordinal) for seed, stamp, ordinal in modern
|
|
133
|
+
if (eid := make_event_id("codex", "steer", seed, stamp,
|
|
134
|
+
offset + ordinal)) in want)
|
|
135
|
+
# Equivalent interpretations need only one bounded replay of the log.
|
|
136
|
+
signature = tuple(sorted(matched))
|
|
137
|
+
profiles[signature] = len({eid for eid, kind, index in matched}), offset
|
|
138
|
+
best = max(count for count, offset in profiles.values())
|
|
139
|
+
return [offset for count, offset in profiles.values() if count == best] if best else []
|
|
140
|
+
|
|
141
|
+
|
|
142
|
+
def _codex_events(path: str, until: datetime | None, keys: RepoKeys, prefix: int,
|
|
143
|
+
imported_until: datetime | None = None,
|
|
144
|
+
) -> Iterable[tuple[str | None, str, int | None]]:
|
|
145
|
+
cur = codex._fresh_state(Path(path).stat())
|
|
146
|
+
legacy, modern_n, seen_modern = 0, 0, False
|
|
147
|
+
for record, kind, payload, copied in _codex_records(path, cur):
|
|
148
|
+
codex._note_context(kind, payload, copied, cur, keys)
|
|
149
|
+
if copied:
|
|
150
|
+
continue
|
|
151
|
+
is_modern = kind == "event_msg" and payload.get("type") == "user_message"
|
|
152
|
+
seen_modern |= is_modern
|
|
153
|
+
prompt = codex.prompt_of(kind, payload, seen_modern)
|
|
154
|
+
if prompt is not None and codex.is_scheduled_automation(prompt[0]):
|
|
155
|
+
yield "", "", None # A software prompt ends the preceding human prompt's reply interval.
|
|
156
|
+
elif prompt is not None:
|
|
157
|
+
if is_modern:
|
|
158
|
+
modern_n += 1
|
|
159
|
+
ordinal = prefix + modern_n
|
|
160
|
+
else:
|
|
161
|
+
legacy += 1
|
|
162
|
+
if legacy > prefix:
|
|
163
|
+
yield "", "", None
|
|
164
|
+
continue
|
|
165
|
+
ordinal = legacy
|
|
166
|
+
if codex.imported_record(record, imported_until):
|
|
167
|
+
yield "", "", None
|
|
168
|
+
elif _in_time(record, until):
|
|
169
|
+
yield make_event_id("codex", "steer", cur["session_id"] or Path(path).name,
|
|
170
|
+
record.get("timestamp") or "", ordinal), prompt[0], ordinal
|
|
171
|
+
elif not codex.imported_record(record, imported_until) and _in_time(record, until):
|
|
172
|
+
if kind == "response_item" and payload.get("type") == "message" \
|
|
173
|
+
and payload.get("role") == "assistant":
|
|
174
|
+
content = payload.get("content")
|
|
175
|
+
blocks = content if isinstance(content, list) else []
|
|
176
|
+
yield None, "\n".join(b["text"] for b in blocks if isinstance(b, dict)
|
|
177
|
+
and b.get("type") in {"text", "output_text"}
|
|
178
|
+
and isinstance(b.get("text"), str)), None
|
|
179
|
+
elif kind == "event_msg" and payload.get("type") == "agent_message":
|
|
180
|
+
message = payload.get("message")
|
|
181
|
+
if isinstance(message, str):
|
|
182
|
+
yield None, message, None
|
|
183
|
+
|
|
184
|
+
|
|
185
|
+
def codex_context(home: Path, sessions: set[str], want: set[str],
|
|
186
|
+
until: datetime | None, keys: RepoKeys, *,
|
|
187
|
+
paths: Iterable[str] | None = None) -> dict[str, Context]:
|
|
188
|
+
out: dict[str, Context] = {}
|
|
189
|
+
imports = codex.imported_until(home)
|
|
190
|
+
for path in codex.session_files(home) if paths is None else paths:
|
|
191
|
+
head = first_record(path)
|
|
192
|
+
payload = head.get("payload") if head else None
|
|
193
