@skitterbyte/skitterspec-jira 2.15.0 → 2.17.0

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.
Files changed (43) hide show
  1. package/MIGRATION.md +27 -0
  2. package/assets/claude-md-section.md +18 -9
  3. package/assets/commands/allow-main.md +1 -2
  4. package/assets/core/SETUP.md +12 -12
  5. package/assets/core/env.config.json.example +0 -1
  6. package/assets/core/env.config.md +28 -26
  7. package/assets/hooks/main-guard.cjs +73 -6
  8. package/assets/hooks/project-guard.cjs +80 -6
  9. package/assets/hooks/review-gate.cjs +73 -6
  10. package/assets/review/page.html +10 -83
  11. package/assets/rules/commit-trailers.md +2 -2
  12. package/assets/rules/spec-planning.md +26 -22
  13. package/assets/rules/spec-reports.md +52 -52
  14. package/assets/skills/no-spec/SKILL.md +2 -2
  15. package/assets/skills/spec/SKILL.md +22 -5
  16. package/assets/skills/spec-bug/SKILL.md +18 -15
  17. package/assets/skills/spec-complete/SKILL.md +23 -1
  18. package/assets/skills/spec-diff/SKILL.md +47 -115
  19. package/assets/skills/spec-hotfix/SKILL.md +18 -15
  20. package/assets/skills/spec-init/SKILL.md +4 -2
  21. package/assets/skills/spec-next/SKILL.md +66 -41
  22. package/assets/skills/spec-park/SKILL.md +92 -0
  23. package/assets/skills/spec-resume/SKILL.md +153 -0
  24. package/assets/skills/spec-review/SKILL.md +2 -2
  25. package/assets/skills/spec-reviewed/SKILL.md +1 -8
  26. package/assets/skills/spec-sync/SKILL.md +10 -10
  27. package/assets/skills/spec-todo/SKILL.md +5 -2
  28. package/package.json +1 -1
  29. package/src/cli.js +361 -208
  30. package/src/env/bootstrap.js +3 -1
  31. package/src/env/checks.js +49 -0
  32. package/src/env/config.js +27 -7
  33. package/src/env/hooks.js +53 -4
  34. package/src/env/live.js +48 -0
  35. package/src/env/provision.js +10 -2
  36. package/src/env/resume.js +234 -0
  37. package/src/env/review.js +71 -173
  38. package/src/env/serve.js +3 -7
  39. package/src/init.js +383 -50
  40. package/src/tada/todo.js +19 -4
  41. package/src/vendor/provider-kit/src/specs.js +46 -1
  42. package/src/vendor/sync-core/src/summary.js +10 -1
  43. package/assets/commands/spec-remote-review.md +0 -37
@@ -35,11 +35,32 @@ Before marking complete, confirm the work is actually finished:
35
35
  - If genuinely incomplete work remains, **stop and tell the user** rather than
36
36
  forcing completion. Offer to finish it (`/spec-next`) or to complete with the
37
37
  remaining items explicitly listed as deferred.
38
+ - **Report any blocker still standing; never refuse over one.** Run
39
+ `skitterspec spec-env resume <spec> --json` and read `blockers`. Each
40
+ `> **Blocked on:**` line still in the spec goes in the report's `Notes`,
41
+ quoted with its `until` date where it has one. A spec can finish with a wait
42
+ still written down: the soak may have been dropped, or the deploy may have
43
+ happened without anyone saying so. The line staying in the completed spec is
44
+ the record of that, and nothing is gated on it. With none, say nothing.
38
45
  - **Check for pre-existing uncommitted changes — before you touch anything.**
39
46
  Run `git status`. Anything already uncommitted is the *user's* work (a
40
47
  half-finished phase, a stray fix) and must not be swept into the completion
41
48
  commit: offer `/commit` and **stop**. Everything this skill writes in steps 3–4
42
49
  is its own, and step 4 commits that.
50
+ - **Run the project's completion checks**, once the suite is green:
51
+
52
+ ```
53
+ skitterspec spec-env checks <spec>
54
+ ```
55
+
56
+ It runs each `complete.checks` command from `env.config.json` in the spec's
57
+ worktree — the project's own end-of-spec questions, such as a docs report —
58
+ and always exits 0. Relay what each printed into `Notes`. A check marked
59
+ `failed` makes the run `⚠️`; it **never stops the completion or the land**,
60
+ because a check's output is a reason to look again, not proof of a fault.
61
+ Where a check names a fix you can make now (re-read a page, then record it),
62
+ offer it — the operator answers. `no completion checks configured` is the
63
+ usual answer and says nothing in the report.
43
64
 
44
65
 
45
66
  ### An epic — complete after its members
@@ -274,7 +295,8 @@ the shape; this section carries only what is specific here.
274
295
 
