joycraft 1.1.2 → 1.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (37) hide show
  1. package/README.md +15 -10
  2. package/dist/joycraft-release.json +1 -1
  3. package/package.json +3 -3
  4. package/payload/.agents/skills/joycraft-decompose/SKILL.md +5 -1
  5. package/payload/.agents/skills/joycraft-entry/SKILL.md +28 -7
  6. package/payload/.agents/skills/joycraft-implement/SKILL.md +16 -0
  7. package/payload/.agents/skills/joycraft-session-end/SKILL.md +1 -1
  8. package/payload/.agents/skills/joycraft-testing/SKILL.md +7 -5
  9. package/payload/.agents/skills/joycraft-ui-slice/SKILL.md +4 -3
  10. package/payload/.agents/skills/joycraft-verify/SKILL.md +1 -1
  11. package/payload/.claude/agents/joycraft-reviewer.md +7 -2
  12. package/payload/.joycraft/assets/artifacts/kit.css +7 -0
  13. package/payload/.joycraft/assets/artifacts/kit.js +23 -7
  14. package/payload/.joycraft/contracts/types.d.ts +22 -0
  15. package/payload/.joycraft/references/archive.md +4 -2
  16. package/payload/.joycraft/references/artifacts.md +66 -9
  17. package/payload/.joycraft/references/completion.md +55 -2
  18. package/payload/.joycraft/references/dispatch.md +12 -0
  19. package/payload/.joycraft/references/setup.md +70 -30
  20. package/payload/.joycraft/references/testing.md +14 -7
  21. package/payload/.joycraft/references/update.md +8 -0
  22. package/payload/.joycraft/references/workflow.md +10 -5
  23. package/payload/.joycraft/runtime/archive.mjs +19 -4
  24. package/payload/.joycraft/runtime/artifacts/cli.mjs +29 -5
  25. package/payload/.joycraft/runtime/artifacts/decisions.mjs +13 -5
  26. package/payload/.joycraft/runtime/artifacts/model.mjs +1 -1
  27. package/payload/.joycraft/runtime/artifacts/preferences.mjs +95 -2
  28. package/payload/.joycraft/runtime/completion.mjs +210 -10
  29. package/payload/.joycraft/runtime/dispatch-preflight.mjs +16 -3
  30. package/payload/.joycraft/runtime/evidence.mjs +4 -0
  31. package/payload/.joycraft/runtime/identity.mjs +11 -0
  32. package/payload/.joycraft/runtime/review-tools.json +110 -30
  33. package/payload/.joycraft/runtime/review-tools.mjs +658 -129
  34. package/payload/.joycraft/runtime/status.mjs +1 -1
  35. package/payload/.joycraft/runtime/worktrees.mjs +1 -1
  36. package/payload/.joycraft/templates/brief.md +1 -1
  37. package/payload/PIN +35 -35
package/README.md CHANGED
@@ -165,20 +165,24 @@ the agent says what it does not cover and installs nothing without your yes.
165
165
 
166
166
  ## Answering questions: the decision page
167
167
 
168
- When the agent needs you to choose between options, it shows the questions on a
169
- decision page instead of in chat. Each question shows its context, its options and
170
- what it affects. You pick an answer, or "Something else" and say what you want, add
171
- a note if you want, and save. You can add or change a note after saving. If the
172
- brief changes while the page is open, the page names the questions that changed and
173
- keeps your other answers. The answers go into the brief, so the agent does not ask
174
- again.
175
-
176
- Sometimes the agent forgets and lists the options in chat. Then say:
168
+ When the agent needs you to choose between options, it asks in your question
169
+ style. Setup asks for it once: one decision at a time with a short talk about each
170
+ answer, or a page of decisions; on a decision page or as numbered prose in chat.
171
+ To change it for one feature, say:
177
172
 
178
173
  ```text
179
- Put these questions in a Joycraft decision artifact so I can answer them there.
174
+ Ask me one question at a time, in chat, for this feature.