|
+
if not isinstance(payload, dict):
|
|
194
|
+
continue
|
|
195
|
+
sid = payload.get("id") or payload.get("session_id")
|
|
196
|
+
if not isinstance(sid, str) or sid not in sessions:
|
|
197
|
+
continue
|
|
198
|
+
if codex.run_origin(payload) in {"subagent", "automated"}:
|
|
199
|
+
continue
|
|
200
|
+
imported_until = imports.get(sid)
|
|
201
|
+
profiles = [_join(_codex_events(path, until, keys, prefix, imported_until), want)
|
|
202
|
+
for prefix in _codex_prefixes(path, keys, want)]
|
|
203
|
+
if not profiles:
|
|
204
|
+
continue
|
|
205
|
+
for eid in set.intersection(*(set(p) for p in profiles)):
|
|
206
|
+
values = [p[eid] for p in profiles]
|
|
207
|
+
if len({v.text for v in values}) == 1:
|
|
208
|
+
out[eid] = Context(values[0].text,
|
|
209
|
+
values[0].before if len({v.before for v in values}) == 1 else "",
|
|
210
|
+
values[0].after if len({v.after for v in values}) == 1 else "",
|
|
211
|
+
values[0].prompt_ordinal if len({v.prompt_ordinal for v in values}) == 1 else None)
|
|
212
|
+
return out
|
|
213
|
+
|
|
214
|
+
|
|
215
|
+
def _pi_context(path: str, want: set[str], until: datetime | None) -> dict[str, Context]:
|
|
216
|
+
head = first_record(path)
|
|
217
|
+
if not head or head.get("type") != "session":
|
|
218
|
+
return {}
|
|
219
|
+
hdr = pi.header(head, path)
|
|
220
|
+
nodes: dict[str, dict[str, Any]] = {}
|
|
221
|
+
for record in _records(path):
|
|
222
|
+
if pi.is_fork_copy(record, hdr) or not _in_time(record, until):
|
|
223
|
+
continue
|
|
224
|
+
rid = record.get("id")
|
|
225
|
+
if isinstance(rid, str):
|
|
226
|
+
nodes[rid] = record
|
|
227
|
+
|
|
228
|
+
def ancestors(record: dict[str, Any]) -> Iterable[str]:
|
|
229
|
+
parent = record.get("parentId")
|
|
230
|
+
seen = set()
|
|
231
|
+
while isinstance(parent, str) and parent in nodes and parent not in seen:
|
|
232
|
+
seen.add(parent)
|
|
233
|
+
yield parent
|
|
234
|
+
parent = nodes[parent].get("parentId")
|
|
235
|
+
|
|
236
|
+
def message(record: dict[str, Any]) -> dict[str, Any]:
|
|
237
|
+
value = record.get("message")
|
|
238
|
+
return value if record.get("type") == "message" and isinstance(value, dict) else {}
|
|
239
|
+
|
|
240
|
+
replies: dict[str, list[str]] = {}
|
|
241
|
+
for rid, record in nodes.items():
|
|
242
|
+
msg = message(record)
|
|
243
|
+
if msg.get("role") == "assistant" and text_of(msg.get("content")).strip():
|
|
244
|
+
for parent in ancestors(record):
|
|
245
|
+
if message(nodes[parent]).get("role") == "user":
|
|
246
|
+
replies.setdefault(parent, []).append(rid)
|
|
247
|
+
break
|
|
248
|
+
out: dict[str, Context] = {}
|
|
249
|
+
for rid, record in nodes.items():
|
|
250
|
+
msg = message(record)
|
|
251
|
+
eid = make_event_id("pi", "steer", f"{hdr['id']}|{rid}")
|
|
252
|
+
if msg.get("role") != "user" or eid not in want:
|
|
253
|
+
continue
|
|
254
|
+
before = next((text_of(message(nodes[p]).get("content")) for p in ancestors(record)
|
|
255
|
+
if message(nodes[p]).get("role") == "assistant"
|
|
256
|
+
and text_of(message(nodes[p]).get("content")).strip()), "")
|
|
257
|
+
candidates = replies.get(rid, [])
|
|
258
|
+
terminal = [p for p in candidates if not any(
|
|
259
|
+
p in ancestors(nodes[q]) for q in candidates if p != q)]