275
296
  `Notes` carries the after-land commands step 6 ran, and where, and any it
276
297
  skipped because the land matched no `when` — or, on a failure, the command that failed and the ones that did not run. A project with
277
- no `afterLand` says nothing about it.
298
+ no `afterLand` says nothing about it. It also carries what each completion
299
+ check reported (§2) — a project with none says nothing about them either.
278
300
 
279
301
  **Report this spec and no other.** A completion is the moment it is most
280
302
  tempting to survey what is left — what else is in flight, what is in the
@@ -14,11 +14,11 @@ A phase is built in its own worktree, and the terminal is somewhere else — so
14
14
  `git diff` answers about the wrong branch and a 350-line diff read as terminal
15
15
  text is scrolling, not review. This renders the diff as a page instead: the
16
16
  engine collects it with `git -C`, writes a self-contained HTML file, and you open
17
- it locally or publish it and read it on a phone.
17
+ it locally, or on a phone over your LAN or tailnet.
18
18
 
19
19
  **The diff never passes through the model.** git writes the patches and the
20
20
  engine splices them into the page. A 266KB patch costs **zero** context tokens,
21
- and a 289KB page costs the same to publish as a 5KB one. Do not "simplify" this
21
+ and a 289KB page costs the same to serve as a 5KB one. Do not "simplify" this
22
22
  by reading the diff and emitting HTML — that is the one change that would undo
23
23
  the whole design.
24
24
 
@@ -354,8 +354,8 @@ spend it.
354
354
 
355
355
  ## 2b. An action changes something, then hands the page back
356
356
 
357
- Only on a pass carrying an **action**. Three of them, and
358
- **all three end the same way**:
357
+ Only on a pass carrying an **action**. Two of them, and
358
+ **both end the same way**:
359
359
  do the thing, re-render (§4), and wait again (§4b) — because the reader
360
360
  still has a decision in front of them and has not made it.
361
361
 
@@ -363,7 +363,11 @@ still has a decision in front of them and has not made it.
363
363
  |--------|----|------|
364
364
  | `live-on` | commit first (below), then `skitterspec spec-env live take <spec>` | re-render `--branch`, wait again |
365
365
  | `allow-network` | `skitterspec spec-env review allow network` | re-render, wait again |
366
- | `allow-remote` | `skitterspec spec-env review allow remote <spec> --set on` | publish (§6), re-render, wait again |
366
+
367
+ **There is no `allow-remote` either.** It permitted publishing the page as a
368
+ claude.ai artifact, and that tier was retired (feat-tailnet-remote-tier):
369
+ `remote` is the tailnet address now, on `network`'s switch, so `allow-network`
370
+ turns both on.
367
371
 
368
372
  **THERE IS NO `live-off`, and its absence is deliberate.** Putting *this* change
369
373
  live is about the diff on screen; handing the whole instance back to `main` is a
@@ -371,7 +375,7 @@ workspace decision with nothing to do with this review. The `live:` line names
371
375
  `/spec-live main` for it — a command the operator types, which is also why no
372
376
  skill runs it.
373
377
 
374
- **AN ACTION IS NOT A VERDICT, and the gate is untouched by all three.** A phase
378
+ **AN ACTION IS NOT A VERDICT, and the gate is untouched by both.** A phase
375
379
  that ended armed the gate, and it is discharged by a committing verdict or a
376
380
  recorded skip and by nothing else — so after any of these the phase still owes
377
381
  an answer, and that is exactly why the run waits again rather than finishing.
@@ -425,17 +429,9 @@ edits `specs/.core/env.config.json` in the **primary checkout**, so it changes
425
429
  behaviour for everyone who pulls and leaves that tree dirty — one server binds
426
430
  for every spec, so the bind cannot be narrower. The engine prints the absolute
427
431
  path; relay that rather than letting a shared setting change land silently.
428
- **`allow remote` is per spec** and writes only a gitignored sidecar beside the
429
- page, so the base branch stays clean and no other spec may publish. And it is **enable-only**: there is no
430
- action that turns a tier off, because turning `network` off from a page reached
431
- over the network kills the page doing the turning.
432
-
433
- **`allow remote` permits publishing. It does not publish by itself.**
434
- Publishing stays an explicit ask (§6), because the page it leaves behind is one
435
- this tooling cannot remove. A reader who presses `Allow remote` on the page, or
436
- types `/spec-remote-review` beside a waiting review, **has** asked: the engine
437
- names the page on its `review:`/`publish:` lines, and that page is published
438
- and the stack re-shown. Nothing it does not name is published.
432
+ And it is **enable-only**: there is no action that turns a tier off, because
433
+ turning `network` off from a page reached over the network kills the page doing
434
+ the turning.
439
435
 
