@assuredloop/orca-bot-kit 0.36.0 → 0.36.1

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@assuredloop/orca-bot-kit",
3
- "version": "0.36.0",
3
+ "version": "0.36.1",
4
4
  "license": "MIT",
5
5
  "repository": {
6
6
  "type": "git",
@@ -127,6 +127,22 @@ the bots folder moves. `obk session add` and `obk session change` say so when a
127
127
  work dir leads out of the bot home; take that back to the user rather than
128
128
  past them.
129
129
 
130
+ ## How its sessions choose a subagent's model
131
+
132
+ A session runs on one model of its own, but the model of each subagent it starts
133
+ can be chosen task by task. A bot that should choose carries its way of working
134
+ for this as a rule unit of its own. The bot can decide that way of working, or
135
+ the user can. Write it in their words as a unit in `rules/` of the bots folder,
136
+ name it in that bot's `rules:` list, and run `obk rules build`: every session
137
+ of the bot then reads it in its `AGENTS.md`. Another bot can carry another
138
+ rule, or none. A bot with none works as before, and each subagent runs on what
139
+ the harness gives it.
140
+
141
+ The rule acts where the harness lets a session choose a model for a subagent,
142
+ as Claude Code's Agent tool does with its `model`. Where the harness does not,
143
+ nothing changes. The kit names no model and no table from a task to a model:
144
+ any model in the rule is the bot's or the user's choice.
145
+
130
146
  ## Offer a line-up rather than a blank page
131
147
 
132
148
  Someone making their first bots does not know what a good set looks like, so
@@ -119,6 +119,9 @@ A usual shape: high effort for the session that orchestrates, medium for the
119
119
  ones doing the work, and the top of the range only for deep read-only
120
120
  analysis.
121
121
 
122
+ A bot can also carry its own rule on the model each subagent gets: see "How
123
+ its sessions choose a subagent's model" in the obk-bot-building skill.
124
+
122
125
  ## When a model is not up to the work
123
126
 
124
127
  The signs are in the work, not in the answers. The same instruction given
package/src/cli.js CHANGED
@@ -1177,6 +1177,10 @@ const commands = {
1177
1177
  answer,
1178
1178
  lines: [
1179
1179
  `${'sent'.padEnd(9)} ${where.padEnd(24)} ${answer.subject}`,
1180
+ // Orca took it, and its answer named no letter (#730).
1181
+ ...(answer.id === undefined
1182
+ ? [' Orca queued it, and its answer gave no message id for it.']
1183
+ : []),
1180
1184
  ...(answer.file === undefined
1181
1185
  ? []
1182
1186
  : [` it was too long to travel as itself, so it went as a file: ${answer.file}`]),
package/src/message.js CHANGED
@@ -203,19 +203,21 @@ export async function sendMessage(bots, { to: target, from: sender, tab, subject
203
203
  const shown = await loaded(to);
204
204
  // The Escape goes first, so the turn has ended when the letter is in.
205
205
  const interrupted = interrupt ? interruptBusy(to) : {};
206
- const message = postMessage({
206
+ const posted = postMessage({
207
207
  to: road.address,
208
208
  from: `run:${from.mailbox}`,
209
209
  subject,
210
210
  body: written.body,
211
211
  thread,
212
212
  });
213
+ // Orca took it: sent, whatever else its answer says or leaves out (#730).
214
+ const message = posted?.message;
213
215
 
