@lmmslab/harness 0.83.0 → 0.85.0
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.
- package/bin/harness.js +6 -1
- package/package.json +1 -1
- package/python/harness/__init__.py +1 -1
- package/python/harness/builtin/prompts/core/core.md +68 -38
- package/python/harness/builtin/prompts/shared/learning-goal.md +1 -1
- package/python/harness/cli.py +140 -18
- package/python/harness/config/agent.py +3 -0
- package/python/harness/core/trace.py +23 -3
- package/python/harness/core/worker.py +6 -0
- package/python/harness/host/client.py +26 -21
- package/python/harness/host/http.py +11 -1
- package/python/harness/host/sessions.py +38 -3
- package/python/harness/host/tasks.py +71 -27
- package/python/harness/instructions.py +329 -0
- package/python/harness/model_list.py +25 -5
- package/python/harness/packages/environments.py +51 -12
- package/python/harness/packages/manager.py +379 -102
- package/python/harness/packages/resolve.py +52 -15
- package/python/harness/packages/sources.py +17 -5
- package/python/harness/packages/state.py +10 -0
- package/python/harness/providers.py +66 -24
- package/python/harness/runtime/decision/model.py +9 -4
- package/python/harness/runtime/events.py +1 -0
- package/python/harness/runtime/execution/entries.py +53 -7
- package/python/harness/runtime/execution/session.py +57 -10
- package/python/harness/runtime/learning/active.py +4 -4
- package/python/harness/runtime/reactive/tree.py +96 -53
- package/python/harness/runtime/task/environments.py +3 -1
- package/python/harness/runtime/task/run.py +93 -4
- package/python/harness/runtime/tick.py +34 -3
- package/python/harness/settings.py +23 -1
- package/python/harness/terminal.py +3 -0
- package/python/harness/transcript.py +12 -0
- package/python/harness/tui/__init__.py +28 -6
- package/python/harness/tui/chat.py +323 -80
- package/python/harness/tui/pty.py +16 -0
- package/python/harness/tui/renderer.py +1 -0
- package/python/harness/tui/views.py +160 -38
|
@@ -8,6 +8,7 @@ tunnel to the server's port. It is also how every frontend drives a session of t
|
|
|
8
8
|
session host (host/): `run` is then the session's id, or "<session>/<agent>".
|
|
9
9
|
"""
|
|
10
10
|
|
|
11
|
+
from http.client import IncompleteRead
|
|
11
12
|
import json
|
|
12
13
|
import time
|
|
13
14
|
from urllib.error import HTTPError
|
|
@@ -112,6 +113,10 @@ class RemoteConversation:
|
|
|
112
113
|
"""Send terminal input to a session started with an interactive PTY."""
|
|
113
114
|
self._post("terminal", text=text)
|
|
114
115
|
|
|
116
|
+
def terminal_resize(self, columns: int, rows: int) -> None:
|
|
117
|
+
"""Match a hosted PTY to the attached terminal's character dimensions."""
|
|
118
|
+
self._post("terminal/resize", columns=columns, rows=rows)
|
|
119
|
+
|
|
115
120
|
def handoff(self, agent: str, context: str, *, folder=None, isolation=None):
|
|
116
121
|
"""Start another agent with the context chosen by the user."""
|
|
117
122
|
answer = self._post("handoff", agent=agent, context=context, folder=str(folder) if folder else None,
|
|
@@ -154,22 +159,24 @@ class RemoteConversation:
|
|
|
154
159
|
turn's end is its turn_result record, after which the conversation waits). With
|
|
155
160
|
`stop` (a threading.Event), it ends once the event is set, within the server's
|
|
156
161
|
keep-alive interval. When the server stops answering, it tries again, waiting longer
|
|
157
|
-
each time, and raises HostGone after GONE_SECONDS without
|
|
162
|
+
each time, and raises HostGone after GONE_SECONDS of failures without records or keep-alives."""