440
436
  ## 3. Gate it on nothing
441
437
 
@@ -493,7 +489,7 @@ could not run is that strip's business rather than yours.
493
489
 
494
490
  **There is no switch left to handle.** The engine serves and prints a **stack** —
495
491
  one line per tier, `local`, `network`, `remote`, each either a URL or the one
496
- command that turns it on. All of it is free and none of it publishes anything.
492
+ command that turns it on. All of it is free, and all of it is one server.
497
493
  You are not choosing between them; you are relaying every line it printed.
498
494
  `.claude/rules/spec-reports.md` carries the shape and why.
499
495
 
@@ -564,7 +560,7 @@ says:
564
560
  |------|---------|
565
561
  | `local` | the loopback URL — or the `file://` page when nothing served, which says it cannot send a verdict |
566
562
  | `network` | the LAN URL and any `also:` alternates under it — or `off` with the command that turns it on |
567
- | `remote` | the published URL and *a verdict here needs `/spec-reviewed`* — or `off` with the command that turns it on |
563
+ | `remote` | the tailnet URL, labelled `(tailnet)` — or `off` with `network`'s command, or `no tailnet address on this machine` |
568
564
 
569
565
  **The `reader:` line decides nothing here any more.** It is still printed and
570
566
  still the only place that question is answered, but the offer no longer branches
@@ -581,23 +577,20 @@ tested and would drift. The ranking and the traps
581
577
  a tty under Claude Code) live in `review.js` beside the code, which is where they
582
578
  belong.
583
579
 
584
- **It authorises serving, never publishing.** A `remote` reader authorises a
585
- local server — one process, ended by one flag, leaving nothing behind — and
586
- nothing more. Publishing leaves a page this tooling cannot remove, so it is an
587
- ask, in every case, always. The two were once one rule, and lumping them together
588
- is what left a remote reader holding a dead link: see
589
- `specs/complete/bug-remote-reader-gets-a-dead-link/`.
580
+ **It authorises serving, and serving is all there is.** A `remote` reader
581
+ authorises a local server — one process, ended by one flag, leaving nothing
582
+ behind. Publishing the page was once a second answer and was retired with the
583
+ published tier (feat-tailnet-remote-tier): the tailnet reaches a reader off the
584
+ wifi without a page this tooling could never remove.
590
585
 
591
586
  When serving fails — a busy port, or a machine with no network address — the
592
- engine falls back to the `file://` URL with its marker and nothing breaks. That
593
- is the one case where naming publishing is worth doing, because it is the only
594
- answer left.
587
+ engine falls back to the `file://` URL with its marker and nothing breaks.
595
588
 
596
589
  `review.reader` in `env.config.json` (`local` · `remote` · `detect`) lets the
597
590
  operator settle where they are reading, and an explicit value is believed without
598
591
  sniffing. What the engine may *do* is settled by the tier settings instead —
599
- `review.serve` (`always` · `never`), `review.allowNetwork` and
600
- `review.allowRemote` — not by the reader.
592
+ `review.serve` (`always` · `never`) and `review.allowNetwork` — not by the
593
+ reader.
601
594
 
602
595
  ## 4b. Wait for the verdict — because asking for one means waiting for it
603
596
 
@@ -623,9 +616,6 @@ transport only decides what carries the answer back:
623
616
  - **A `file://` page** has nothing to post to, so the reader pastes the pass and
624
617
  their next message carries it. Say you are holding and **end the turn** — that
625
618
  is the same wait, carried by the conversation. It is not a lesser one.
626
- - **A published page** writes to the artifact's own store, which nothing reaches
627
- from here. That is the one case where the honest sentence is *press a verdict,
628
- then type `/spec-reviewed`* — see the published-page paragraph below.
629
619
 
630
620
  **Do not ask, and therefore do not wait, on a render nobody is waiting behind** —
631
621
  a bare `--page-only`, a page produced alongside other work, a pass that has
@@ -673,18 +663,15 @@ the stranger's pass the old rule was written about, and two arrivals refuse
673
663
  rather than pick. What is left of the old rule is unchanged and still absolute —
674
664
  outside this window, a pass is claimed because a person named it.
675
665
 
676
- **The wait covers the SERVED page and nothing else.** It watches the engine's
677
- local store, which is where a served page POSTs. A **published** page writes to
678
- the artifact's own store instead, and no watch of any kind reaches that — so a
679
- verdict pressed there is invisible until someone asks for it. Never start a
680
- wait and then hand over a published link under it: that reads as a promise to
681
- notice, and it was made three times in a row while three verdicts sat unread.
682
- Where the page is published, say plainly that `/spec-reviewed` is what picks it
683
- up.
666
+ **The wait covers every served door.** It watches the engine's local store,
667
+ which is where a served page POSTs — from `local`, `network` or `remote` alike,
668
+ since all three are the same server. A published page once wrote to a store no
669
+ watch reached, and three verdicts sat unread under a line claiming the run was
670
+ waiting; that is why the published tier was retired rather than caveated.
684
671
 
