@skitterbyte/skitterspec-jira 2.1.0 → 2.3.0

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/MIGRATION.md CHANGED
@@ -1,5 +1,47 @@
1
1
  # Migration guide
2
2
 
3
+ ## `issue` and `incident` spec kinds, per-type states, incident log (additive — nothing breaks)
4
+
5
+ Specs can now be **issues** (`issue-` prefix — non-code work done as recorded
6
+ manual steps) and **incidents** (`incident-` prefix, captured live by the new
7
+ `/spec-incident` skill; `/spec-complete` ends one by writing `incident-log.md`
8
+ and posting it onto the tracker item as a comment, via the new
9
+ `spec-sync comment <spec> --file <f>` verb). A lifecycle bucket can map to a
10
+ different state per work item type (`statesByType` on Azure Boards and Jira)
11
+ or per spec kind (`statesByKind` on GitHub and Linear), both top-level beside
12
+ `states`, with `null` meaning "send no state" — and the pre-write check now
13
+ validates each type's effective states against **that type's own** list, so a
14
+ state only one type lacks refuses before anything is written. On Azure and
15
+ Jira, `spec-sync states --json` now prints a per-type map
16
+ (`{ "<type>": [names] }`); a flat array from an older file is still accepted
17
+ and checked as the union. A fix spec can carry
18
+ `> **Incident:** incident-<name>` (linked as related on push) and
19
+ `> **Severity:**` (`1 - Critical` … `4 - Low`). Configs without any of the new
20
+ keys behave exactly as before, and `skitterspec update` ships the new skill
21
+ and rules.
22
+
23
+ Consumer config, by provider:
24
+
25
+ - **Azure Boards** —
26
+ `azure.workItemTypeByKind: { "issue": "Issue", "incident": "Incident" }`,
27
+ plus
28
+ `statesByType: { "Issue": { "complete": "Closed" }, "Incident": { "backlog": null, "complete": "Closed" } }`.
29
+ Severity writes the built-in `Microsoft.VSTS.Common.Severity` — nothing to
30
+ configure; a type without the field skips it with a note. The incident link
31
+ is `System.LinkTypes.Related`.
32
+ - **GitHub** — `github.labelsByKind` accepts the new kinds;
33
+ `statesByKind: { "issue": { "complete": "Done" } }` routes a kind's board
34
+ Status (`null` sends none); `github.severityLabels: { "2": "sev-high" }`
35
+ maps the header's leading digit to a label. The incident rides in the
36
+ generated body as a `Related incident: #N` line.
37
+ - **Jira** — `jira.issueTypeByKind` accepts the new kinds; `statesByType`
38
+ keys by issue type; `jira.severityField` names the field to write (e.g.
39
+ `customfield_10900`). The incident link is a `Relates` issue link, added
40
+ once.
41
+ - **Linear** — `linear.labelsByKind` accepts the new kinds; `statesByKind`
42
+ keys by spec kind; `linear.severityLabels` maps digits to team labels. The
43
+ incident link is a related-issue relation.
44
+
3
45
  ## The `chore` spec kind, and per-type bodies (additive — nothing breaks)
4
46
 
5
47
  Specs can now be **chores** (`chore-` prefix, `> **Type:** Chore`) and every
@@ -13,6 +13,8 @@ NAMES — never a credential. Credentials live in the user-level store
13
13
  | `jira.subtaskType` | The type a phase mints as (must be a subtask type). | `"Subtask"` |
14
14
  | `mapping.specFolder` / `mapping.phases` / `mapping.tasks` | Spec → issue, phase → subtask, tasks → ADF checklist. Same modes as the Linear provider (`subissue` \| `deferred` \| `inline`; `checklist` \| `none`). | `issue` / `subissue` / `checklist` |
15
15
  | `states.<bucket>` | The Jira STATUS name for each lifecycle bucket. Must match the project's workflow statuses exactly; moves are applied by workflow transition, and an unreachable status refuses loudly. `cancelled` defaults to `Done` because default Jira workflows carry no Cancelled status. Pick work statuses, not pipeline stages — see **Choosing your status map** below. | see example |
16
+ | `statesByType.<issue type>.<bucket>` | Per-TYPE override laid over `states` — statuses belong to the type's workflow, so `{ "Service Request": { "complete": "Closed" } }` completes that type differently. A bucket set to `null` sends no status; validation checks each type against its own workflow. | `{}` |
17
+ | `jira.severityField` | The field id the `Severity:` header is written to (e.g. `customfield_10900`). Empty, the header stays repo-only. A fix spec's `Incident:` header adds a `Relates` issue link to its incident, once. | `""` |
16
18
  | `auth.emailEnv` / `auth.tokenEnv` | Env-var NAMES for the email + API-token pair (basic auth). | `JIRA_EMAIL` / `JIRA_API_TOKEN` |
17
19
  | `apply.transport` | `api` \| `mcp`, or empty for "api when credentials resolve, mcp when they don't". | `""` |
18
20
  | `sync.baseDir` | Where last-pushed snapshots live. | `specs/.core/jira-base` |
@@ -7,9 +7,10 @@ and lifecycle stay consistent. Each sets a status on the spec header
7
7
 