|
|
158
163
|
failing, delay = None, RETRY_SECONDS
|
|
159
164
|
while True:
|
|
160
165
|
if stop is not None and stop.is_set():
|
|
161
166
|
return
|
|
162
|
-
|
|
167
|
+
reason = "the records stream ended before its end event"
|
|
163
168
|
try:
|
|
164
169
|
with open_url(f"{self.url}/api/stream?run={quote(self.run)}", timeout=self.timeout,
|
|
165
170
|
headers={"X-Session-API-Key": self.token, "Last-Event-ID": str(after)}) as response:
|
|
166
|
-
failing, delay = None, RETRY_SECONDS
|
|
167
171
|
for event, data in read_sse(response):
|
|
168
172
|
if stop is not None and stop.is_set():
|
|
169
173
|
return
|
|
174
|
+
# Headers alone do not prove a working stream. Records and
|
|
175
|
+
# keep-alive comments do, including on an otherwise quiet turn.
|
|
176
|
+
if event in (None, "record", "end"):
|
|
177
|
+
failing, delay = None, RETRY_SECONDS
|
|
170
178
|
if event == "end":
|
|
171
|
-
|
|
172
|
-
break
|
|
179
|
+
return
|
|
173
180
|
if event == "record":
|
|
174
181
|
record = json.loads(data)
|
|
175
182
|
after = record["seq"]
|
|
@@ -178,23 +185,21 @@ class RemoteConversation:
|
|
|
178
185
|
yield record
|
|
179
186
|
except HTTPError as error:
|
|
180
187
|
raise RemoteError(_reason(error)) from None
|
|
181
|
-
except
|
|
182
|
-
|
|
183
|
-
|
|
184
|
-
now = time.monotonic()
|
|
185
|
-
failing = failing or now
|
|
186
|
-
if now - failing >= GONE_SECONDS:
|
|
187
|
-
raise HostGone(f"the session host at {self.url} does not answer "
|
|
188
|
-
f"({getattr(error, 'reason', None) or error})") from None
|
|
189
|
-
if stop is not None:
|
|
190
|
-
if stop.wait(delay):
|
|
191
|
-
return
|
|
192
|
-
else:
|
|
193
|
-
time.sleep(delay)
|
|
194
|
-
delay = min(delay * 2, RETRY_LIMIT)
|
|
195
|
-
continue
|
|
196
|
-
if ended:
|
|
188
|
+
except (OSError, IncompleteRead) as error: # Timeouts, truncated bodies and connection failures.
|
|
189
|
+
reason = getattr(error, "reason", None) or error
|
|
190
|
+
if stop is not None and stop.is_set():
|
|
197
191
|
return
|
|
192
|
+
now = time.monotonic()
|
|
193
|
+
if failing is None:
|
|
194
|
+
failing = now
|
|
195
|
+
if now - failing >= GONE_SECONDS:
|
|
196
|
+
raise HostGone(f"the session host at {self.url} does not answer ({reason})") from None
|
|
197
|
+
if stop is not None:
|
|
198
|
+
if stop.wait(delay):
|
|
199
|
+
return
|
|
200
|
+
else:
|
|
201
|
+
time.sleep(delay)
|
|
202
|
+
delay = min(delay * 2, RETRY_LIMIT)
|
|
198
203
|
|
|
199
204
|
|
|
200
205
|
def _reason(error: HTTPError) -> str:
|
|
@@ -480,6 +480,7 @@ class Handler(BaseHTTPRequestHandler):
|
|
|
480
480
|
|
|
481
481
|
def post_conversations_message(self, body):
|
|
482
482
|
self._conversation(body).send_message(str(body.get("text") or ""), files=body.get("files") or ())
|
|
483
|
+
self.server.conversations.interact(self.server.conversations.resolve(body.get("run")))
|
|
483
484
|
self._json({"ok": True})
|
|
484
485
|
|
|
485
486
|
def _control(self, body, capability, method, *args):
|
|
@@ -490,6 +491,7 @@ class Handler(BaseHTTPRequestHandler):
|
|
|
490
491
|
|
|
491
492
|
def post_conversations_steer(self, body):
|
|
492
493
|
self._control(body, "steer", "steer", str(body.get("text") or ""), body.get("files") or ())
|
|
494
|
+
self.server.conversations.interact(self.server.conversations.resolve(body.get("run")))
|
|
493
495
|
self._json({"ok": True})
|
|
494
496
|
|
|
495
497
|
def post_conversations_cancel(self, body):
|
|
@@ -516,12 +518,17 @@ class Handler(BaseHTTPRequestHandler):
|
|
|
516
518
|
self._control(body, "terminal_input", "terminal_input", body.get("text"))
|
|
517
519
|
self._json({"ok": True})
|
|
518