685
672
  **`/spec-reviewed` is not replaced by this.** It stays the way in for a pass
686
- that arrived when nobody was waiting, for the two-passes case, and for every
687
- published page. It is user-only, and that is still the
673
+ that arrived when nobody was waiting, and for the two-passes case. It is
674
+ user-only, and that is still the
688
675
  enforcement that makes a named claim a person's decision.
689
676
 
690
677
  **A harness with no file-watch is no longer on that list**, and that is the
@@ -743,78 +730,23 @@ Then re-render with it spliced in — the engine renders it, you never emit HTML
743
730
  skitterspec spec-env review <spec> --review <file>
744
731
  ```
745
732
 
746
- ## 6. Publish only when asked
747
-
748
- **Never publish unprompted.** Publishing leaves something behind that this
749
- tooling cannot remove, and the page already works as a local file. Offer it;
750
- do not assume it.
751
-
752
- When the user asks:
753
-
754
- - **One spec is one page.** `--json` reports `url` when this spec has been
755
- published before. If it has, **update that URL** rather than creating a
756
- second entry, and label the new version for the phase (`phase-3`).
757
- - **On the first publish**, write the returned URL to the path `--json` reports
758
- as `urlFile` — one line, no formatting. That file is how every later phase
759
- finds the same page. Never construct the path yourself.
760
- - **Say, once, that it is theirs now.** When you report a URL, say in the same
761
- breath that skitterspec cannot remove the page and that `/artifacts` (or the
762
- gallery at `claude.ai/code/artifacts`) is where it goes. This is the moment the
763
- decision is being made, so it is the moment worth saying it — `spec-env down`
764
- repeats it at teardown, by which point the page has outlived the spec.
765
- - **Degrade in one line.** If the harness cannot publish — no capability, an
766
- error — say so, report the local file path, and carry on. That is a working
767
- outcome, not a failure.
768
- - **Reach for `--publish-copy`, never a hand transform.** The engine writes the
769
- page as a complete HTML document and an artifact host wraps page *content*, so
770
- publishing the page as-written nests two documents. `spec-env review <spec>
771
- --publish-copy` emits the body-only copy and names its path on a `publish:`
772
- line. Do not split the document yourself: a rendered page contains the diff,
773
- and a diff of this project contains `<!doctype html>` as ordinary patch text.
774
-
775
- The engine knows nothing about publishing and cannot do it. It writes a file and
776
- reads a URL back as an opaque string; everything about what that string means
777
- lives here.
778
-
779
- ### Publish it so the buttons still work
780
-
781
- A published page is the answer for a reader **no local server can reach** — a
782
- phone on mobile data, anyone away from the LAN the engine serves on. It is also
783
- the one surface where the verdict buttons have nowhere to POST: the page's own
784
- URL is on claude.ai, so a POST fails and the reader gets "could not reach the
785
- server" after pressing a verdict they meant.
786
-
787
- So **publish it with a store**, and the page uses it:
788
-
789
- - Declare `capabilities: {db: {}}`. The page checks for `window.claude.use` and
790
- writes the pass into the artifact's own `passes` collection instead of
791
- POSTing. Nothing else about it changes — same marks, same verdicts, same
792
- refusal to commit over an open comment.
793
- - **Do not declare `user`.** The pass is not per-viewer private state, a shared
794
- collection is what it wants, and a declaration nothing uses is a grant asked
795
- for nothing.
796
- - **Same file path every time**, so a redeploy reuses the URL. That is what
797
- answers the clear-down worry: one page per spec, replaced at each render,
798
- rather than an artifact per phase accumulating in the gallery.
799
- - **Send a push notification with the URL** when you publish. The reader is by
800
- definition somewhere else — that is why it was published.
801
-
802
- ### Taking a stored pass back
803
-
804
- A pass in the store is claimed the same way in spirit and a different way in
805
- mechanism, because the engine cannot see it:
806
-
807
- 1. Read the `passes` collection with the Artifact tool's `read_db`.
808
- 2. **Apply step 0 unchanged.** One waiting pass, offered by what it says and
809
- when it was sent; two is a refusal to guess. Nothing here is automatic —
810
- there is no wait window on this transport, because nothing pushes from the
811
- store to this conversation.
812
- 3. Write the pass's `blob` to a scratch file **verbatim** and merge it with
813
- `spec-env review <spec> --notes <file>`. It rejoins the ordinary path at §2
814
- step 1, so the verdict is judged and routed exactly as any other.
815