8
8
  | Skill | Purpose | Status | Folder |
9
9
  |-------|---------|--------|--------|
10
- | `/spec` | (Feature/Chore) Grill to a clear shared understanding, propose the kind, write a groomed spec in its own worktree, then land it on the verdict | `Ready` (or `Draft`) | `specs/backlog/` (on the base branch, by a fast-forward) |
10
+ | `/spec` | (Feature/Chore/Issue) Grill to a clear shared understanding, propose the kind, write a groomed spec in its own worktree, then land it on the verdict | `Ready` (or `Draft`) | `specs/backlog/` (on the base branch, by a fast-forward) |
11
11
  | `/spec-bug` | (Bug) Reproduce with a failing test, capture spec, drive red→green | `In Progress` | `specs/in-progress/` |
12
12
  | `/spec-hotfix` | (Hotfix) Fork a worktree from a release tag, red→green, land by tag + cherry-pick | `In Progress` | `specs/in-progress/` |
13
+ | `/spec-incident` | (Incident) Capture a live outage as a record written during it — grill, write in flight, fixes as separate linked specs | `In Progress` | `specs/in-progress/` |
13
14
  | `/no-spec` | Work that genuinely has no spec — its own branch, worktree, page and land. Writes no spec and moves nothing through the lifecycle | `—` | (none) |
14
15
  | `/spec-review` | Re-validate a spec against the codebase; refresh stale parts | `—` | (unchanged) |
15
16
  | `/spec-start` | Put a spec in flight — provision its branch, then build phase 1 | `In Progress` | `specs/in-progress/` |
@@ -587,26 +588,34 @@ none of them describes.
587
588
  - **Other rules specs must honour:** link the relevant `.claude/rules/*.md`
588
589
  (architecture, code style, testing, database, etc.) rather than restating them.
589
590
 
590
- ## Spec types — Feature, Chore, Bug, Hotfix
591
+ ## Spec types — Feature, Chore, Issue, Bug, Hotfix, Incident
591
592
 
592
- Every spec is one of four types, recorded **both** in the header and the filename:
593
+ Every spec is one of six types, recorded **both** in the header and the filename:
593
594
 
594
595
  - **Header field:** `> **Type:** Feature`, `> **Type:** Chore`,
595
- `> **Type:** Bug`, or `> **Type:** Hotfix` (authoritative, greppable:
596
+ `> **Type:** Issue`, `> **Type:** Bug`, `> **Type:** Hotfix`, or
597
+ `> **Type:** Incident` (authoritative, greppable:
596
598
  `grep -rl 'Type:.*Hotfix' specs/`).
597
599
  - **Filename prefix:** `feat-<name>` for features, `chore-<name>` for chores,
598
- `bug-<name>` for bugs, `hotfix-<name>` for hotfixes (visible in listings;
600
+ `issue-<name>` for issues, `bug-<name>` for bugs, `hotfix-<name>` for
601
+ hotfixes, `incident-<name>` for incidents (visible in listings;
599
602
  glob-safe — never use `[BUG]`/`[FEATURE]` brackets).
600
603
 
601
- A **Chore** is work no end user sees — tooling, CI, dependency bumps, internal
602
- refactors. The line: anything that is not work end users see or functions they
603
- use within the app is a chore; everything user-facing is a feature. There is no
604
- separate entry skill — `/spec` **proposes** feature or chore from what the
604
+ A **Chore** is code work no end user sees — tooling, CI, dependency bumps,
605
+ internal refactors. An **Issue** is work with **no code change at all** —
606
+ in-app configuration, a data fix, a support task — written as a spec of manual
607
+ steps so there is a record and a history; everything user-facing that changes
608
+ code is a feature. There is no
609
+ separate entry skill for either — `/spec` **proposes** feature, chore or issue
610
+ from what the
605
611
  grilling defined, with a one-line reason, and the user confirms. A chore
606
612
  defaults its `Gating:` to `none: not user-facing` (where gating is configured),
607
613
  and its commits default to `chore`/`refactor`/`ci` types with **no**
608
614
  `Release-Note:` footer — a chore is by definition not user-visible, so it has
609
- no release note to write. Feature specs keep `feat`; bug specs keep `fix`.
615
+ no release note to write. An issue defaults its `Gating:` to `none: no code`
616
+ and its commits to `docs` with no `Release-Note:`, and it closes when the spec
617
+ completes — nothing is released. Feature specs keep `feat`; bug specs keep
618
+ `fix`.
610
619
  Where a ticketing provider maps kinds to tracker types (e.g. Azure's
611
620
  `workItemTypeByKind`), the chore mapping is what keeps tooling work off the
612
621
  customer-facing release ladder.
@@ -616,14 +625,75 @@ carries an extra `> **Base version:** <tag>` header, forks its worktree from tha
616
625
  tag, and lands by tagging a new patch + cherry-picking onto `main` (never a
617
626
  fast-forward merge). `/spec-live` refuses a hotfix — test it with `/spec-connect`.
618
627
 