520
|
|
|
521
|
+
def post_conversations_terminal_resize(self, body):
|
|
522
|
+
self._control(body, "terminal_resize", "terminal_resize", body.get("columns"), body.get("rows"))
|
|
523
|
+
self._json({"ok": True})
|
|
524
|
+
|
|
519
525
|
def post_conversations_pause(self, body):
|
|
520
526
|
self._conversation(body).pause()
|
|
521
527
|
self._json({"ok": True})
|
|
522
528
|
|
|
523
529
|
def post_conversations_resume(self, body):
|
|
524
530
|
self._conversation(body).resume()
|
|
531
|
+
self.server.conversations.interact(self.server.conversations.resolve(body.get("run")))
|
|
525
532
|
self._json({"ok": True})
|
|
526
533
|
|
|
527
534
|
def post_conversations_confirm(self, body):
|
|
@@ -584,11 +591,14 @@ class Handler(BaseHTTPRequestHandler):
|
|
|
584
591
|
row = self._session((query.get("id") or [""])[0])
|
|
585
592
|
lines = int((query.get("lines") or ["40"])[0])
|
|
586
593
|
for agent in row["agents"]:
|
|
594
|
+
agent["lines"] = []
|
|
595
|
+
if lines <= 0:
|
|
596
|
+
continue
|
|
587
597
|
directory = Path(agent["directory"]) if agent["directory"] else None
|
|
588
598
|
records = runs.read_records(directory) if directory and (directory / "events.jsonl").is_file() else []
|
|
589
599
|
printer = TextTranscript()
|
|
590
600
|
text = [line for record in records for line in printer.lines(record)]
|
|
591
|
-
agent["lines"] = text[-lines:]
|
|
601
|
+
agent["lines"] = text[-lines:]
|
|
592
602
|
self._json(row)
|
|
593
603
|
|
|
594
604
|
def post_sessions_stop(self, body):
|
|
@@ -17,6 +17,7 @@ listed after the host exits (history()).
|
|
|
17
17
|
import json
|
|
18
18
|
from pathlib import Path
|
|
19
19
|
import secrets
|
|
20
|
+
import sys
|
|
20
21
|
import threading
|
|
21
22
|
import time
|
|
22
23
|
|
|
@@ -54,7 +55,10 @@ def waiting(status, note=None) -> str | None:
|
|
|
54
55
|
|
|
55
56
|
def ended_status(directory: Path) -> str:
|
|
56
57
|
"""The status of an agent nothing runs any more, from how its episode ended (none: interrupted)."""
|
|
57
|
-
|
|
58
|
+
summary = episodes.summary(directory)
|
|
59
|
+
if summary.get("open"):
|
|
60
|
+
return "INTERRUPTED"
|
|
61
|
+
return {"error": "ERROR", "interrupted": "INTERRUPTED", None: "INTERRUPTED"}.get(episodes.ending(summary), "FINISHED")
|
|
58
62
|
|
|
59
63
|
|
|
60
64
|
def _usage(conversation) -> dict:
|
|
@@ -91,6 +95,8 @@ class Sessions(dict):
|
|
|
91
95
|
# Run -> (session id, agent id).
|
|
92
96
|
self._runs = {}
|
|
93
97
|
self._lock = threading.RLock()
|
|
98
|
+
# Last session whose user selection was persisted; repeated messages need no new row.
|
|
99
|
+
self._last_interacted = None
|
|
94
100
|
|
|
95
101
|
# The dict, as servers use it.
|
|
96
102
|
|
|
@@ -127,7 +133,8 @@ class Sessions(dict):
|
|
|
127
133
|
now = created if created is not None else time.time()
|
|
128
134
|
if new:
|
|
129
135
|
entry = {"id": session, "title": title or _title(conversation),
|
|
130
|
-
"frontend": frontend or self.default_frontend, "created": now,
|
|
136
|
+
"frontend": frontend or self.default_frontend, "created": now,
|
|
137
|
+
"last_interaction": time.time(), "agents": {}}
|
|
131
138
|
self._sessions[session] = entry
|
|
132
139
|
agent, parent = MAIN, None
|
|
133
140
|
else:
|
|
@@ -141,6 +148,8 @@ class Sessions(dict):
|
|
|
141
148
|
except AttributeError:
|
|
142
149
|
pass
|
|
143
150
|
self.save_metadata(run)
|
|
151
|
+
if new:
|
|
152
|
+
self._last_interacted = session
|
|
144
153
|
self.registered(run, conversation)
|
|
145
154
|
return session
|
|
146
155
|
|
|
@@ -232,6 +241,27 @@ class Sessions(dict):
|
|
|
232
241
|
"""(session id, agent id) of a run."""