- 4. **Delete the document** with `write_db` once it is merged. A claim consumes:
816
- a pass left in the store is claimable twice, which is the one property the
817
- six-digit code has always had and this transport must not lose.
733
+ ## 6. There is no publishing
734
+
735
+ **This skill does not publish the page, and nothing in this workflow does.**
736
+ Publishing it as a claude.ai artifact was retired with the published tier
737
+ (feat-tailnet-remote-tier). It wrote to a store no wait could watch, and left a
738
+ page this tooling could never remove. A reader off the wifi opens the `remote`
739
+ line — the review server's tailnet address — and sends the verdict into the
740
+ same store as every other door.
741
+
742
+ **Where no door reaches the reader** — no network, no tailnet, a server that
743
+ could not start — say so plainly with the stack the engine printed, and stop.
744
+ Do not reach for the Artifact tool as a fallback: that is the second store this
745
+ design removed.
746
+
747
+ A page published before this change still exists, and teardown still names it
748
+ (`spec-env down` reads the `.url` sidecar it left). Nothing writes that sidecar
749
+ any more.
818
750
 
819
751
  ## 7. Report
820
752
 
@@ -42,8 +42,14 @@ Then:
42
42
 
43
43
  1. **Exclude what's already adopted.** Run `skitterspec spec-sync linked --json`
44
44
  and drop any issue whose identifier is already stamped on a spec. If the user
45
- named such an issue directly, **stop** and point them at that spec — one issue
46
- never becomes two specs.
45
+ named such an issue directly, it already has a spec, and one issue never
46
+ becomes two specs. **Offer `/spec-resume <ref>`** instead, carrying the rest
47
+ of what they wrote as its message: feedback on shipped work, or "the deploy
48
+ is out", belongs to the spec that is already in flight. Confirm it, as kind
49
+ routing does (`.claude/rules/spec-planning.md`), and invoke it in-session on
50
+ a yes. On a no, stop and name the spec. That spec may exist only on its
51
+ branch, which `linked` reads too, so a missing `specs/` folder here is not
52
+ evidence that the issue is free.
47
53
  2. **Read the issue** with the discovered issue-read tool (`getJiraIssue`):
48
54
  title, description, labels, reporter, URL. If Jira isn't connected, say so
49
55
  and offer to carry on without it — a spec written from the user's own
@@ -587,10 +593,9 @@ line with it where the engine printed one, and nothing where it did not.
587
593
  `.claude/rules/spec-reports.md` carries the shape, including why a rigid
588
594
  contract took that line.
589
595
 
590
- - **Never write the review unasked**, and **never publish**. Publishing leaves
591
- something behind that this tooling cannot remove, so it is always an ask. A
592
- `file://` link is no use on a phone, and saying so **is** the ask —
593
- `/spec-diff` §6 owns how.
596
+ - **Never write the review unasked**, and **never publish** — there is no
597
+ publishing (`/spec-diff` §6). A `file://` link is no use on a phone, and the
598
+ stack says why nothing was served.
594
599
  **The `reader:` line no longer decides anything here, and that is the point.**
595
600
  It is still printed, and it is still the only place that question is answered —
596
601
  but the offer does not change with it, because the stack lists every tier
@@ -600,22 +605,20 @@ that branching is exactly what produced a `file://` page on a session detected
600
605
  underneath a reader when detection flipped mid-session.
601
606
 
602
607
  So: **relay all three tier lines, every time.** Any `also:` lines sit under
603
- `network` — the other addresses this machine has, offered because the
604
- best-guess one can be wrong — so pass them on rather than editing them out: in
605
- the banner they are indented sub-bullets under **network**, labels
606
- (`(tailnet)`) included, exactly as `.claude/rules/spec-reports.md` shows.
608
+ their tier — the other addresses this machine has of that kind — so pass them
609
+ on rather than editing them out: in the banner they are indented sub-bullets
610
+ under that tier, labels included, exactly as `.claude/rules/spec-reports.md`
611
+ shows.
607
612
 
608
613
  **Never read an environment variable to decide anything about the offer** — not
609
614
  `SSH_CONNECTION`, not `CLAUDE_CODE_*`, not a tty check. There is nothing left
610
615
  here for a detection to decide, and a second implementation of one could not be
611
616
  tested and would drift.
612
617
 
613
- **Serving is the engine's to do; publishing is never.** A `remote` reader
618
+ **Serving is the engine's to do, and there is nothing else.** A `remote` reader
614
619
  authorises a local server — one process, ended by one flag, leaving nothing
615
- behind — and authorises nothing else. Publishing leaves a page this tooling
616
- cannot remove, so it stays an ask in every case, always. If the engine could not