619
- All four types share the same lifecycle folders below — type is orthogonal to
628
+ An **Incident** is an outage or degradation of a live service, captured by
629
+ **`/spec-incident`** during the outage, so it starts straight in
630
+ `specs/in-progress/`. It is a record, not delivery work: its fixes are
631
+ separate hotfix, bug, chore or issue specs, each carrying
632
+ `> **Incident:** incident-<name>` so the push links them, and they are never
633
+ its phases. It carries `> **Severity:**` (Azure's literal values,
634
+ `1 - Critical` … `4 - Low`) and a `## Timeline` appended to as the incident
635
+ unfolds; `/spec-complete` writes `incident-log.md` from it and pushes the log
636
+ onto the tracker item as a comment.
637
+
638
+ ### Kind routing — any door, the right spec
639
+
640
+ `/spec` is the default door, so the kind of work someone describes does not
641
+ always match the skill they typed — especially for someone new, for whom
642
+ "spec it" is the reflex. Every entry skill therefore
643
+ **watches its grill for the signals that mark the other kinds**, and proposes a
644
+ hand-off the moment one appears — never at the end of a grill already shaping
645
+ the wrong spec:
646
+
647
+ | In | Watches for | Hands to |
648
+ |----|-------------|----------|
649
+ | `/spec` | defect · defect on a released version · live outage | `/spec-bug` · `/spec-hotfix` · `/spec-incident` |
650
+ | `/spec-bug` | on a released version · live outage · not a defect | `/spec-hotfix` · `/spec-incident` · `/spec` |
651
+ | `/spec-hotfix` | live outage · not on a released version · not a defect | `/spec-incident` · `/spec-bug` · `/spec` |
652
+ | `/spec-incident` | no live outage to record | defect → `/spec-bug` (on a released version → `/spec-hotfix`), else `/spec` |
653
+
654
+ Feature, chore and issue all route to `/spec` — one door for the three kinds it
655
+ writes, where its kind proposal picks between them. Example signals, weighed in
656
+ context rather than matched as keywords: a **defect** is behaviour differing
657
+ from intended — "it's broken", a failing flow, a stack trace of a failure;
658
+ **on a released version** is that defect where what is deployed is named — "in
659
+ prod", "since v2.1"; a **live outage** is a service down or degraded
660
+ **right now**, users affected as we speak; **not a defect** is an ask for new or
661
+ changed behaviour rather than for restoring intended behaviour.
662
+
663
+ Three rules, the same in every skill:
664
+
665
+ - **Always confirm; never silently switch.** Name the signal in the user's own
666
+ words and propose the destination with a one-line reason; the user confirms
667
+ or declines. A decline ends the matter — the grill carries on where it was,
668
+ and the same signal is not raised again. A signal is a question, never a
669
+ verdict: a pasted stack trace in "handle this error nicely" is a feature
670
+ request, not a bug report.
671
+ - **Precedence: incident > hotfix > bug.** One description can carry several
672
+ signals; propose the most specific destination — the same reasoning the
673
+ intake label routing documents for hotfix-over-bug.
674
+ - **Hand off in-session, carrying the context.** On a confirmed switch the
675
+ current skill invokes the destination skill directly and seeds it with
676
+ everything the grilling has already established — symptom, version, impact,
677
+ repro material, decisions so far — so the destination grills only what it
678
+ still needs ("seed, don't skip, the grill", the principle issue adoption
679
+ already uses). Stop-and-retype is the **label** routing's shape, and right
680
+ there because it fires before any grilling; mid-grill it would throw away
681
+ the answers already given.
682
+
683
+ This is judgment, not configuration: the signals are examples, there is no
684
+ list to maintain, and nothing here reads a label. A ticketing provider's
685
+ intake label routing (`intake.bugLabels` / `intake.hotfixLabels`) is a
686
+ separate, earlier check and is unchanged by this.
687
+
688
+ All six types share the same lifecycle folders below — type is orthogonal to
620
689
  status.
621
690
 
622
691
  ## Header fields & State log (audit trail)
623
692
 
624
693
  Every spec header carries:
625
694
 
626
- - `> **Name:**` — the spec's folder name (`feat-`/`bug-`/`hotfix-<kebab-name>`).
695
+ - `> **Name:**` — the spec's folder name
696
+ (`feat-`/`chore-`/`issue-`/`bug-`/`hotfix-`/`incident-<kebab-name>`).
627
697
  It's the handle you pass to `/spec-start` and the other lifecycle skills, surfaced
628
698
  in the header so it's copy-pasteable without digging for the folder name.
629
699
  - `> **Author:**` — who created the spec (set at `/spec` / `/spec-bug`, defaults
@@ -634,6 +704,14 @@ Every spec header carries:
634
704
  **only when `specs/.core/gating.config.json` exists**: a flag name, or
635
705
  `none: <one-line reason>`. Absent entirely in a project that has not adopted
636
706
  gating.
707
+ - `> **Severity:**` — on bug, hotfix and incident specs, optional except where
708
+ `/spec-incident` grills it: Azure's literal values
709
+ (`1 - Critical` … `4 - Low`). The push writes it where the provider has a
710
+ home for it; absent, nothing is sent.
711
+ - `> **Incident:**` — on a spec that is a **fix** for an open incident:
712
+ `incident-<name>`. The push links the two as related on the tracker
713
+ (one-way; the repo owns the link), and `/spec-complete` on the incident
714
+ counts the spec among its fixes. Absent on everything else.
637
715
 