|
|
260
|
+
# An append-only Pi file can contain several branches. Only a unique reply path is safe.
|
|
261
|
+
after = text_of(message(nodes[terminal[0]]).get("content")) if len(terminal) == 1 else ""
|
|
262
|
+
out[eid] = Context(text_of(msg.get("content")), before, after)
|
|
263
|
+
return out
|
|
264
|
+
|
|
265
|
+
|
|
266
|
+
def pi_context(home: Path, sessions: set[str], want: set[str],
|
|
267
|
+
until: datetime | None, *, paths: Iterable[str] | None = None) -> dict[str, Context]:
|
|
268
|
+
out: dict[str, Context] = {}
|
|
269
|
+
for path in pi.session_files(home) if paths is None else paths:
|
|
270
|
+
head = first_record(path)
|
|
271
|
+
if not head or head.get("type") != "session" or pi.header(head, path)["id"] not in sessions:
|
|
272
|
+
continue
|
|
273
|
+
out.update(_pi_context(path, want, until))
|
|
274
|
+
return out
|
|
275
|
+
|
|
276
|
+
|
|
277
|
+
def cursor_context(home: Path, sessions: set[str], want: set[str],
|
|
278
|
+
until: datetime | None, keys: RepoKeys, *,
|
|
279
|
+
paths: Iterable[str] | None = None) -> dict[str, Context]:
|
|
280
|
+
out: dict[str, Context] = {}
|
|
281
|
+
ambiguous: set[str] = set()
|
|
282
|
+
for path in cursor_cli.chat_stores(home) if paths is None else paths:
|
|
283
|
+
if Path(path).parent.name not in sessions:
|
|
284
|
+
continue
|
|
285
|
+
con = connect_ro(path)
|
|
286
|
+
if con is None:
|
|
287
|
+
raise CollectorUnavailable("cursor_cli: context store unreadable")
|
|
288
|
+
try:
|
|
289
|
+
meta = cursor_cli.chat_meta(con)
|
|
290
|
+
if not meta or meta.get("subagentInfo"):
|
|
291
|
+
continue
|
|
292
|
+
sid = meta.get("agentId") or Path(path).parent.name
|
|
293
|
+
if sid not in sessions:
|
|
294
|
+
continue
|
|
295
|
+
lists = cursor_cli.root_lists(con)
|
|
296
|
+
messages = cursor_cli.messages(con, {i for seq in lists for i in seq})
|
|
297
|
+
if sid in automated_sessions():
|
|
298
|
+
continue
|
|
299
|
+
first_ids: set[str] = set()
|
|
300
|
+
listed = {mid for seq in lists for mid in seq}
|
|
301
|
+
for seq in lists:
|
|
302
|
+
seen_requests: set[str] = set()
|
|
303
|
+
for mid in seq:
|
|
304
|
+
rid = cursor_cli.request_id(messages.get(mid, {}))
|
|
305
|
+
if rid and rid not in seen_requests:
|
|
306
|
+
first_ids.add(mid)
|
|
307
|
+
seen_requests.add(rid)
|
|
308
|
+
bases: dict[str, set[str]] = {}
|
|
309
|
+
prompts: list[tuple[str, str]] = []
|
|
310
|
+
for mid, prompt in messages.items():
|
|
311
|
+
rid = cursor_cli.request_id(prompt)
|
|
312
|
+
if not rid:
|
|
313
|
+
continue
|
|
314
|
+
text = text_of(prompt.get("content"))
|
|
315
|
+
query = cursor_cli.USER_QUERY.search(text)