617
- serve (a busy port, a machine with no network address) it falls back to the
618
- `file://` URL with its marker, and that is when publishing is worth naming.
620
+ behind. If the engine could not serve (a busy port, a machine with no network
621
+ address) it falls back to the `file://` URL with its marker.
619
622
 
620
623
  - **Never fatal.** A failed render — no worktree, a git error — is one line, and
621
624
  the fix is still done. The page is a convenience; the repo is the record.
@@ -27,14 +27,16 @@ clobber customised content. Finish with a summary of created vs already-present.
27
27
  > **Leave alone**. Non-interactively it only adds what's missing; `--resync` /
28
28
  > `--reset` (reset needs `--yes`) drive the stronger actions.
29
29
 
30
- The system is **eighteen skills**: `spec` (feature, chore or issue), `spec-bug` (bug),
30
+ The system is **twenty skills**: `spec` (feature, chore or issue), `spec-bug` (bug),
31
31
  `spec-hotfix` (a bug fixed on a released tag),
32
32
  `spec-incident` (a live outage, recorded as it happens),
33
33
  `no-spec` (work that genuinely has none — off the base branch all the same),
34
34
  `spec-review`, `spec-start`,
35
35
  `spec-next`, `spec-diff` (read a spec's diff as a page),
36
36
  `spec-reviewed` (pick up a review you approved on that page),
37
- `spec-epic-next` (name the epic member that can start now), `spec-to-main`,
37
+ `spec-epic-next` (name the epic member that can start now),
38
+ `spec-resume` (back into a spec in flight, by name or ticket),
39
+ `spec-park` (record what a spec in flight is waiting on), `spec-to-main`,
38
40
  `spec-complete`, `spec-cancel`, `spec-tada` (a ta-da list of what got done),
39
41
  `spec-todo` (where you left off, and what is ahead), `spec-idea` (an idea
40
42
  parked without a spec),
@@ -50,8 +50,10 @@ evidence.
50
50
  Otherwise resolve **in this order**, and stop at the first that answers:
51
51
 
52
52
  1. **The live spec of this checkout** — run
53
- `skitterspec spec-env live status` and read its `live:` line. `live: yes`
54
- names the spec whose branch is checked out here; that is the one to build.
53
+ `skitterspec spec-env live status` and read its `in-flight:` line. A spec
54
+ name there is the spec whose branch is checked out here; that is the one to
55
+ build. `none` and `unknown` answer nothing — carry on down the list. (The
56
+ bare report's first line is a bare `spec-env live:` header, not an answer.)
55
57
  2. **The worktree you are standing in** — if this session's cwd is inside a
56
58
  spec's worktree, that spec is in flight *for this session*. This is the
57
59
  manual-parallel path: several specs may be provisioned, and a terminal tab
@@ -62,10 +64,17 @@ Otherwise resolve **in this order**, and stop at the first that answers:
62
64
  named, from wherever you happen to be standing:
63
65
 
64
66
  ```
65
- skitterspec spec-env resolve
67
+ skitterspec spec-env resolve --json
66
68
  ```
67
69
 
68
- Take its `spec:` line **only when it names exactly one spec**.
70
+ Take its `spec` **only when it names exactly one spec** (`ok: true`); `via`
71
+ says which rung answered.
72
+
73
+ From the primary checkout that includes the spec `/spec-live main` last
74
+ released here (`via: "last-live"`): releasing leaves the session standing
75
+ in the primary checkout, which says nothing on its own, so the engine
76
+ remembers the spec it handed back — for as long as that spec's worktree
77
+ exists — ahead of counting worktrees.
69
78
 
70
79
  This is the **durable** rung, and that is the whole reason it exists. Rungs 1
71
80
  to 3 all read *session* state, and session state does not survive a `/clear`,
@@ -81,13 +90,14 @@ Otherwise resolve **in this order**, and stop at the first that answers:
81
90
  `.claude/rules/spec-planning.md` states has none left.
82
91
 
83
92
  **Say which spec you resolved and how**, before writing a line of it:
84
- *"not standing in a worktree — `<spec>` is the only spec provisioned"*. Rungs
93
+ *"not standing in a worktree — `<spec>` is the only spec provisioned"*, or
94
+ *"`<spec>` was last live on this checkout"*. Rungs
85
95
  1 to 3 are self-evident to whoever typed the command; this one is not, and
86
96
  the operator cannot see from where they sit what you picked.
87
97
 
88
- **Several worktrees stay a refusal.** The engine names them and resolves
89
- nothing — exactly the ambiguity the refusal below exists for. Relay its list
90
- unchanged and stop; never pick from it.
98
+ **Several worktrees stay a refusal.** The engine answers `ok: false` with
99
+ `reason: "ambiguous"` and lists the `candidates`, each with its `worktree`,
100
+ and resolves nothing. Relay that list unchanged and stop; never pick from it.
91
101
 
92
102
  WHAT WOULD FOOL THIS: a worktree left behind by a declined teardown is still
93
103
  a worktree, so a finished spec can go on counting as provisioned. That widens