638
716
  Every spec also has a **State log** table — the audit trail of folder/status
639
717
  transitions. Each lifecycle skill appends exactly one row when it changes state:
@@ -668,7 +746,7 @@ When asked for a plan, implementation strategy, or feature breakdown:
668
746
 
669
747
  ```
670
748
  specs/backlog/ Ready (or Draft) specs (/spec)
671
- specs/in-progress/ under active implementation (/spec-start, /spec-bug)
749
+ specs/in-progress/ under active implementation (/spec-start, /spec-bug, /spec-incident)
672
750
  specs/complete/ finished (/spec-complete)
673
751
  specs/cancelled/ abandoned, with a reason on the header (/spec-cancel)
674
752
  specs/.core/ project rules — ALWAYS APPLY, never moved
@@ -188,31 +188,78 @@ not write the spec until this is resolved.
188
188
  Stop grilling when there are no unresolved branches that would change the spec.
189
189
  Briefly play back the agreed understanding before writing.
190
190
 
191
- ### Propose the kind — feature or chore
192
-
193
- This skill writes two kinds of spec, and
191
+ ### Watch for the other doors — bug, hotfix, incident
192
+
193
+ `/spec` is the default door, so work that belongs to another skill arrives here
194
+ described in the user's own words — no issue, no labels for Phase 0 to route
195
+ on. While grilling, watch for the signals that mark the other kinds and raise
196
+ the hand-off **the moment one appears**, not at the end of the grill: for a
197
+ live outage especially, finishing a feature grill first is actively harmful,
198
+ because `/spec-incident` exists to be written *during* the outage.
199
+
200
+ - **Defect** — behaviour differing from intended ("it's broken", a failing
201
+ flow, a stack trace of a failure) → propose **`/spec-bug`**.
202
+ - **Defect on a released version** — the defect is on what is deployed ("in
203
+ prod", "since v2.1") → propose **`/spec-hotfix`**.
204
+ - **Live outage** — a service down or degraded **right now**, users affected
205
+ as we speak → propose **`/spec-incident`**.
206
+
207
+ **Always confirm; never silently switch.** Name the signal in the user's own
208
+ words, propose the destination with a one-line reason, and let them decide. A
209
+ decline carries the grill on where it was, and the same signal is not raised
210
+ again. A signal is a question, never a verdict: a pasted stack trace in
211
+ "handle this error nicely" is a feature request. When one description carries
212
+ several signals, propose the most specific destination —
213
+ **incident > hotfix > bug**.
214
+
215
+ **On a confirmed switch, hand off in-session and carry the context.** Invoke
216
+ the destination skill directly and seed it with everything the grilling has
217
+ already established — symptom, version, impact, repro material, decisions so
218
+ far — so it grills only what it still needs ("seed, don't skip, the grill",
219
+ exactly as issue adoption does). Do not stop and tell the user to retype: that
220
+ shape belongs to Phase 0's label routing, which fires before any grilling has
221
+ happened and so has nothing to lose. The full routing matrix lives in
222
+ `.claude/rules/spec-planning.md`.
223
+
224
+ ### Propose the kind — feature, chore or issue
225
+
226
+ This skill writes three kinds of spec, and
194
227
  **you propose which from what the grilling defined, with a one-line reason; the user confirms or overrides.**
195
228
  The
196
- line that decides it: a **chore** is anything that is not work end users see or
197
- functions they use within the app — tooling, CI, dependency bumps, internal
198
- refactors. Everything user-facing is a **feature**. So the proposal reads like
199
- *"no user-visible surface → chore"* or *"adds an endpoint users call → feature"*.
229
+ lines that decide it: a **chore** is code work no end user sees — tooling, CI,
230
+ dependency bumps, internal refactors. An **issue** is work with
231
+ **no code change at all** — in-app configuration, a data fix, a support task —
232
+ written as a spec of manual steps so there is a record and a history.
233
+ Everything user-facing that changes code is a **feature**. So the proposal
234
+ reads like *"no user-visible surface → chore"*, *"no code change, manual
235
+ steps → issue"* or *"adds an endpoint users call → feature"*.
200
236
 
201
237
  The kind is recorded twice, and both follow from the answer: the folder prefix
202
- (`feat-` / `chore-`) and the `> **Type:**` header (`Feature` / `Chore`). A chore
238
+ (`feat-` / `chore-` / `issue-`) and the `> **Type:**` header
239
+ (`Feature` / `Chore` / `Issue`). A chore
203
240
  also defaults its `Gating:` to `none: not user-facing` (only when gating is
204
241
  configured — offer the default, let the user override it like any other
205
242
  answer), and its commits default to `chore`/`refactor`/`ci` types with no
206
- `Release-Note:` (see `.claude/rules/spec-planning.md`).
207
-
208
- There is no `/spec-chore` and none should be added: one entry point, one grill,
209
- and the kind is an output of understanding the work rather than a door the user
210
- must pick before describing it.
243
+ `Release-Note:` (see `.claude/rules/spec-planning.md`). An issue defaults its
244
+ `Gating:` to `none: no code`, and its commits to `docs` with no
245
+ `Release-Note:` — its phases are manual steps, not code, though `/spec-start`
246
+ still gives it a worktree, because the document needs one and the history is
247
+ the point.
248
+
249
+ There is no `/spec-chore` or `/spec-issue` and none should be added: one entry
250
+ point, one grill, and the kind is an output of understanding the work rather
251
+ than a door the user must pick before describing it. The kinds with their own
252
+ doors — bug, hotfix, and the **incident**, an outage or degradation of a live
253
+ service that needs a different grill and starts mid-flight — are the watch's
254
+ business above: when the grilling uncovers one, hand off to **`/spec-bug`**,
255
+ **`/spec-hotfix`** or **`/spec-incident`** rather than writing it here, and it
256
+ is never too late to propose that — reaching this proposal with the signal
257
+ unraised is a missed watch, not a reason to write the wrong kind.
211
258
 
212
259
  ## Phase B — write the spec
213
260
 
214
- This skill is for **features and chores**. For bugs, use `/spec-bug`
215
- (test-first, red→green).
261
+ This skill is for **features, chores and issues**. For bugs, use `/spec-bug`
262
+ (test-first, red→green); for a live outage, `/spec-incident`.
216
263
 
217
264
  ### Write it in its own worktree
218
265
 
@@ -280,7 +327,8 @@ move to, **warn, don't refuse**, and say the trailer for that commit wants
280
327
  notes. **Even a single-phase spec gets `01-….md`** — never lump phase tasks
281
328
  into `00-overview.md`. This keeps each phase easy to dive into on its own.
282
329
  - Choose a short kebab-case name and **prefix it with the kind** — `feat-` for
283
- a feature, `chore-` for a chore (the bug counterpart uses `bug-`).
330
+ a feature, `chore-` for a chore, `issue-` for an issue (the bug counterpart
331
+ uses `bug-`, the incident one `incident-`).
284
332
 
285
333
  Use this template (keep it **as concise as possible** — no filler, no restating
286
334
  the codebase, link rather than duplicate):
@@ -288,8 +336,8 @@ the codebase, link rather than duplicate):
288
336
  ```markdown
289
337
  # <Feature title>
290
338
 