180
175
  ```
181
176
 
177
+ A consequential question gives the context, the question, two to four options each
178
+ with an example and a trade-off, and the agent's recommendation with its reason. On
179
+ the decision page, you pick an answer, or "Something else" and say what you want,
180
+ add a note if you want, and save. You can add or change a note after saving. If the
181
+ brief changes while the page is open, the page names the questions that changed and
182
+ keeps your other answers. In chat or on the page, the answers go into the brief, so
183
+ the agent does not ask again. If a question is not clear, say "wait, what?" and the
184
+ agent explains it again with more context.
185
+
182
186
  **Colors.** The page uses a light or dark theme. To match your project, ask:
183
187
 
184
188
  ```text
@@ -263,6 +267,7 @@ should forget:
263
267
  - **Lauren** (Cursor): prefer fixes in this order: architecture, then lint and CI, then skills and rules, then asking a human.
264
268
  - **Robert C. Martin**: impose values, not disciplines.
265
269
  - **aka-luan** ([doc-cleanup](https://github.com/aka-luan/doc-cleanup), MIT), brought to Joycraft by **David Moshal**: Joycraft's doc audit follows its method. Treat every doc claim as a hypothesis and the code as the evidence.
270
+ - **David Moshal**, with his `questions-dm` skill: ask one question at a time, give the context first, and never assume the user remembers what the agent remembers. **Matt Pocock** ([`wait-what`](https://github.com/mattpocock/skills/tree/main/skills/productivity/wait-what), MIT): when a message does not land, say it again in plain words with the missing context.
266
271
 
267
272
  The levels this project aims at are Dan Shapiro's. See [The levels](docs/guides/levels.md).
268
273
 
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "schemaVersion": 1,
3
- "releaseVersion": "1.1.2",
3
+ "releaseVersion": "1.2.1",
4
4
  "manifestSchemas": [
5
5
  1
6
6
  ],
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "joycraft",
3
- "version": "1.1.2",
3
+ "version": "1.2.1",
4
4
  "description": "Install and update Joycraft, a skill bundle and runtime for agent-led software work, in any project",
5
5
  "type": "module",
6
6
  "bin": {
@@ -13,8 +13,8 @@
13
13
  "payload"
14
14
  ],
15
15
  "joycraft": {
16
- "release": "1.1.2",
17
- "summary": "After install, your agent builds Joycraft's context map with you and keeps it current; init, update and doctor report an unwired project; cleanup follows doc-cleanup; the update prompt lists every release since yours.",
16
+ "release": "1.2.1",
17
+ "summary": "Joycraft asks questions your way: one at a time or as a page, in chat or on the decision page, set at setup and changeable per feature; each question gives context, options with an example and a trade-off, and a recommendation.",
18
18
  "autoSafeEligible": false
19
19
  },
20
20
  "scripts": {
@@ -18,7 +18,11 @@ decision status or ask for approval again when the session already supplies it.
18
18
  Use `.joycraft/templates/spec.md` and `queue.json` as small starting
19
19
  points. Keep each spec independently testable and put requirement text in stable
20
20
  brief/spec anchors. Each queued criterion points to its source and declares the
21
- needed runtime, visual or structural evidence. A check that only a deployed
21
+ needed runtime, visual or structural evidence. Write each criterion so the
22
+ package's recorded review tools (the `Review loop` section of `AGENTS.md`) can
23
+ check it within their `canShow`. A main claim those tools cannot show at all
24
+ goes back to the brief as a tooling-change question, asked once in the
25
+ interview and recorded; it does not enter the queue. A check that only a deployed
22
26
  environment can show is not a criterion; list it under the spec's "Deferred until
23
27
  deploy" section instead. Include the brief, spec and exact
24
28
  relevant code, test/configuration and fixture paths as verification inputs. Plan
@@ -70,13 +70,34 @@ Settle each goal's "Done when" with the user while writing the brief; if it
70
70
  cannot be filled yet, record it as an open decision so dependent specs wait for
71
71
  the answer instead of a reviewer discovering it.
72
72
 
73
- **Ask answerable questions in an artifact.** When the user must choose between
74
- options, record each choice as a brief decision and render the open decisions
75
- with the decision artifact (`.joycraft/references/artifacts.md`). Do not list
76
- option questions in chat: the artifact shows each question with its context and
77
- options, and saves the answers to the brief. Plain chat is only for one quick
78
- yes/no or an open question with no options. If you already asked option
79
- questions in chat and they are unanswered, move them to an artifact now.