@@ -96,10 +106,18 @@ Otherwise resolve **in this order**, and stop at the first that answers:
96
106
  manufacture an *absence*, which is why the absence below is still worth
97
107
  refusing on.
98
108
 
99
- **If none answers, refuse and stop:**
100
- `no spec in flight — run /spec-start <name> to start one`.
109
+ **If none answers, refuse and stop — and say which absence it is.** Rung 4's
110
+ `reason` tells them apart, and they send the reader to different places:
111
+
112
+ - **`reason: "none"`** — no spec has a worktree anywhere:
113
+ `no spec in flight — run /spec-start <name> to start one`.
114
+ - **`reason: "ambiguous"`** — several specs are in flight and this session is
115
+ standing in none of them: `several specs are in flight and this session is
116
+ not in one of them`, then the candidates as the engine listed them (spec and
117
+ worktree), and `/spec-next --worktree <path>` as the way to name one.
118
+ `/spec-start` is wrong here: every one of them is already started.
101
119
 
102
- That now answers a real absence — no worktree anywhere, the engine included —
120
+ The first answers a real absence — no worktree anywhere, the engine included —
103
121
  rather than a session that merely lost track of where it was standing.
104
122
 
105
123
  **Never fall back to the spec "in context".** A spec discussed in conversation
@@ -215,6 +233,24 @@ once at the start, so each phase is asked as it is built.
215
233
  not matter for this phase, and the bypass is that knowledge on the record. What
216
234
  it rules out is the silent build.
217
235
 
236
+ **Then ask whether the spec is parked.** Run
237
+ `skitterspec spec-env resume <spec> --json` and read `blocking`. It is the
238
+ `> **Blocked on:**` line that holds this build: the spec's own line, or this
239
+ phase's. A line on a later phase is not this build's business, and the engine
240
+ leaves it out.
241
+
242
+ - **`blocking: null`**: carry on in silence. This is the usual case.
243
+ - **Anything else**: stop with the `⏸` block before marking anything started.
244
+ `Why` quotes the blocker, with its `until` date where it has one. `Next` is
245
+ `/spec-resume <spec> <what happened>`, which is the exit. It hears that the
246
+ condition is met, clears the line on the record, and builds.
247
+
248
+ **No bypass is offered here, and none is needed.** Unlike an epic wait, a
249
+ blocker is a condition only a person can judge (a deploy, another project's
250
+ release, a week in prod), so the person who knows it is met says so to
251
+ `/spec-resume`, and that clears it with a Changelog line. A build past one
252
+ without a word would be exactly the silent step this check exists to stop.
253
+
218
254
  Mark it started: set the phase-file heading to `🔄` and its `> **Status:**` to
219
255
  `In progress`, and flip the matching row in the overview phase index to `🔄`.
220
256
 
@@ -464,11 +500,11 @@ so one subject lives in one place:
464
500
 
465
501
  - **local** — <the `local:` URL>
466
502
  - **network** — <the `network:` URL, or off with the command that turns it on>
467
- - **remote** — <the `remote:` URL, or off with the command that turns it on>
503
+ - **remote** — <the `remote:` tailnet URL, or what that line says when there is none>
468
504
 
469
505
  <the `live:` line, exactly as the engine printed it — omitted when it printed none>
470
506
 
471
- I'm holding here until you send a verdict — the wait covers local and network.
507
+ I'm holding here until you send a verdict — the wait covers <the tiers the engine's `covers` names>.
472
508
 
473
509
  `/spec-reviewed` picks it up · `spec-env review skip "<reason>"` moves on
474
510
 
@@ -489,22 +525,14 @@ reviewer sees the heading they have always seen.
489
525
  printed one line per tier, in that order, and `--json` carries the same thing as
490
526
  `tiers`. Relay all three whatever they say, including the ones that are off with
491
527
  the command that turns them on: a reader who has left the house cannot ask for a
492
- remote page they were never shown.
528
+ remote door they were never shown.
493
529
 
494
530
  **Do not pick one for them.** The engine used to, and the guess failed three
495
531
  separate ways in one day. `.claude/rules/spec-reports.md` carries the shape and
496
532
  why the *one link, never two* rule now reads
497
- **one link per reachable store, each labelled** —
498
- local and network are two doors into one room, so the wait
499
- covers both. `remote` is a second store, so its own line carries its own caveat:
500
- *a verdict here needs `/spec-reviewed`*, because nothing pushes from
501
- the artifact store into this conversation. Never let the holding line promise a
502
- wait over that tier.
503
-
504
- **Publishing is still never yours to do.** `remote` reading `off` is the normal
505