291
- > **Type:** <Feature or Chore — the kind proposed after Phase A>
292
- > **Name:** <feat- or chore-><kebab-name> (the spec folder name — the handle you paste into `/spec-start`)
339
+ > **Type:** <Feature, Chore or Issue — the kind proposed after Phase A>
340
+ > **Name:** <feat-, chore- or issue-><kebab-name> (the spec folder name — the handle you paste into `/spec-start`)
293
341
  > **Status:** Ready — not started
294
342
  > **Author:** <git user.name — `git config user.name`>
295
343
  > **Developer:** —
@@ -297,6 +345,10 @@ the codebase, link rather than duplicate):
297
345
  > **Area:** <comma-separated files/modules this touches>
298
346
  > **Stack:** <worktree — or "worktree + docker" if it touches the DB/stateful
299
347
  > services; only acted on when isolation is enabled — see Phase A item 9>
348
+ > **Incident:** <incident-<name> — only when this spec is a fix for an open
349
+ > incident spec; the next push links the two as related on the tracker, and
350
+ > `/spec-complete` on the incident counts this spec among its fixes. Omit the
351
+ > line entirely otherwise>
300
352
  > **Gating:** <flag name — or "none: <one-line reason>". Only when release gating
301
353
  > is configured; omit the line entirely otherwise. A chore defaults to
302
354
  > "none: not user-facing". An empty value or a bare
@@ -150,6 +150,38 @@ Bugs are concrete — confirm, don't over-grill. Establish:
150
150
  against the broken one (the bug usually lives in the differential). Do NOT
151
151
  patch a symptom before you understand the cause.
152
152
 