73
+ **Review defaults stay strict.** Every feature starts with the review loop on
74
+ and its tools required. Record a waiver only from the user's own words in that
75
+ feature's interview: put them in the note of a settled brief decision and write
76
+ the feature's `waiver.json` naming that decision, before any review runs
77
+ (`.joycraft/references/completion.md`). A `no-review` waiver blocks reviewer
78
+ dispatch for the feature. Never propose or ask about one, and never carry it
79
+ into the next feature.
80
+
81
+ **Feature sessions stay quiet about the review loop.** A feature interview
82
+ asks nothing about review tools: read the recorded `## Review loop` and write
83
+ criteria its tools can check. Ask once, before any spec, only when the
84
+ feature's main claim needs something the recorded tools cannot show (a drag on a
85
+ screenshots-only tool). That is a tooling change: follow
86
+ [change a tool](../../../.joycraft/references/setup.md#review-loop-tools) and
87
+ record the answer, so it is not asked again.
88
+
89
+ **Ask in the user's question style.** When the user must choose between
90
+ options, record each choice as a brief decision. Then run
91
+ `node .joycraft/runtime/artifacts/cli.mjs questions --brief <brief>` and ask as
92
+ it says: one decision at a time with a short discussion after each answer, or a
93
+ page of decisions; in the decision artifact or as numbered prose in chat. Give
94
+ each consequential question its context, options with an example and a
95
+ trade-off, and your recommendation with the reason, in Simplified Technical
96
+ English. Never assume the user remembers what you remember. Save every answer
97
+ to the brief. Follow
98
+ [asking questions](../../../.joycraft/references/artifacts.md#asking-questions).
99
+ Plain chat with no decision is only for one quick yes/no or an open question
100
+ with no options.
80
101
 
81
102
  **Use the project's folders.** If `joycraft.yaml` exists, run
82
103
  `node .joycraft/runtime/config.mjs` and use its `paths` wherever Joycraft guidance
@@ -81,6 +81,22 @@ If it returns unenforced, or a separate review context/capability is unavailable
81
81
  leave the spec `in-review`; do not invent an ID or prompt to continue. This holds
82
82
  the claimed review, not unrelated authorized or hooks-disabled/manual work.
83
83
 
84
+ When a recorded tool breaks during a run (the dev server does not start, a
85
+ port is taken, the browser or simulator crashes), repair it with what the user
86
+ already agreed to: restart the dev server, free the port, relaunch the browser
87
+ or simulator, then retry once. A repair never installs, upgrades or
88
+ reconfigures anything; that needs the user's yes. If the repair fails, record an
89
+ `UNVERIFIED` observation with `cause: 'tool-break'` and the reason, leave that
90
+ spec `in-review`, and continue with the specs that do not depend on it. At the
91
+ first hold, tell the user in the conversation while workers run: the spec, the
92
+ tool, the cause and the repair tried. The session-end summary lists every held
93
+ spec again.
94
+
95
+ A tool break stops the whole queue only when the project will not build or
96
+ install, the worktree or repository is broken, or the next step would touch data
97
+ or systems outside the project (a shared database, a deploy). Every other tool
98
+ break is repair, then hold.
99
+
84
100
  Only the guarded completion path can move an `in-review` spec to `done`. Use its manual snapshot/complete commands only at an explicit non-native boundary; read [completion guidance](../../../.joycraft/references/completion.md) before doing so.
85
101
 
86
102
  When a current independent review is `FAIL` or `UNVERIFIED`, first fix
@@ -23,7 +23,7 @@ force local state to a remote tip. Start a replacement only from the durable
23
23
  run journal and a newly printed dispatch plan. This guidance describes the
24
24
  project/native runtime boundary; it does not enforce lifecycle behavior.
25
25
 
26
- For each completed spec, report the implementation revision, independent reviewer session ID, report and receipt paths, actual commands and outcomes, visual inspection status where required, and known limitations. For owner-accepted work, also report the owner-acceptance record and its inherited limits as `done, owner-accepted with <limits>`; preserve the original `FAIL` or `UNVERIFIED` report rather than calling it PASS or verified.