- state, and the fix for it is the command on that line — typed by the operator,
506
- or pressed on the page, because a published page is one this tooling cannot
507
- remove and permitting it is not publishing it.
533
+ **one link per door, each labelled, all into one store** —
534
+ local, network and remote are three doors into one room, so the wait
535
+ covers every one the stack shows.
508
536
 
509
537
  **The `live:` line goes with the stack**, and it is the engine's line too —
510
538
  copied as printed, and **absent when the engine printed none**. The page can act
@@ -519,7 +547,7 @@ verdict.
519
547
  **Where you are not waiting, it stays the `Review` row** — the counts and the
520
548
  page link, and **no question**:
521
549
 
522
- | **Review** | <N> files, +<a> −<d> · **local** <URL> · **network** <URL> · **remote** off |
550
+ | **Review** | <N> files, +<a> −<d> · **local** <URL> · **network** <URL> · **remote** <URL> (tailnet) |
523
551
 
524
552
  At the end of a phase you are always waiting, so this shape belongs to the
525
553
  renders that are not this step: a mid-phase `/spec-diff`, a page produced
@@ -633,10 +661,9 @@ Relay the engine's **stack** — the `local:`, `network:` and `remote:` lines
633
661
  never the bare `page:` path: a path is not clickable in any terminal, and a page
634
662
  nobody can open is a page nobody reads.
635
663
 
636
- - **Never write the review unasked**, and **never publish**. Publishing leaves
637
- something behind that this tooling cannot remove, so it is always something
638
- someone asks for. A `file://` link is no use on a phone, and saying so **is**
639
- the ask — publishing is the answer to it, and `/spec-diff` §6 owns how.
664
+ - **Never write the review unasked**, and **never publish** — there is no
665
+ publishing (`/spec-diff` §6). A `file://` link is no use on a phone, and the
666
+ stack says why nothing was served.
640
667
  **The `reader:` line no longer decides anything here, and that is the point.**
641
668
  It is still printed, and it is still the only place that question is answered —
642
669
  but the offer does not change with it, because the stack lists every tier
@@ -646,23 +673,20 @@ that branching is exactly what produced a `file://` page on a session detected
646
673
  underneath a reader when detection flipped mid-session.
647
674
 
648
675
  So: **relay all three tier lines, every time.** Any `also:` lines sit under
649
- `network` — the other addresses this machine has, offered because the
650
- best-guess one can be wrong — so pass them on rather than editing them out: in
651
- the banner they are indented sub-bullets under **network**, labels
652
- (`(tailnet)`) included, exactly as `.claude/rules/spec-reports.md` shows.
676
+ their tier — the other addresses this machine has of that kind — so pass them
677
+ on rather than editing them out: in the banner they are indented sub-bullets
678
+ under that tier, labels included, exactly as `.claude/rules/spec-reports.md`
679
+ shows.
653
680
 
654
681
  **Never read an environment variable to decide anything about the offer** — not
655
682
  `SSH_CONNECTION`, not `CLAUDE_CODE_*`, not a tty check. There is nothing left
656
683
  here for a detection to decide, and a second implementation of one could not be
657
684
  tested and would drift.
658
685
 
659
- **Serving is the engine's to do; publishing is never.** The server is one
686
+ **Serving is the engine's to do, and there is nothing else.** The server is one
660
687
  process, ended by one flag, leaving nothing behind, and `review.allowNetwork`
661
- decides whether it binds wide — no detection involved. A reader the server CAN
662
- reach is not a reason to publish as well: the tiers they can already open are on
663
- the stack. Publishing leaves a page this tooling cannot remove, so it stays an
664
- ask in every case, always — which is why `remote` shows the command rather than
665
- a URL until someone types it.
688
+ decides whether it binds wide — on the LAN and the tailnet alike — with no
689
+ detection involved.
666
690
 
667
691
  - **Never fatal.** A failed render — no worktree, a git error — is one line and
668
692
  the phase is still done. The page is a convenience; the repo is the record.
@@ -684,7 +708,8 @@ the shape; this section carries only what is specific here.
684
708
  **Some failed** or **Not yet**: the phase stays `🔄`, `Notes` says which steps
685
709
  are open, and `Next` is `/commit, then /spec-next` to pick them up.
686
710
  - `❌` — the phase's tests are red, or it stopped part-way. Quote the failure.
687
- - `⏸` — no spec in flight, or the name given does not match the one that is.
711
+ - `⏸` — no spec in flight, several in flight with this session in none of
712
+ them, or the name given does not match the one that is.
688
713
 
689
714
  **Fields:** `Tracker` · `Branch` · `Built` · `Tests` · `Notes` · `Review` ·
690
715
  `Follow-ups` · `Next`