153
+ ### Watch for the other doors — hotfix, incident, or not a bug at all
154
+
155
+ The investigation above is where mis-entry shows, so watch it for the signals
156
+ that mark the other kinds and raise the hand-off the moment one appears —
157
+ before the failing test is written, while nothing is on disk
158
+ (`.claude/rules/spec-planning.md` carries the full routing matrix):
159
+
160
+ - **On a released version** — the defect is on what is deployed ("in prod",
161
+ "since v2.1") → propose **`/spec-hotfix`**. This is the escalation that must
162
+ not be missed: a fix built here lands on `main` and never reaches the
163
+ running version.
164
+ - **Live outage** — a service down or degraded **right now** → propose
165
+ **`/spec-incident`**; the record comes first, and the fix returns as a
166
+ linked spec.
167
+ - **Not a defect** — the ask is new or changed behaviour, not restoring
168
+ intended behaviour → propose **`/spec`**, whose kind proposal picks feature,
169
+ chore or issue from there.
170
+
171
+ **Always confirm; never silently switch** — name the signal in the user's own
172
+ words with a one-line reason, and a decline carries this skill on where it
173
+ was, the same signal not raised again. When several signals appear at once,
174
+ propose the most specific destination — incident over hotfix.
175
+ **On a confirmed switch, hand off in-session and carry the context**: invoke
176
+ the destination skill directly and seed it with what §1 established — symptom,
177
+ repro, root-cause notes, any version named — so it grills only what it still
178
+ needs.
179
+
180
+ **Once the failing test exists, the repro is evidence, not a question to re-litigate**
181
+ — but evidence can still escalate: a red test that also fails at
182
+ the released tag is the hotfix signal with proof attached, and the hand-off
183
+ carries the failing test across.
184
+
153
185
  ## 2. Isolate the fix in a worktree — when isolation is enabled
154
186
 
155
187
  **Only when per-spec isolation is enabled** (`specs/.core/env.config.json`
@@ -269,6 +301,10 @@ because work is already underway. Keep it lean:
269
301
  > **Developer:** <git user.name — you, since you're fixing it now>
270
302
  > **Raised:** <YYYY-MM-DD (today)>
271
303
  > **Area:** <files/modules>
304
+ > **Severity:** <Azure's literal values, "1 - Critical" … "4 - Low" — optional;
305
+ > the push writes it where the provider has a home for it. Omit otherwise>
306
+ > **Incident:** <incident-<name> — only when this bug is a fix for an open
307
+ > incident spec; the next push links the two as related. Omit otherwise>
272
308
  > **Gating:** <flag name — or "none: <one-line reason>". Only when release gating
273
309
  > is configured; omit the line entirely otherwise>
274
310
 
@@ -42,6 +42,45 @@ Before marking complete, confirm the work is actually finished:
42
42
  is its own, and step 4 commits that.
43
43
 
44
44
 
45
+ ### An incident spec — the log, and the open-fix check
46
+
47
+ **Only for `Type: Incident` specs** (folder prefix `incident-`); every other
48
+ kind skips this in silence.
49
+
50
+ 1. **Check for open linked fixes.** Grep `specs/backlog/` and
51
+ `specs/in-progress/` for specs carrying `> **Incident:** <this spec's name>`.
52
+ Any hit is a fix still open: **refuse**, naming each one, and offer to
53
+ complete anyway with it listed as outstanding in the log — the operator
54
+ answers; record the choice in the spec's Changelog either way.
55
+ BLIND SPOT, named: a fix spec authored but not yet landed on the base branch
56
+ is invisible to this grep — acceptable, because `/spec`'s committing
57
+ verdicts land authored specs on the base
58
+ (`.claude/rules/negative-checks.md`).
59
+ 2. **Write `incident-log.md`** into the spec folder — the stakeholder- and
60
+ support-facing record, built from the spec and its linked fixes:
61
+ summary · severity · timeline (started, detected, mitigated, resolved) ·
62
+ impact (who, what, for how long) · root cause · the fixes (each linked
63
+ spec, its ticket, and the release it shipped in — `spec-sync released`
64
+ answers that where a provider is installed; a fix completed but not yet
65
+ released is listed as such) · what went well and what didn't · follow-ups,
66
+ each offered as a `/spec` on the spot.
67
+ 3. **Push the log onto the incident item as a comment** (only when a provider
68
+ is installed and the spec is linked):
69
+
70
+ ```
71
+ skitterspec spec-sync comment <spec> --file specs/in-progress/<spec>/incident-log.md
72
+ ```
73
+
74
+ A comment, deliberately: the item's description is generated from the spec,
75
+ so a log written there would be clobbered by the next push — comments are
76
+ the one surface one-way sync never touches. The command exits 0 on every
77
+ cannot-tell (unlinked, offline, no credentials) and says so; relay the line
78
+ and finish either way — the log in the repo is the record.
79
+
80
+ Completion then proceeds as normal: the per-type state map decides what the
81
+ tracker is sent (an Incident completes to whatever `statesByType` names for
82
+ it), and steps 3–7 below are unchanged.
83
+
45
84
  ### Release gating (only when configured)
46
85
 
47
86
  **Only when `specs/.core/gating.config.json` exists.** Run
@@ -177,6 +177,27 @@ Hotfixes are concrete — confirm, don't over-grill. Establish:
177
177
  - **Root cause:** read the code **at the base tag** and trace it to `file:line`.
178
178
  The fix belongs on the tag's line, so reason about that code, not `main`'s.
179
179
 