26
+ For each completed spec, report the implementation revision, independent reviewer session ID, report and receipt paths, actual commands and outcomes, visual inspection status where required, and known limitations. For owner-accepted work, also report the owner-acceptance record and its inherited limits as `done, owner-accepted with <limits>`; preserve the original `FAIL` or `UNVERIFIED` report rather than calling it PASS or verified. List each spec completed under a feature waiver as `done, waived by the user: "<instruction>"`, with any report's real verdict; never call it reviewed or verified.
27
27
 
28
28
  If a completion operation failed or was interrupted, do not declare success or hand-edit statuses. Retain the helper's error and journal/recovery information, then use the bounded local recovery process in [completion guidance](../../../.joycraft/references/completion.md). It only restores the two status surfaces and accepted-report receipt captured by the journal; it does not retry unrelated work.
29
29
 
@@ -25,11 +25,13 @@ remain valid; only assertions intended to detect the new behavior need meaningfu
25
25
  red/green. If a red check was not run, say so. Do not weaken assertions to obtain
26
26
  green or alter unrelated tests to manufacture universal failure.
27
27
 
28
- For UI requirements, use the tools and commands in the `Review loop` section of
29
- `AGENTS.md` (also in the worker packet as `reviewLoop`); a tool it marks skipped
30
- or missing leaves that UI claim UNVERIFIED.
31
- Render and inspect the relevant route, state and viewport.
32
- Perform the interactions the claim needs and observe their outcomes. Fix observed
28
+ For UI requirements, use the package's recorded tool and commands from the
29
+ worker packet's `reviewLoop` (the `Review loop` section of `AGENTS.md`, plus
30
+ `local` host tools), within its `canShow`. A claim beyond `canShow`, or in a
31
+ package the user skipped, is an agreed limit, not UNVERIFIED: check what the
32
+ tool can show and say how you checked. Render and inspect the relevant route,
33
+ state and viewport. When the tool can interact, perform the interactions the
34
+ claim needs and observe their outcomes. Fix observed
33
35
  problems within scope, then rerender and reinspect affected states. A saved
34
36
  screenshot, DOM check or browser launch alone is not visual inspection. Preserve
35
37
  tool access boundaries: if a check is blocked or unavailable, record UNVERIFIED
@@ -19,9 +19,10 @@ current render cannot answer them. Otherwise build or modify the smallest runnab
19
19
  slice in the project's actual stack. Show the consequential state and a useful
20
20
  interaction when interaction matters. One viewport is a useful starting size, not
21
21
  a rule for all products. Put explanatory detail behind disclosure. The slice is
22
- exploration, not production approval. Render and inspect it early with permitted
23
- browser tools; record the route, state, viewport, interactions and observed
24
- outcomes. If blocked, record UNVERIFIED and the reason; respect denials without
22
+ exploration, not production approval. Render and inspect it early with the
23
+ package's recorded review tool (`Review loop` in `AGENTS.md`) or the host's
24
+ permitted tools; record the route, state, viewport, interactions the tool can
25
+ perform and observed outcomes. If blocked, record UNVERIFIED and the reason; respect denials without
25
26
  alternate access routes.
26
27
 
27
28
  The optional local helper below needs the installed shared runtime. If it is
@@ -51,7 +51,7 @@ weakened assertions that could hide the requested behavior.
51
51
 
52
52
  Review scope as well as behavior: determine whether the listed inputs and criteria cover the requested work, including relevant newly changed files. For a cross-boundary change, inspect the combined observable behavior across the relevant producer, consumer, state or storage boundary; separate green helpers, snapshots, or assertion counts do not establish that behavior. If a material counterpart or owning decision/context source was omitted, report the scope gap and require its exact path to be added before a fresh current review subject is prepared. Preserve the snapshot's implementation revision as provenance while checking its current bound inputs. A changed, missing, or renamed input makes the subject stale; do not substitute a newer HEAD as its identity.