|
|
233
242
|
return self._runs.get(run)
|
|
234
243
|
|
|
244
|
+
def interact(self, run: str) -> None:
|
|
245
|
+
"""Keep a successful user message or resume, independently of background events."""
|
|
246
|
+
with self._lock:
|
|
247
|
+
found = self._runs.get(run)
|
|
248
|
+
if found is None:
|
|
249
|
+
return
|
|
250
|
+
session, _ = found
|
|
251
|
+
self._sessions[session]["last_interaction"] = time.time()
|
|
252
|
+
if self._last_interacted == session:
|
|
253
|
+
return
|
|
254
|
+
try:
|
|
255
|
+
self.save_metadata(run)
|
|
256
|
+
except OSError as error:
|
|
257
|
+
# The message/resume already succeeded. Leave the selection dirty for retry.
|
|
258
|
+
try:
|
|
259
|
+
print(f"harness: could not save session activity: {error}", file=sys.stderr, flush=True)
|
|
260
|
+
except OSError:
|
|
261
|
+
pass
|
|
262
|
+
else:
|
|
263
|
+
self._last_interacted = session
|
|
264
|
+
|
|
235
265
|
def save_metadata(self, run: str) -> None:
|
|
236
266
|
"""Persist the public identity once a provider has returned its external ID."""
|
|
237
267
|
with self._lock:
|
|
@@ -242,6 +272,7 @@ class Sessions(dict):
|
|
|
242
272
|
self._write({"id": session, "agent": agent, "parent": item["parent"], "run": run,
|
|
243
273
|
"directory": str(Path(conversation.directory).resolve()), "title": item["title"],
|
|
244
274
|
"frontend": entry["frontend"], "created": item["created"],
|
|
275
|
+
"last_interaction": entry.get("last_interaction") or entry["created"],
|
|
245
276
|
"metadata": session_metadata(conversation)})
|
|
246
277
|
|
|
247
278
|
# Describing.
|
|
@@ -268,7 +299,8 @@ class Sessions(dict):
|
|
|
268
299
|
"model": main.get("model"), "reasoning_effort": main.get("reasoning_effort"),
|
|
269
300
|
"workspace": getattr(conversation, "workspace_kind", None),
|
|
270
301
|
"directory": str(Path(conversation.directory).resolve()) if conversation is not None else None,
|
|
271
|
-
"created": item["created"],
|
|
302
|
+
"created": item["created"],
|
|
303
|
+
"usage": _usage(conversation) if conversation is not None else None}
|
|
272
304
|
|
|
273
305
|
def describe(self, session: str) -> dict | None:
|
|
274
306
|
"""One session as it is listed: its id, title, the frontend that started it, its
|
|
@@ -325,6 +357,7 @@ def _session_row(entry: dict, agents: dict) -> dict:
|
|
|
325
357
|
"cost_usd": round(sum(item["cost_usd"] for item in usage), 6)} if usage else None)
|
|
326
358
|
return {"id": entry["id"], "title": entry["title"], "frontend": entry["frontend"], "created": entry["created"],
|
|
327
359
|
"run": main.get("run", entry["id"]), "directory": main.get("directory"), "status": status,
|
|
360
|
+
"last_interaction": entry.get("last_interaction") or entry["created"],
|
|
328
361
|
"live": any(live(agent["status"]) for agent in tree),
|
|
329
362
|
"waiting": ("approval" if approvals else main.get("waiting")), "approvals": approvals, "usage": total,
|
|
330
363
|
"kind": main.get("metadata", {}).get("kind", "harness"),
|
|
@@ -377,6 +410,8 @@ def history(index: Path, skip=()) -> list[dict]:
|
|
|
377
410
|
entry = entries.setdefault(item["id"], {"id": item["id"], "title": item.get("title") or "",
|
|
378
411
|
"frontend": item.get("frontend") or "", "created": item.get("created"),
|
|
379
412
|
"agents": {}})
|
|
413
|
+
if item.get("last_interaction") is not None:
|
|
414
|
+
entry["last_interaction"] = max(entry.get("last_interaction") or 0, item["last_interaction"])
|
|
380
415
|
if item.get("agent", MAIN) == MAIN:
|
|
381
416
|
entry.update(title=item.get("title") or "", frontend=item.get("frontend") or "", created=item.get("created"))
|
|
382
417
|
entry["agents"][item.get("agent", MAIN)] = item
|
|
@@ -37,7 +37,8 @@ CONVERSATION = ROOT / "configs" / "assistant.toml"
