@tablation/crew 0.14.1 → 0.15.0

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@tablation/crew",
3
- "version": "0.14.1",
3
+ "version": "0.15.0",
4
4
  "description": "A standing team of headless agents that picks work off a Tablation board, does it on a machine you control, and reports back.",
5
5
  "type": "module",
6
6
  "license": "MIT",
@@ -200,7 +200,8 @@ since your last comment: an answer to a question, new direction, or a
200
200
  rebasing) — another ship may have pushed to the branch since you left it.
201
201
  If it reports your branch as diverged (local commits not on the remote AND
202
202
  the remote has moved), stop: comment on the ticket with what it printed and
203
- set `needs_info`; do not merge or rebase it yourself.
203
+ park it for a person (`needs_info` plus the operator as assignee — see
204
+ "Stopping: say it in data, not prose"); do not merge or rebase it yourself.
204
205
  - A ticket QA has bounced back to you comes in as `in_progress`,
205
206
  reassigned to your row (or unassigned, when your ship holds it), with a comment saying what still fails. Treat
206
207
  that exactly like new direction from a hold: read the comment, fix what
@@ -228,13 +229,16 @@ before; whether *this* run should too is Step 1's call, not a filesystem
228
229
  check.
229
230
 
230
231
  - For a ticket Step 1 cleared for resumption: if its worktree already
231
- exists, `cd` into it and resume; otherwise this shouldn't normally
232
- happen for an in_progress/fixed ticket (report it as unusual rather than
233
- guessing).
234
- - If an unassigned `in_progress` ticket has *no* worktree left (removed,
235
- or a genuinely new pickup), that's unusual for anything but a
236
- freshly-`accepted` ticket — report it rather than reconstructing state
237
- from nothing.
232
+ exists, `cd` into it and resume. If it does not, apply the next bullet.
233
+ - **"Mine, but nothing exists" is a start, not a stop.** An `in_progress`
234
+ ticket assigned to you (or unassigned) with no worktree here, no
235
+ `origin/<branch>`, and no progress comment from any seat has no state to
236
+ reconstruct: cut the worktree per Step 3.1 and begin, with one comment saying
237
+ you did. If `origin/<branch>` exists, cut the worktree from it instead of
238
+ from the base. Only evidence of work somewhere else — a progress comment
239
+ from another ship, or a branch on the remote this ship did not push — is a
240
+ reason to pause, and then it is a stop for a person (see "Stopping: say it in
241
+ data, not prose"), never a bare comment.
238
242
  - Otherwise, pick a ticket to work from the digest's "Step 2" table, which
239
243
  is already `status=accepted`, already narrowed to your lane, and already
240
244
  in the order below. Absent a digest, fetch those tickets yourself and
@@ -354,9 +358,12 @@ if `report_type` is either of those.
354
358
  branch (local commits the remote lacks, or one that cannot fast-forward)
355
359
  makes `crew sync` print `DO NOT CUT A WORKTREE FROM THIS BASE` and exit
356
360
  non-zero. That is a hard stop: do not run `git worktree add`, do not try to
357
- reconcile it yourself. Post a comment on the ticket naming the repo and the
358
- counts `crew sync` printed, leave the ticket as you found it, and end the
359
- run — the operator reconciles the checkout. Otherwise, from
361
+ reconcile it yourself. `crew sync` has already flagged the ship in data
362
+ (`base_unsafe`, shown by `crew status`). Post a comment on the ticket naming
363
+ the repo and the counts it printed, in the "resumes automatically" form
364
+ ("Resumes automatically when the primary checkout's base is level with the
365
+ remote. No status change needed."), leave the ticket as you found it, and end
366
+ the run — the operator reconciles the checkout and the next run resumes. Otherwise, from
360
367
  the primary checkout, `git worktree add <dir> -b <branch> <remote>/<base>`
361
368
  a sibling worktree for this ticket, on a new branch cut from the base
362
369
  branch's REMOTE-TRACKING ref (`origin/main`, say), never from the primary
@@ -727,6 +734,44 @@ to true, and stop that ticket's work for this run. That is a report for the
727
734
  operator, not an accusation to litigate; let them decide what's actually
728
735
  going on.
729
736
 