53
53
 
54
- Create a schema-v1 verification report tied to that subject, using the shared fields in [.joycraft/contracts/types.d.ts](../../../.joycraft/contracts/types.d.ts). For every declared criterion, provide PASS observations for its required evidence kinds. Keep unfiltered command output and its real exit status in hashed, project-relative receipts. A source scan supports only structural observations. Visual observations require a reviewer-inspected render plus route, state, viewport, interactions, and outcomes; take the render with the tools and start command in the packet's `reviewLoop` (the `Review loop` section of `AGENTS.md`), and mark a claim whose tool is skipped or missing `UNVERIFIED`. Mark unavailable inspection, a failing result, or an unresolved concern as `UNVERIFIED` or `FAIL` truthfully.
54
+ Create a schema-v1 verification report tied to that subject, using the shared fields in [.joycraft/contracts/types.d.ts](../../../.joycraft/contracts/types.d.ts). For every declared criterion, provide PASS observations for its required evidence kinds. Keep unfiltered command output and its real exit status in hashed, project-relative receipts. A source scan supports only structural observations. Visual observations require a reviewer-inspected render plus route, state, viewport, interactions, and outcomes. Take evidence with the package's recorded tool, start command and evidence command from the packet's `reviewLoop` (`section`: the `Review loop` section of `AGENTS.md`; `local`: this machine's host tools), within that tool's `canShow`. A claim beyond `canShow`, or in a package the user skipped, is an agreed limit, not `UNVERIFIED`: check what the agreed tool can show, say in plain words how you checked, and leave what is left to the parent agent. A picture-only tool gives `interactions: []`. `UNVERIFIED` means a check that should have run and did not: the tool broke, the app did not start, or a render was saved but not inspected. On a tool break, first repair with what the user agreed to: restart the dev server, free the port, relaunch the browser or simulator, retry once. A repair never installs, upgrades or reconfigures anything. If it still fails, record `UNVERIFIED` with `cause: 'tool-break'` and the reason. Mark a failing result or an unresolved concern as `FAIL` or `UNVERIFIED` truthfully.
55
55
 
56
56
  Run each test command once in single-run mode with `CI=1` and a 300000 ms
57
57
  timeout; follow [running tests](../../../.joycraft/references/testing.md#running-tests)
@@ -34,6 +34,11 @@ criterion by what you observe.
34
34
  Report every finding with a severity and the criterion it touches. Only a failing
35
35
  stated criterion, or a reproducible defect or regression in the changed code with
36
36
  file:line or a receipt, blocks. A case the spec does not decide is "decision
37
- needed", not FAIL. Do not repeat an earlier round's finding as new. Record only
38
- what you observed; mark anything unobserved UNVERIFIED. Keep the final message
37
+ needed", not FAIL. Do not repeat an earlier round's finding as new.
38
+
39
+ Take evidence with each package's recorded tool from the packet's `reviewLoop`,
40
+ within its `canShow`, and say in plain words how you checked. A limit the user
41
+ agreed to (a claim beyond `canShow`, or a package the user skipped) is not
42
+ UNVERIFIED; the parent agent judges what is left. UNVERIFIED is a check that
43
+ should have run and did not. Record only what you observed. Keep the final message
39
44
  short: verdict, blocking findings, decision-needed items, report path.