180
+ ### Watch for the other doors — incident, bug, or not a defect
181
+
182
+ This is the most specific defect door, so most mis-entries route *down*
183
+ (`.claude/rules/spec-planning.md` carries the full routing matrix):
184
+
185
+ - **Live outage** — the released version is not just wrong but down or
186
+ degraded **right now** → propose **`/spec-incident`**; the record comes
187
+ first, and the fix returns here as a linked hotfix spec.
188
+ - **Not on a released version** — §1 cannot name a deployed tag the defect is
189
+ on (it exists only on `main`, or prod runs something newer that is fine) →
190
+ propose **`/spec-bug`**; a hotfix branch for something `main` can carry is
191
+ waste, not safety.
192
+ - **Not a defect** — the ask is new or changed behaviour, not restoring
193
+ intended behaviour → propose **`/spec`**.
194
+
195
+ **Always confirm; never silently switch** — name the signal in the user's own
196
+ words with a one-line reason; a decline carries this skill on where it was.
197
+ **On a confirmed switch, hand off in-session and carry the context**
198
+ established so far — the symptom, the repro, any version discussion — so the
199
+ destination grills only what it still needs.
200
+
180
201
  ## 3. Provision the worktree from the tag, then write in it
181
202
 
182
203
  **Provision before you write, not after.** §1 settled the tag and the name and
@@ -293,6 +314,10 @@ is usually a single-pass fix, so the `## Fix` block can live directly in
293
314
  > **Base version:** <tag prod is running, e.g. v33.16.4>
294
315
  > **Raised:** <YYYY-MM-DD (today)>
295
316
  > **Area:** <files/modules>
317
+ > **Severity:** <Azure's literal values, "1 - Critical" … "4 - Low" — optional;
318
+ > the push writes it where the provider has a home for it. Omit otherwise>
319
+ > **Incident:** <incident-<name> — only when this hotfix is a fix for an open
320
+ > incident spec; the next push links the two as related. Omit otherwise>
296
321
  > **Gating:** <pre-filled "none: hotfix — restoring released behaviour"; only
297
322
  > when release gating is configured, and overridable — see below>
298
323
 