|
|
|
37
37
|
# --window-details, --think; and `learn`, a conversation that learns its environment for live
|
|
38
38
|
# minutes (true: learning.active_minutes) instead of doing a task (runtime/learning/active.py).
|
|
39
39
|
# `agents`: sub-agents on in this conversation whatever session.agents says.
|
|
40
|
-
RUN_KEYS = ("physical", "privacy", "think_while_acting", "think_interval", "learn", "managed_workspace", "agents"
|
|
40
|
+
RUN_KEYS = ("physical", "privacy", "think_while_acting", "think_interval", "learn", "managed_workspace", "agents",
|
|
41
|
+
"instructions")
|
|
41
42
|
# The keys of a conversation profile's file (profiles()).
|
|
42
43
|
PROFILE_KEYS = {
|
|
43
44
|
"description": "one line: what conversations of this profile are for",
|
|
@@ -82,10 +83,20 @@ class TaskHost:
|
|
|
82
83
|
|
|
83
84
|
def llm(self) -> dict:
|
|
84
85
|
"""The model the user chose, and whether its provider's key is there."""
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
|
|
88
|
-
|
|
86
|
+
from .. import model_list
|
|
87
|
+
|
|
88
|
+
try:
|
|
89
|
+
chosen = model_list.select(self.directory, native=True)
|
|
90
|
+
except model_list.SelectionError as error:
|
|
91
|
+
return {"provider": None, "model": None, "base_url": None, "key_env": None,
|
|
92
|
+
"key": {"set": False, "source": None}, "ready": False,
|
|
93
|
+
"detail": str(error), "selection_error": True}
|
|
94
|
+
settings = chosen if chosen is not None else providers.load(self.directory)
|
|
95
|
+
provider = providers.PROVIDERS.get(settings["provider"])
|
|
96
|
+
key_env = provider.key_env if provider else None
|
|
97
|
+
key = self.credentials.status(key_env) if key_env else {"set": True, "source": None}
|
|
98
|
+
return {**settings, "key_env": key_env, "key": key,
|
|
99
|
+
"ready": chosen is not None and (key["set"] or not providers.requires_key(settings["provider"]))}
|
|
89
100
|
|
|
90
101
|
def status(self) -> dict:
|
|
91
102
|
workspaces = self.available_workspaces()
|
|
@@ -252,7 +263,7 @@ class TaskHost:
|
|
|
252
263
|
settings = providers.check(body)
|
|
253
264
|
provider = providers.PROVIDERS[settings["provider"]]
|
|
254
265
|
key = str(body.get("key") or "").strip() or (os.environ.get(provider.key_env) if provider.key_env else None)
|
|
255
|
-
if provider
|
|
266
|
+
if providers.requires_key(settings["provider"]) and not key:
|
|
256
267
|
return {"ok": False, "detail": "no key"}
|
|
257
268
|
result = providers.reach(settings, key)
|
|
258
269
|
if result["ok"]:
|
|
@@ -280,14 +291,22 @@ class TaskHost:
|
|
|
280
291
|
kind = self.kind(kind)
|
|
281
292
|
from .. import model_list
|
|
282
293
|
|
|
283
|
-
|
|
294
|
+
try:
|
|
295
|
+
current = providers.load(self.directory)
|
|
296
|
+
except model_list.SelectionError:
|
|
297
|
+
if choice is None:
|
|
298
|
+
raise
|
|
299
|
+
current = {}
|
|
284
300
|
advanced = values["model"]
|
|
285
301
|
if choice is not None:
|
|
286
302
|
advanced = {**advanced, "reasoning_effort": effort}
|
|
287
|
-
|
|
288
|
-
|
|
289
|
-
|
|
290
|
-
choice
|
|
303
|
+
choice = choice if choice is not None else model_list.select(self.directory, native=True)
|
|
304
|
+
if choice is None:
|
|
305
|
+
raise ValueError("add an enabled Harness model in Settings")
|
|
306
|
+
if any(choice.get(key) != current.get(key) for key in ("provider", "base_url")):
|
|
307
|
+
advanced = dict(advanced)
|
|
308
|
+
advanced.pop("checker", None)
|
|
309
|
+
advanced.pop("vision", None)
|
|
291
310
|
if model:
|
|
292
311
|
choice = {**choice, "model": str(model)}
|
|
293
312
|
# A model the list says reads no images is sent none (its role's `images`).
|
|
@@ -316,6 +335,10 @@ class TaskHost:
|
|
|
316
335
|
"agent_skills": addons.agent_skill_options(self.packages, skills["agent"])}
|
|
317
336
|
# No implicit cross-task knowledge. A caller can explicitly bind a learning library.
|
|
318
337
|
embodiment_config["runtime"] = {**embodiment_config.get("runtime", {}), "library": None}
|
|
338
|
+
# The user's and the project's instruction files (instructions.py): a conversation a user
|
|
339
|
+
# starts reads them; an experiment or a benchmark, whose config has none of this, does not.
|
|
340
|
+
embodiment_config["runtime"] = {**embodiment_config.get("runtime", {}),
|
|
341
|
+
"instructions": dict(values["instructions"])}
|
|
319
342
|
servers = self.mcp_servers()
|
|
320
343
|
if servers:
|
|
321
344
|
# The MCP servers a package connects (the app's connectors, through its gateway).
|
|
@@ -403,6 +426,18 @@ class TaskHost:
|
|
|
403
426
|
enable.append("agents")
|
|
404
427
|
return {"enable": enable, "installed": installed, "agent": agent}
|
|
405
428
|
|
|
429
|
+
def _package_context(self):
|
|
430
|
+
"""Optional workspace context for package management, which also works before setup."""
|
|
431
|
+
from ..bench import bench_spec
|
|
432
|
+
|
|
433
|
+
available = self.workspaces.providers()
|
|
434
|
+
matches = [provider for provider in available if provider.accepts({})]
|
|
435
|
+
if len(matches) != 1 and len(available) != 1:
|
|
436
|
+
# Installing the first environment or managing several needs no default.
|
|
437
|
+
return None, None
|
|
438
|
+
kind = self.kind()
|
|
439
|
+
return kind, bench_spec(self.workspaces.bench(kind)).kind
|
|
440
|
+
|
|
406
441
|
def skills(self) -> dict:
|
|
407
442
|
"""The skills as Settings lists them: the built-in ones of this computer and the
|
|
408
443
|
installed ones, each on or off; why an installed one cannot load."""
|
|
@@ -410,18 +445,19 @@ class TaskHost:
|
|
|
410
445
|
|
|
411
446
|
off = set((self.settings.get().get("skills") or {}).get("off", []))
|
|
412
447
|
loaded = load_skills()
|
|
413
|
-
|
|
414
|
-
|
|
415
|
-
|
|
416
|
-
|
|
417
|
-
|
|
418
|
-
|
|
419
|
-
for name in
|
|
420
|
-
if name != embodiment:
|
|
448
|
+
kind, embodiment = self._package_context()
|
|
449
|
+
rows = ([{"name": embodiment, "format": "built-in", "summary": loaded[embodiment].summary,
|
|
450
|
+
"locked": True, "on": True}] if embodiment in loaded else [])
|
|
451
|
+
rows.append({"name": "checklist", "format": "built-in", "summary": loaded["checklist"].summary,
|
|
452
|
+
"on": not {"checklist", "plan"} & off})
|
|
453
|
+
builtin = self.workspaces.provider(kind).builtin_skills(kind) if kind is not None else ()
|
|
454
|
+
for name in builtin:
|
|
455
|
+
if name != embodiment and name in loaded:
|
|
421
456
|
rows.append({"name": name, "format": "built-in", "summary": loaded[name].summary,
|
|
422
457
|
"locked": True, "on": True})
|
|
423
458
|
for row in addons.skills(self.packages):
|
|
424
|
-
if row["format"] == "harness" and "embodiments" in row
|
|
459
|
+
if (embodiment is not None and row["format"] == "harness" and "embodiments" in row
|
|
460
|
+
and embodiment not in row["embodiments"]):
|
|
425
461
|
continue
|
|
426
462
|
problem = row.get("problem") or USER_PROBLEMS.get(row["path"]) if row["format"] == "harness" else \
|
|
427
463
|
row.get("problem")
|
|
@@ -451,7 +487,6 @@ class TaskHost:
|
|
|
451
487
|
installed only with `trust`, after the page showed what it brings. A folder of
|
|
452
488
|
skills becomes a package per skill, taken only when it runs on this computer."""