@@ -89,6 +89,13 @@ html[data-jca-theme] body { margin: 0; background: var(--canvas); color: var(--i
89
89
  .jca-option input { margin-top: .28rem; accent-color: var(--accent); }
90
90
  .jca-option strong, .jca-option small { display: block; }
91
91
  .jca-option small { margin-top: .1rem; color: var(--muted); }
92
+ .jca-option .jca-option-line { margin-top: .3rem; }
93
+ .jca-option-line b, .jca-recommendation b { color: var(--ink); font-weight: 650; }
94
+ .jca-badge { display: inline-block; margin-left: .5rem; padding: .05rem .45rem; border-radius: 999px; background: var(--accent); color: var(--accent-text); font-size: .7rem; font-weight: 700; letter-spacing: .04em; vertical-align: .1em; }
95
+ .jca-option-recommended { border-color: var(--accent); }
96
+ .jca-context { margin: .8rem 0 0; }
97
+ .jca-context p { margin: .55rem 0; }
98
+ .jca-recommendation { margin: 0 0 1rem; padding: .7rem .8rem; border-left: 3px solid var(--accent); background: var(--accent-soft); }
92
99
  .jca-note-label { display: block; margin-bottom: .35rem; font-size: .88rem; font-weight: 700; }
93
100
  .jca-note { display: block; width: 100%; resize: vertical; border: 1px solid var(--control-line); border-radius: 8px; padding: .7rem; color: var(--ink); background: var(--input); font: inherit; }
94
101
  .jca-note::placeholder { color: var(--muted); opacity: 1; }
@@ -46,8 +46,9 @@
46
46
  return JSON.stringify({
47
47
  question: question.question || '',
48
48
  options: (question.options || []).map(function (option) {
49
- return { value: option.value, label: option.label || '', description: option.description || '', requiresNote: Boolean(option.requiresNote) };
49
+ return { value: option.value, label: option.label || '', description: option.description || '', example: option.example || '', tradeoff: option.tradeoff || '', recommended: Boolean(option.recommended), requiresNote: Boolean(option.requiresNote) };
50
50
  }),
51
+ recommendation: question.recommendation || '',
51
52
  status: question.status || '',
52
53
  answer: question.answer == null ? null : question.answer,
53
54
  note: question.note == null ? '' : question.note
@@ -336,23 +337,38 @@
336
337
  body.append(node('p', { class: 'jca-status jca-status-' + (question.status || 'open'), text: statusLabel(question.status) }));
337
338
  if (question.answer) body.append(node('p', { class: 'jca-recorded', text: 'Brief record: ' + question.answer }));
338
339
  (model.comparisons || []).filter(function (comparison) { return comparison.decisionId === question.id; }).forEach(function (comparison) { body.append(renderComparison(comparison)); });
339
- if (question.context || question.why || (question.affected || []).length) {
340
- var detail = node('details', { class: 'jca-details' }, [node('summary', { text: 'Context and impact' })]);
341
- if (question.context) detail.append(node('p', { text: question.context }));
342
- if (question.why) detail.append(node('p', { text: question.why }));
340
+ // Context comes before the choice, never behind a click: a person should not have to remember it.
341
+ if (question.context || question.why) {
342
+ body.append(node('div', { class: 'jca-context' }, [
343
+ question.context ? node('p', { text: question.context }) : null,
344
+ question.why ? node('p', { text: question.why }) : null
345
+ ]));
346
+ }
347
+ if (question.details || (question.affected || []).length) {
348
+ var detail = node('details', { class: 'jca-details' }, [node('summary', { text: 'More detail' })]);
349
+ if (question.details) detail.append(node('p', { text: question.details }));
343
350
  if ((question.affected || []).length) detail.append(node('p', { text: 'Affected: ' + question.affected.join(', ') }));
344
351
  body.append(detail);
345
352
  }
346
353
  var fieldset = node('fieldset', { class: 'jca-options', disabled: readOnly });
347
354
  fieldset.append(node('legend', { text: readOnly ? 'Responses are read-only' : 'Choose one response' }));
348
- (question.options || []).forEach(function (option) {
355
+ // The recommended option leads; "Something else" always closes the list.