737
+ ## Stopping: say it in data, not prose
738
+
739
+ A stop that lives only in a comment is invisible to every view that filters on
740
+ data: the operator reads the board's status, assignee and flags, and the ship's
741
+ attention flags, not every ticket's comments. There are two kinds of stop, and
742
+ each is written where the operator looks:
743
+
744
+ 1. **The ticket needs a person** (a question, a decision, an ambiguous spec, a
745
+ branch only a human can reconcile): set `status` to `needs_info` **and**
746
+ `assignee_id` to the operator's Crew row, and `needs_planning` to true when
747
+ it is a decision rather than a fact. These fields are mandatory whenever a
748
+ comment of yours addresses "Operator:" or a named person — a comment alone
749
+ is not a stop.
750
+ 2. **This ship cannot proceed** (the base branch diverged from the remote, a
751
+ tool or hook is missing, no disk, authentication failed) but the ticket
752
+ itself is fine: leave the ticket exactly as it is, `in_progress` and
753
+ assigned to you, so no other ship takes it, and do not set `needs_info`.
754
+ The **ship** carries the flag instead: `crew sync` raises it itself for a
755
+ diverged base (`crew status`, the Ships row and any `hooks.notify` show it
756
+ until the base is level again, and it clears on its own). Post one short
757
+ comment saying which ship is held and why, e.g. "held on <ship>, which
758
+ needs attention: main diverged from origin".
759
+
760
+ **Every stop comment names what un-parks it, and whether anyone has to act.**
761
+ Use one of two templates:
762
+
763
+ - *Resumes on its own:* "Resumes automatically when <condition>. No status
764
+ change needed." Use this for anything a later run can re-check by itself
765
+ (the base is level, a tool is installed, a setup hook passes). The next run
766
+ re-checks the condition first and, once it holds, carries on with the ticket
767
+ with no board edit from anyone.
768
+ - *Needs a person:* "Needs <person> to <action>; set status back to
769
+ `accepted` / clear `needs_info` when done."
770
+
771
+ If an operator sets a ticket you hold back to `accepted` after fixing an
772
+ environment problem, that is a resume, not a new claim: pick it up in its
773
+ existing worktree and note in a comment that the status change was not needed.
774
+
730
775
  ## Guardrails
731
776
 
732
777
  - One ticket at a time. Never mix two tickets' changes in one branch,
@@ -181,7 +181,8 @@ since your last comment: an answer to a question, new direction, or a
181
181
  rebasing) — another ship may have pushed to the branch since you left it.
182
182
  If it reports your branch as diverged (local commits not on the remote AND
183
183
  the remote has moved), stop: comment on the ticket with what it printed and
184
- set `needs_info`; do not merge or rebase it yourself.
184
+ park it for a person (`needs_info` plus the operator as assignee — see
185
+ "Stopping: say it in data, not prose"); do not merge or rebase it yourself.
185
186
  - A ticket QA has bounced back to you comes in as `in_progress`,
186
187
  reassigned to your row (or unassigned, when your ship holds it), with a comment saying what still fails. Treat
187
188
  that exactly like new direction from a hold: read the comment, fix what
@@ -209,13 +210,16 @@ before; whether *this* run should too is Step 1's call, not a filesystem
209
210
  check.
210
211
 