214
216
  return {
215
217
  ...answer,
216
218
  sent: true,
217
- id: message.id,
218
- thread: message.thread_id ?? thread,
219
+ id: message?.id ?? posted?.messageId,
220
+ thread: message?.thread_id ?? thread,
219
221
  file: written.file,
220
222
  ...interrupted,
221
223
  ...shown,
@@ -356,12 +358,14 @@ function mailOf(bots, who, peek, until) {
356
358
  trouble: `${who.bot}/${who.session} is not up, and its mailbox is bound to no tab, so there is nothing to read it as. Its mail waits until its tab is back: \`up\` brings it back, or \`unpause\` when it is paused.`,
357
359
  };
358
360
  }
359
- // Orca hands mail over a batch at a time and replays the oldest until it is
360
- // acknowledged; the acknowledgement answers with the next batch. So a read
361
- // takes every batch in turn, or newer mail waits behind an older one (#299).
362
- //
363
- // A batch is given back once it is acknowledged, and not before: one this
364
- // check stops short of acknowledging is handed over again to the next.
361
+ // Orca hands mail over a batch at a time, so a read takes every batch in
362
+ // turn, or newer mail waits behind an older one (#299). Up to 1.4.245 it
363
+ // replays the oldest until it is acknowledged, and the acknowledgement
364
+ // answers with the next batch: a batch this check stops short of
365
+ // acknowledging is handed over again to the next. From 1.4.246 a read
366
+ // acknowledges the batch it hands over and says so (`deliveryAcknowledged`),
367
+ // and an acknowledgement answers with no batch, so the next is read again
368
+ // (#729; read in the 1.4.246 and 1.4.255 source).
365
369
  const messages = [];
366
370
  const read = { bots, bot: who.bot, session: who.session, mailbox: who.mailbox, read: !peek, messages };
367
371
  let found = readMailbox(who.mailbox, { peek, handle }, timeLeft(until));
@@ -372,6 +376,24 @@ function mailOf(bots, who, peek, until) {
372
376
  return read;
373
377
  }
374
378
  const wait = timeLeft(until, false);
379
+ if (found.deliveryAcknowledged === true) {
380
+ // Read already, so shown whatever comes after.
381
+ messages.push(...batch);
382
+ if (wait === undefined) {
383
+ return { ...read, trouble: `${who.bot}/${who.session} has more mail than one check reads in ${HOLD_MS / 1000} seconds, so it stopped there. The rest is still waiting: check again.` };
384
+ }
385
+ try {
386
+ found = readMailbox(who.mailbox, { handle }, wait);
387
+ } catch (error) {
388
+ // A read Orca did not answer about may have taken the next batch all
389
+ // the same, and taken is read: that batch is not handed over again.
390
+ if (error.code === TIMED_OUT) {
391
+ return { ...read, trouble: `${error.message} Orca may have taken the next letters as read without the kit seeing them. The mail shown above is read.` };
392
+ }
393
+ return { ...read, trouble: `${error.message} The mail shown above is read, and the rest waits for the next check.` };
394
+ }
395
+ continue;
396
+ }
375
397
  if (wait === undefined) {
376
398
  return { ...read, trouble: `${who.bot}/${who.session} has more mail than one check reads in ${HOLD_MS / 1000} seconds, so it stopped there. The rest is still waiting: check again.` };
377
399
  }
package/src/orca.js CHANGED
@@ -1203,6 +1203,33 @@ export const pressEscape = (handle) =>
1203
1203
  /** Where Orca names the terminal a process runs in, in every pane it opens. */
1204
1204
  export const TERMINAL_ENV = 'ORCA_TERMINAL_HANDLE';
1205
1205
 
1206
+ /**
1207
+ * One `orca orchestration` command, answered in full. From 1.4.246 Orca prints
1208
+ * a short answer for each of these unless it is given `--full`, and the short
1209
+ * one leaves out what the kit reads: a Run's `id` and `coordinator_handle`, a
1210
+ * letter's `from_handle`, `created_at`, `thread_id`, `read` and `delivered_at`,
1211
+ * and a sent letter's `message` (#729, #730; read in the 1.4.246 and 1.4.255
1212
+ * source, and seen live on 1.4.255). Orca 1.4.245 and older do not know the
1213
+ * flag on any of them but `inbox`, and refuse it before they do anything, as
1214
+ * `deleteProject` finds with `--force` (#528). Only that refusal sends the
1215
+ * command again without it: one Orca took, or did not answer, is never sent a
1216
+ * second time. Once refused, the rest of this run sends no `--full`.
1217
+ */
1218
+ function orchestration(args, options) {
1219
+ if (fullKnown) {
1220
+ try {
1221
+ return orca([...args, '--full'], options);
1222
+ } catch (error) {
1223
+ if (error.code !== 'invalid_argument' || !error.reason?.startsWith('Unknown flag --full for command: ')) throw error;
1224
+ fullKnown = false;
1225
+ }
1226
+ }
1227
+ return orca(args, options);
1228
+ }
1229
+
1230
+ /** Whether this Orca may know `--full` on its orchestration commands: false once it has refused it. */
1231
+ let fullKnown = true;
1232
+
1206
1233
  /**
1207
1234
  * A mailbox of one session's own: an Orca Run, which is a name and an inbox and
1208
1235
  * nothing else — it schedules nothing and runs nobody. The objective is what a
@@ -1215,7 +1242,7 @@ export const TERMINAL_ENV = 'ORCA_TERMINAL_HANDLE';
1215
1242
  * tab and names no `--from` (tech notes, section 1; #317).
1216
1243
  */
1217
1244
  export const makeMailbox = (objective, options) =>
1218
- orca(['orchestration', 'run-create', '--objective', `obk ${objective}`], options).run.id;
1245
+ orchestration(['orchestration', 'run-create', '--objective', `obk ${objective}`], options).run.id;
1219
1246
 
1220
1247
  /**
1221
1248
  * Bind a mailbox to the terminal `handle`, which makes it the Run's coordinator
@@ -1226,7 +1253,7 @@ export const makeMailbox = (objective, options) =>
1226
1253
  * tab may name no other (#317).
1227
1254
  */
1228
1255
  export const useMailbox = (id, handle, options) =>
1229
- orca(['orchestration', 'run-use', '--id', id, ...(handle === undefined ? [] : ['--from', handle])], options).run;
1256
+ orchestration(['orchestration', 'run-use', '--id', id, ...(handle === undefined ? [] : ['--from', handle])], options).run;
1230
1257
 