356
+ var rank = function (option) { return option.value === 'custom' ? 2 : option.recommended ? 0 : 1; };
357
+ var ordered = (question.options || []).map(function (option, position) { return { option: option, position: position }; })
358
+ .sort(function (left, right) { return rank(left.option) - rank(right.option) || left.position - right.position; })
359
+ .map(function (item) { return item.option; });
360
+ var optionLine = function (name, value) { return value ? node('small', { class: 'jca-option-line' }, [node('b', { text: name + ': ' }), text(value)]) : null; };
361
+ ordered.forEach(function (option) {
349
362
  var id = 'jca-' + question.id + '-' + option.value;
350
363
  var radio = node('input', { type: 'radio', id: id, name: 'jca-answer-' + question.id, value: option.value, checked: current.answer === option.value,
351
364
  onchange: function () { draft[question.id] = { answer: option.value, note: (draft[question.id] || {}).note || '' }; setMessage('quiet', 'Unsaved changes.'); syncLiveUi(); syncNoteRequirement(); } });
352
- var label = node('label', { class: 'jca-option', for: id }, [radio, node('span', { class: 'jca-option-copy' }, [node('strong', { text: option.label }), option.description ? node('small', { text: option.description }) : null])]);
365
+ var title = node('strong', { text: option.label }, [option.recommended ? node('span', { class: 'jca-badge', text: 'Recommended' }) : null]);
366
+ var label = node('label', { class: 'jca-option' + (option.recommended ? ' jca-option-recommended' : ''), for: id }, [radio, node('span', { class: 'jca-option-copy' }, [title, option.description ? node('small', { text: option.description }) : null, optionLine('Example', option.example), optionLine('Trade-off', option.tradeoff)])]);
353
367
  fieldset.append(label);
354
368
  });
355
369
  body.append(fieldset);
370
+ var pick = (question.options || []).find(function (option) { return option.recommended; });
371
+ if (pick && question.recommendation) body.append(node('p', { class: 'jca-recommendation' }, [node('b', { text: 'Recommendation: ' + pick.label + '. ' }), text(question.recommendation)]));
356
372
  var note = node('textarea', { class: 'jca-note', id: 'jca-note-' + question.id, rows: 3, disabled: !canEditNote(question), placeholder: 'Optional note' });
357
373
  note.value = current.note || '';
358
374
  function syncNoteRequirement() {
@@ -68,6 +68,8 @@ export interface QueueSpec {
68
68
  acceptedReport?: { path: string; sha256: string };
69
69
  /** Explicit owner exception, mutually exclusive with acceptedReport; not a PASS. */
70
70
  ownerAcceptance?: Receipt;
71
+ /** The feature's FeatureWaiver file, plus the UNVERIFIED report a tool-break waiver completed on; mutually exclusive with acceptedReport and ownerAcceptance. */
72
+ waiver?: Receipt & { review?: Receipt };
71
73
  }
72
74
  export interface SpecQueue { schemaVersion: 1; feature: string; specs: QueueSpec[] }
73
75
  export interface BoundInput extends VerificationInput {
@@ -103,6 +105,8 @@ export interface Observation {
103
105
  result: Result;
104
106
  actual: string;
105
107
  receipts: Receipt[];
108
+ /** Set only on an UNVERIFIED observation whose recorded tool did not work during the run. */
109
+ cause?: 'tool-break';
106
110
  /** Mandatory for test/command observations; null means unavailable. */
107
111
  command?: { argv: string[]; exitCode: number | null };
108
112
  /** Mandatory for visual observations. Interactions name actual outcomes. */
@@ -156,6 +160,24 @@ export interface OwnerAcceptance {
156
160
  limitations: string[];
157
161
  createdAt: string;
158
162
  }
163
+ /**
164
+ * The user's own words, from one feature's interview, loosening that feature's queue only. Never covers a FAIL.
165
+ * Lives at `<feature folder>/waiver.json`. While it covers no-review, reviewer dispatch for the queue is refused.
166
+ */
167
+ export interface FeatureWaiver {
168
+ schemaVersion: 1;
169
+ kind: 'feature-waiver';
170
+ queuePath: string;
171
+ /** Nonempty subset: complete without a review report, or with an UNVERIFIED report whose gaps are all tool breaks. */
172
+ covers: Array<'no-review' | 'tool-break'>;
173
+ /** The user's exact words: whole words of the named decision's note or answer, at least three. */
174
+ instruction: string;
175
+ /** Id of the clarified or assigned brief decision whose note holds the instruction. */
176
+ decision: string;
177
+ /** Receipt of the feature brief that holds the decision. */
178
+ source: Receipt;
179
+ createdAt: string;
180
+ }
159
181
  export interface ValidationResult { ok: boolean; errors: string[] }
160
182
  export type Need = 'repair' | 'feature' | 'context' | 'decision' | 'ui-exploration' | 'implementation' | 'verification' | 'unknown';
161
183
  export type NextInteraction = 'continue' | 'inspect' | 'render-slice' | 'ask-focused' | 'verify' | 'abstain';
@@ -33,8 +33,10 @@ another decision or approval authority.