|
|
453
489
|
import base64
|
|
454
|
-
from ..bench import bench_spec
|
|
455
490
|
|
|
456
491
|
zip_bytes = base64.b64decode(body["zip"]) if body.get("zip") else None
|
|
457
492
|
if not zip_bytes and not body.get("source"):
|
|
@@ -459,11 +494,13 @@ class TaskHost:
|
|
|
459
494
|
|
|
460
495
|
def kernel_python():
|
|
461
496
|
# A skill's requirements are for the kernel: the environment of the computer's mode.
|
|
462
|
-
|
|
497
|
+
kind = self.kind()
|
|
498
|
+
return self.workspaces.provider(kind).python(kind, log=lambda message: None)
|
|
463
499
|
|
|
500
|
+
_, embodiment = self._package_context()
|
|
464
501
|
result = self.packages.install([body["source"]] if body.get("source") else [], zip_bytes=zip_bytes,
|
|
465
502
|
consent=lambda found: bool(body.get("trust")), python=kernel_python,
|
|
466
|
-
embodiment=
|
|
503
|
+
embodiment=embodiment,
|
|
467
504
|
log=lambda message: None)
|
|
468
505
|
if result.get("found"):
|
|
469
506
|
result = {**result, "found": [{**found, "shown": describe(found)} for found in result["found"]]}
|
|
@@ -609,10 +646,12 @@ class TaskHost:
|
|
|
609
646
|
provider = providers.PROVIDERS[choice["provider"]]
|
|
610
647
|
if provider.key_env:
|
|
611
648
|
self.credentials.load()
|
|
612
|
-
if not os.environ.get(provider.key_env):
|
|
649
|
+
if providers.requires_key(choice["provider"]) and not os.environ.get(provider.key_env):
|
|
613
650
|
raise ValueError(f"configure credentials for {provider.name} in Settings")
|
|
614
|
-
|
|
615
|
-
|
|
651
|
+
else:
|
|
652
|
+
state = self.llm()
|
|
653
|
+
if not state["ready"]:
|
|
654
|
+
raise ValueError(state.get("detail") or "no model key yet")
|
|
616
655
|
kind, folder = self.environment(workspace, folder, last=frontend == "taptab", model_entry=model_entry)
|
|
617
656
|
request = {"folder": folder} if folder is not None else {}
|
|
618
657
|
self.workspaces.require_choices(kind, runtime)
|
|
@@ -640,6 +679,9 @@ class TaskHost:
|
|
|
640
679
|
if unknown:
|
|
641
680
|
raise ValueError(f"a conversation's own runtime keys are {', '.join(RUN_KEYS)}, not {sorted(unknown)}")
|
|
642
681
|
runtime = dict(runtime or {})
|
|
682
|
+
if "instructions" in runtime:
|
|
683
|
+
# harness run --bare: {"load": "off"} over the settings' instructions.
|
|
684
|
+
runtime["instructions"] = {**(config.runtime.get("instructions") or {}), **dict(runtime["instructions"])}
|
|
643
685
|
learn = runtime.pop("learn", None)
|
|
644
686
|
runtime.pop("agents", None)
|
|
645
687
|
managed = runtime.pop("managed_workspace", False)
|
|
@@ -793,7 +835,8 @@ class TaskHost:
|
|
|
793
835
|
rows = [row for row in active + (history(index, skip={row["id"] for row in active}) if index else [])
|
|
794
836
|
if row.get("kind", "harness") == "harness" and row.get("directory")]
|
|
795
837
|
if identifier is None:
|
|
796
|
-
found = sorted(rows, key=lambda row: row["created"] or 0
|
|
838
|
+
found = sorted(rows, key=lambda row: (row.get("last_interaction") or row["created"] or 0,
|
|
839
|
+
row["created"] or 0))[-1:]
|
|
797
840
|
else:
|
|
798
841
|
name = str(identifier)
|
|
799
842
|
found = ([row for row in rows if name in (row["id"], row["run"])]
|
|
@@ -803,6 +846,7 @@ class TaskHost:
|
|
|
803
846
|
f"no session {identifier} to continue (harness sessions ls --all lists them)")
|
|
804
847
|
row = found[0]
|
|
805
848
|
if row["live"]:
|
|
849
|
+
server.conversations.interact(row["run"])
|
|
806
850
|
return row["run"]
|
|
807
851
|
directory = Path(row["directory"])
|
|
808
852
|
kept = saved(directory)
|