@@ -0,0 +1,164 @@
1
+ ---
2
+ name: spec-incident
3
+ description: Capture a live incident — an outage or degradation of a running service — as an Incident-type spec written during the outage. Grills for what is down, since when, who is affected, severity and what was tried, then writes the spec straight into specs/in-progress/ with a Timeline appended to as it unfolds. Fixes are separate linked specs, never phases. Use when the user says "/spec-incident", "we have an outage", "prod is down and I need an incident record", or asks to track an incident.
4
+ ---
5
+
6
+ # /spec-incident — capture a live incident
7
+
8
+ > Stay silent while this runs — speak only to ask something you cannot answer
9
+ > yourself, or to report a failure at the moment it happens. Read
10
+ > `.claude/rules/spec-reports.md` before reporting; it defines the block this
11
+ > skill ends with.
12
+
13
+ An **incident** is an outage or degradation of a live service. It is not
14
+ delivery work: the fixes are separate hotfix, bug, chore or issue specs, each
15
+ linked back to it, and this spec is the **record** — what happened, when, to
16
+ whom, and what was done. It is written **during** the outage, so it starts in
17
+ `specs/in-progress/`, there is no backlog stop and no review page to wait on:
18
+ speed matters, and the record is reviewed at completion, when
19
+ `/spec-complete` writes the incident log from it.
20
+
21
+ **This skill is not the response.** Mitigating comes first, and nothing here
22
+ blocks it — the grill is short, the spec is small, and everything else is
23
+ appended as it happens.
24
+
25
+ ## 1. Grill — short, and only what the record needs
26
+
27
+ Ask for exactly these, batched, with nothing speculative:
28
+
29
+ 1. **What is down or degraded** — the service, and how it is failing.
30
+ 2. **Since when** — first report, and detection time where they differ.
31
+ 3. **Who is affected** — which users or systems, and how badly.
32
+ 4. **Severity** — Azure's literal values: `1 - Critical`, `2 - High`,
33
+ `3 - Medium`, `4 - Low`. Propose one from the impact described; the user
34
+ confirms.
35
+ 5. **What has been tried** — anything already attempted, and its effect.
36
+
37
+ A root cause is **not** grilled: nobody has it yet, and the Timeline is where
38
+ the investigation lands as it happens.
39
+
40
+ ### No live outage to record — route to the right door
41
+
42
+ This is the costliest door to enter wrongly — no backlog stop, no review page,
43
+ straight into `specs/in-progress/` — so the grill's own answers are also its
44
+ exit. When items 1 and 2 establish there is **no outage to record** — nothing
45
+ is down or degraded, it never was live, or it is a risk rather than an event —
46
+ propose the right door instead of writing an incident
47
+ (`.claude/rules/spec-planning.md` carries the full routing matrix):
48
+
49
+ - a **defect** → **`/spec-bug`**, or **`/spec-hotfix`** when it is on a
50
+ released version;
51
+ - anything else → **`/spec`**, whose kind proposal picks feature, chore or
52
+ issue from there.
53
+
54
+ **A past outage someone wants a record of stays an incident**: "it is over" is
55
+ not "it never happened" — the signal is *no outage to record*, not *no outage
56
+ now*. This check is the grill's existing questions read honestly, never an
57
+ extra round trip, and the hand-off contract is the same as everywhere else:
58
+ confirm first, naming the signal in the user's own words; a decline carries
59
+ this skill on; on a confirmed switch invoke the destination skill in-session,
60
+ seeded with everything already established.
61
+
62
+ ## 2. Write it, in flight
63
+
64
+ Pick a short kebab name prefixed `incident-` (e.g. `incident-api-outage`).
65
+
66
+ **With per-spec isolation** (`specs/.core/env.config.json` present), provision
67
+ before writing, exactly as `/spec` Phase B does: run
68
+ `skitterspec spec-env up <name> --docs`, run the commands it prints, `cd` into
69
+ the worktree, and confirm with a bare `skitterspec spec-env resolve` naming
70
+ this spec. Without isolation, write where you stand.
71
+
72
+ The spec is a folder in **`specs/in-progress/`** — never backlog — with
73
+ `00-overview.md` and one phase file:
74
+
75
+ ```markdown
76
+ # Incident: <short title>
77
+
78
+ > **Type:** Incident
79
+ > **Name:** incident-<kebab-name>
80
+ > **Status:** In Progress — live (started <YYYY-MM-DD>)
81
+ > **Author:** <git user.name>
82
+ > **Developer:** <git user.name — whoever is running the incident>
83
+ > **Raised:** <YYYY-MM-DD (today)>
84
+ > **Severity:** <1 - Critical … 4 - Low>
85
+ > **Gating:** <pre-filled "none: incident record, no code"; only when release
86
+ > gating is configured, omit otherwise>
87
+
88
+ ## Impact
89
+
90
+ <who is affected, what is failing, since when — from the grill>
91
+
92
+ ## Timeline
93
+
94
+ - <YYYY-MM-DD HH:MM> — <first report / detected / what was tried / what changed>
95
+
96
+ ## Fixes
97
+
98
+ <one line per fix spec as they are opened — see step 4. "None yet" until then.>
99
+
100
+ ## State log
101
+
102
+ | Date | Status | Folder | By |
103
+ |------|--------|--------|----|
104
+ | <YYYY-MM-DD> | In Progress | in-progress | <author> |
105
+
106
+ ## Changelog
107
+
108
+ - <YYYY-MM-DD> — Incident captured.
109
+ ```
110
+
111
+ The phase file `01-response.md` holds the response as tasks — mitigate,
112
+ diagnose, open fix specs, verify recovery — manual steps, not code. Commits in
113
+ this spec default to `docs` with no `Release-Note:`.
114
+
115
+ ## 3. Link to the tracker (only if a provider is installed)
116
+
117
+ Run the provider's linking push exactly as `/spec` Phase E describes — the
118
+ spec is already in `in-progress`, so the mint lands in the incident's
119
+ in-flight state (with Azure's `statesByType`, an Incident is created Active
120
+ and never sent New; the `Severity:` header is written where the provider has a
121
+ home for it). No provider, or an unreachable tracker, is one line and the
122
+ record stands in the repo.
123
+
124
+ ## 4. Run the incident from the spec
125
+
126
+ - **Append to the Timeline** as things happen — dated lines, cheap and often.
127
+ - **Fixes are separate specs**: a prod-version fix is `/spec-hotfix`, a
128
+ mainline defect `/spec-bug`, tooling `/spec` (chore), manual remediation
129
+ `/spec` (issue). Each fix spec carries `> **Incident:** incident-<name>` in
130
+ its header, which the next push turns into a related link on the tracker.
131
+ They are never phases of this spec.
132
+ - **Mid-incident board states stay manual**: skitterspec writes states at
133
+ lifecycle moves only, so moving the ticket to a Mitigated-style state during
134
+ the outage is done on the board, and the next push leaves it until the spec
135
+ itself moves.
136
+
137
+ ## 5. End of the outage is not the end of the spec
138
+
139
+ When the service is recovered, say so in the Timeline and keep the spec in
140
+ flight until the record is complete — root cause written, fixes linked.
141
+ `/spec-complete` is the ending: it writes `incident-log.md` from this spec and
142
+ its linked fixes, refuses while a linked fix is still open (and offers to
143
+ complete anyway with it listed as outstanding), and pushes the log onto the
144
+ incident item as a comment.
145
+
146
+ ## Report
147
+
148
+ End with the block defined in `.claude/rules/spec-reports.md`. That file
149
+ carries the shape; this section carries only what is specific here.
150
+
151
+ **Verdicts**
152
+
153
+ - `✅` — the incident spec is written, in flight, and (where a provider is
154
+ installed) linked.
155
+ - `⚠️` — written and in flight, with something worth knowing (a mint that
156
+ failed, a worktree that could not be provisioned so it was written in
157
+ place).
158
+ - `⏸` — the grill could not establish what is down, so nothing was written.
159
+
160
+ **Fields:** `Tracker` · `Branch` · `Spec` · `Built` · `Worktree` ·
161
+ `Follow-ups` · `Next`
162
+
163
+ `Next` is the response itself — the first task of `01-response.md` — never a
164
+ review page: nothing here waits on a verdict while a service is down.