|
|
316
|
+
text = query[1] if query else text
|
|
317
|
+
# Blob-suffixed IDs identify their text even if a later root starts with that blob.
|
|
318
|
+
prompts.append((make_event_id("cursor_cli", "steer", f"{sid}|{rid}|{mid}"), text))
|
|
319
|
+
if mid in first_ids or mid not in listed:
|
|
320
|
+
bases.setdefault(rid, set()).add(text)
|
|
321
|
+
for rid, texts in bases.items():
|
|
322
|
+
eid = make_event_id("cursor_cli", "steer", f"{sid}|{rid}")
|
|
323
|
+
if len(texts) != 1:
|
|
324
|
+
ambiguous.add(eid)
|
|
325
|
+
else:
|
|
326
|
+
prompts.append((eid, next(iter(texts))))
|
|
327
|
+
for eid, text in prompts:
|
|
328
|
+
if eid not in want:
|
|
329
|
+
continue
|
|
330
|
+
value = Context(text)
|
|
331
|
+
# Reply order is unknown; conflicting base prompt interpretations are omitted too.
|
|
332
|
+
if eid in out and out[eid] != value:
|
|
333
|
+
ambiguous.add(eid)
|
|
334
|
+
out[eid] = value
|
|
335
|
+
except (sqlite3.DatabaseError, TypeError, ValueError, IndexError) as exc:
|
|
336
|
+
raise CollectorUnavailable("cursor_cli: invalid context store") from exc
|
|
337
|
+
finally:
|
|
338
|
+
con.close()
|
|
339
|
+
return {eid: value for eid, value in out.items() if eid not in ambiguous}
|
|
340
|
+
|
|
341
|
+
|
|
342
|
+
def t3_context(home: Path, sessions: set[str], want: set[str],
|
|
343
|
+
until: datetime | None) -> dict[str, Context]:
|
|
344
|
+
"""Resolve app message IDs locally; text never enters collector rows or watermarks."""
|
|
345
|
+
out: dict[str, Context] = {}
|
|
346
|
+
threads = tuple(sorted(sid[3:] for sid in sessions if sid.startswith("t3:")))
|
|
347
|
+
if not threads:
|
|
348
|
+
return out
|
|
349
|
+
try:
|
|
350
|
+
with t3.connection(home) as con:
|
|
351
|
+
placeholders = ','.join('?' for _ in threads)
|
|
352
|
+
condition = f'thread_id IN ({placeholders})'
|
|
353
|
+
params: tuple[Any, ...] = threads
|
|
354
|
+
if until is not None:
|
|
355
|
+
condition += ' AND julianday(created_at)<julianday(?)'
|
|
356
|
+
params += (until.isoformat(),)
|
|
357
|
+
rows = t3._read(con, f"SELECT message_id,thread_id,run_id,role,created_at,payload_json "
|
|
358
|
+
f"FROM {t3._PREFIX}messages WHERE {condition} "
|
|
359
|
+
"ORDER BY created_at,message_id", params)
|
|
360
|
+
by_thread: dict[str, list[tuple[str | None, str, int | None]]] = {}
|
|
361
|
+
counts: Counter[str] = Counter()
|
|
362
|
+
for mid, tid, runid, role, ts, raw in rows:
|
|
363
|
+
if "t3:" + tid not in sessions or not _in_time({"timestamp": ts}, until):
|
|
364
|
+
continue
|
|
365
|
+
payload = t3._payload(raw)
|
|
366
|
+
text_value = payload.get("text")
|
|
367
|
+
text = text_value if isinstance(text_value, str) else ""
|
|
368
|
+
if role == "user":
|
|
369
|
+
if not t3.confirmed_human(payload):
|
|
370
|
+
# An automated message ends any prior human reply interval.
|
|
371
|
+
by_thread.setdefault(tid, []).append(("", "", None))
|
|
372
|
+
continue
|
|
373
|
+
counts[tid] += 1
|
|
374
|
+
by_thread.setdefault(tid, []).append(
|
|
375
|
+
(make_event_id("t3", "steer", mid), text, counts[tid]))
|
|
376
|
+
elif role == "assistant":
|
|
377
|
+
by_thread.setdefault(tid, []).append((None, text, None))
|
|
378
|
+
for events in by_thread.values():
|
|
379
|
+
out.update(_join(events, want))
|
|
380
|
+
except (OSError, sqlite3.Error, ValueError, TypeError) as exc:
|
|
381
|
+
raise CollectorUnavailable("t3: local context unavailable") from exc
|
|
382
|
+
return out
|