211
212
  - For a ticket Step 1 cleared for resumption: if its worktree already
212
- exists, `cd` into it and resume; otherwise this shouldn't normally
213
- happen for an in_progress/fixed ticket (report it as unusual rather than
214
- guessing).
215
- - If an unassigned `in_progress` ticket has *no* worktree left (removed,
216
- or a genuinely new pickup), that's unusual for anything but a
217
- freshly-`accepted` ticket — report it rather than reconstructing state
218
- from nothing.
213
+ exists, `cd` into it and resume. If it does not, apply the next bullet.
214
+ - **"Mine, but nothing exists" is a start, not a stop.** An `in_progress`
215
+ ticket assigned to you (or unassigned) with no worktree here, no
216
+ `origin/<branch>`, and no progress comment from any seat has no state to
217
+ reconstruct: cut the worktree per Step 3.1 and begin, with one comment saying
218
+ you did. If `origin/<branch>` exists, cut the worktree from it instead of
219
+ from the base. Only evidence of work somewhere else — a progress comment
220
+ from another ship, or a branch on the remote this ship did not push — is a
221
+ reason to pause, and then it is a stop for a person (see "Stopping: say it in
222
+ data, not prose"), never a bare comment.
219
223
  - Otherwise, pick a ticket to work from the digest's "Step 2" table, which
220
224
  is already `status=accepted`, already narrowed to your lane, and already
221
225
  in the order below. Absent a digest, fetch those tickets yourself and
@@ -335,9 +339,12 @@ if `report_type` is either of those.
335
339
  branch (local commits the remote lacks, or one that cannot fast-forward)
336
340
  makes `crew sync` print `DO NOT CUT A WORKTREE FROM THIS BASE` and exit
337
341
  non-zero. That is a hard stop: do not run `git worktree add`, do not try to
338
- reconcile it yourself. Post a comment on the ticket naming the repo and the
339
- counts `crew sync` printed, leave the ticket as you found it, and end the
340
- run — the operator reconciles the checkout. Otherwise, from
342
+ reconcile it yourself. `crew sync` has already flagged the ship in data
343
+ (`base_unsafe`, shown by `crew status`). Post a comment on the ticket naming
344
+ the repo and the counts it printed, in the "resumes automatically" form
345
+ ("Resumes automatically when the primary checkout's base is level with the
346
+ remote. No status change needed."), leave the ticket as you found it, and end
347
+ the run — the operator reconciles the checkout and the next run resumes. Otherwise, from
341
348
  the primary checkout, `git worktree add <dir> -b <branch> <remote>/<base>`
342
349
  a sibling worktree for this ticket, on a new branch cut from the base
343
350
  branch's REMOTE-TRACKING ref (`origin/main`, say), never from the primary
@@ -709,6 +716,44 @@ to true, and stop that ticket's work for this run. That is a report for the
709
716
  operator, not an accusation to litigate; let them decide what's actually
710
717
  going on.
711
718
 
719
+ ## Stopping: say it in data, not prose
720
+
721
+ A stop that lives only in a comment is invisible to every view that filters on
722
+ data: the operator reads the board's status, assignee and flags, and the ship's
723
+ attention flags, not every ticket's comments. There are two kinds of stop, and
724
+ each is written where the operator looks:
725
+
726
+ 1. **The ticket needs a person** (a question, a decision, an ambiguous spec, a
727
+ branch only a human can reconcile): set `status` to `needs_info` **and**
728
+ `assignee_id` to the operator's Crew row, and `needs_planning` to true when
729
+ it is a decision rather than a fact. These fields are mandatory whenever a
730
+ comment of yours addresses "Operator:" or a named person — a comment alone
731
+ is not a stop.
732
+ 2. **This ship cannot proceed** (the base branch diverged from the remote, a
733
+ tool or hook is missing, no disk, authentication failed) but the ticket
734
+ itself is fine: leave the ticket exactly as it is, `in_progress` and
735
+ assigned to you, so no other ship takes it, and do not set `needs_info`.
736
+ The **ship** carries the flag instead: `crew sync` raises it itself for a
737
+ diverged base (`crew status`, the Ships row and any `hooks.notify` show it
738
+ until the base is level again, and it clears on its own). Post one short
739
+ comment saying which ship is held and why, e.g. "held on <ship>, which
740
+ needs attention: main diverged from origin".
741
+
742
+ **Every stop comment names what un-parks it, and whether anyone has to act.**
743
+ Use one of two templates:
744
+
745
+ - *Resumes on its own:* "Resumes automatically when <condition>. No status
746
+ change needed." Use this for anything a later run can re-check by itself
747
+ (the base is level, a tool is installed, a setup hook passes). The next run
748
+ re-checks the condition first and, once it holds, carries on with the ticket
749
+ with no board edit from anyone.
750
+ - *Needs a person:* "Needs <person> to <action>; set status back to
751
+ `accepted` / clear `needs_info` when done."
752
+
753
+ If an operator sets a ticket you hold back to `accepted` after fixing an
754
+ environment problem, that is a resume, not a new claim: pick it up in its
755
+ existing worktree and note in a comment that the status change was not needed.
756
+
712
757
  ## Guardrails
713
758
 
714
759
  - One ticket at a time. Never mix two tickets' changes in one branch,