1231
1258
  /**
1232
1259
  * The terminal a mailbox is bound to, or undefined when it has none. A closed
@@ -1234,7 +1261,7 @@ export const useMailbox = (id, handle, options) =>
1234
1261
  * nothing, only `run-use` does (tech notes, section 1).
1235
1262
  */
1236
1263
  export const coordinatorOf = (id, options) =>
1237
- orca(['orchestration', 'run-show', '--id', id], options).run.coordinator_handle ?? undefined;
1264
+ orchestration(['orchestration', 'run-show', '--id', id], options).run.coordinator_handle ?? undefined;
1238
1265
 
1239
1266
  /**
1240
1267
  * Whether this Orca has no Run of that id at all, as when the book was written
@@ -1245,29 +1272,36 @@ export const coordinatorOf = (id, options) =>
1245
1272
  */
1246
1273
  export function runMissing(id, options) {
1247
1274
  try {
1248
- orca(['orchestration', 'run-show', '--id', id], options);
1275
+ orchestration(['orchestration', 'run-show', '--id', id], options);
1249
1276
  return false;
1250
1277
  } catch (error) {
1251
1278
  return error.code === 'run_not_found';
1252
1279
  }
1253
1280
  }
1254
1281
 
1255
- /** Queue one message. `to` and `from` are mailboxes, written `run:<id>`. */
1282
+ /**
1283
+ * Queue one message. `to` and `from` are mailboxes, written `run:<id>`. Orca's
1284
+ * whole answer comes back: the letter it queued is its `message`, where Orca
1285
+ * says so. Once Orca has taken it, an answer the kit cannot read is no reason
1286
+ * to say it was not sent (#730).
1287
+ */
1256
1288
  export function postMessage({ to, from, subject, body, type = 'status', thread }) {
1257
1289
  const args = ['orchestration', 'send', '--to', to, '--subject', subject, '--body', body, '--type', type];
1258
1290
  if (from !== undefined) args.push('--from', from);
1259
1291
  if (thread !== undefined) args.push('--thread-id', thread);
1260
- return orca(args).message;
1292
+ return orchestration(args);
1261
1293
  }
1262
1294
 
1263
1295
  /**
1264
1296
  * What is waiting in a mailbox: the oldest batch that has not been
1265
- * acknowledged, or, peeking, whatever is unread without touching it. Read as
1297
+ * acknowledged, or, peeking, whatever is unread without touching it. From
1298
+ * Orca 1.4.246 a read that is not a peek acknowledges the batch it hands over,
1299
+ * and says so with `deliveryAcknowledged: true` (#729). Read as
1266
1300
  * the terminal `handle` when one is given, which is how a read works from a tab
1267
1301
  * bound to another Run; otherwise as this process's own terminal.
1268
1302
  */
1269
1303
  export const readMailbox = (id, { peek = false, handle } = {}, options) =>
1270
- orca(['orchestration', 'check', '--run', id, ...as(handle), ...(peek ? ['--peek'] : [])], options);
1304
+ orchestration(['orchestration', 'check', '--run', id, ...as(handle), ...(peek ? ['--peek'] : [])], options);
1271
1305
 
1272
1306
  /**
1273
1307
  * The newest `limit` messages in a mailbox, newest first, with every column
@@ -1278,11 +1312,14 @@ export const readMailbox = (id, { peek = false, handle } = {}, options) =>
1278
1312
  * cap in Orca, so the answer is taken whatever its size.
1279
1313
  */
1280
1314
  export const mailIn = (id, limit) =>
1281
- orca(['orchestration', 'inbox', '--terminal', `run:${id}`, '--limit', String(limit)], { whole: true }).messages ?? [];
1315
+ orchestration(['orchestration', 'inbox', '--terminal', `run:${id}`, '--limit', String(limit)], { whole: true }).messages ?? [];
1282
1316
 
1283
- /** Say a batch has been read. Orca answers with the batch after it. */
1317
+ /**
1318
+ * Say a batch has been read. Up to Orca 1.4.245 Orca answers with the batch
1319
+ * after it; from 1.4.246 with no batch, since its read acknowledged it (#729).
1320
+ */
1284
1321
  export const ackMailbox = (id, delivery, handle, options) =>
1285
- orca(['orchestration', 'check', '--run', id, ...as(handle), '--ack', delivery], options);
1322
+ orchestration(['orchestration', 'check', '--run', id, ...as(handle), '--ack', delivery], options);
1286
1323
 
1287
1324
  /** The reader of a `check`: the terminal named, or this process's own. */
1288
1325
  const as = (handle) => (handle === undefined ? [] : ['--terminal', handle]);