33
33
 
34
34
  The preview validates the complete queue before any write: the canonical queue
35
35
  must be the only queue anywhere in the feature tree, all specs must be
36
- `done`, and each spec must have exactly one of an accepted PASS report or an
37
- explicit `ownerAcceptance` receipt. A normal accepted report must retain its
36
+ `done`, and each spec must have exactly one of an accepted PASS report, an
37
+ explicit `ownerAcceptance` receipt, or a feature `waiver` receipt. A waiver
38
+ keeps its record bytes, the user's quoted instruction as its limitation, and
39
+ for a tool-break completion its UNVERIFIED `waiver.review` report and receipts. A normal accepted report must retain its
38
40
  PASS evidence. An owner-acceptance record is historical exception metadata:
39
41
  it binds the queue/spec/subject identity to a hashed independent review whose
40
42
  verdict remains `FAIL` or `UNVERIFIED`, carries nonempty limitations, and is
@@ -1,11 +1,64 @@
1
1
  # Decision artifacts
2
2
 
3
- Use an artifact for every question that asks the user to choose between
4
- options: record the choices as brief decisions, then render them. Also use one
5
- for an editorial comparison of existing screenshots or a review summary. Plain
6
- chat is only for one quick yes/no or an open question with no options. It renders a selected view of the brief; the brief and queue remain
7
- the source of truth. Do not use one for a routine native question, a product UI
8
- slice whose real interaction must be explored, or a long questionnaire.
3
+ Record every choice the user must make between options as a brief decision,
4
+ then ask it in the user's [question style](#asking-questions): on this page or
5
+ in chat. Also use a page for an editorial comparison of existing screenshots or
6
+ a review summary. A page renders a selected view of the brief; the brief and
7
+ queue remain the source of truth. Do not use one for a routine native question,
8
+ a product UI slice whose real interaction must be explored, or a long
9
+ questionnaire.
10
+
11
+ ## Asking questions
12
+
13
+ Do not assume the user remembers what you remember. Before you ask, run:
14
+
15
+ ```sh
16
+ node .joycraft/runtime/artifacts/cli.mjs questions --brief docs/features/<slug>/brief.md
17
+ ```
18
+
19
+ It prints `pacing`, `surface` and where each came from. With `unset: true`, the
20
+ project has no recorded style yet: ask for it as [setup](setup.md) step 2 says,
21
+ then continue.
22
+
23
+ - **Pacing `one-at-a-time`.** Ask one decision. After the answer, tell the user
24
+ what it changes and ask if that is correct, and stop there: do not put the
25
+ next decision in the same message. Ask the next decision after the user
26
+ replies. Ask first the decision that the others depend on.
27
+ - **Pacing `page`.** Show the open decisions together. After the save, say in
28
+ one line per answer what it changes.
29
+ - **Surface `artifact`.** Serve the page for the decisions you ask now (the
30
+ view's `decisionIds`). Discuss the answers in chat.
31
+ - **Surface `chat`.** Write the question as prose with numbered options, in the
32
+ order below. Save each answer to its brief decision with the user's exact
33
+ words as the note. The brief stays the only record of the answer.
34
+
35
+ The user can change the style for one feature ("ask me one at a time here"):
36
+ `questions --brief <brief> --set pacing=one-at-a-time`. That writes a
37
+ `question-style:` line in that brief only. `--clear` removes it. A request to
38
+ change it for every feature is `questions --set ...` without `--brief`.
39
+
40
+ **Order for a consequential choice** (one that changes scope, cost or what is
41
+ built):
42
+
43
+ 1. Context: what you are doing, the current state, and why you are blocked.
44
+ 2. The question: one sentence, one decision.
45
+ 3. Two to four options. Each has a short explanation, a concrete example of the
46
+ result, and the main trade-off.
47
+ 4. Your recommendation: the option and the reason.
48
+
49
+ On the page, put these in `context` (and `why`), `question`, each option's
50
+ `description`, `example` and `tradeoff`, and `recommended: true` on one option
51
+ with the reason in `recommendation`. A small choice can be shorter, but still
52
+ give the context the user needs to answer it.
53
+
54
+ **Words.** Write questions in
55
+ [ASD-STE100 Simplified Technical English](https://www.asd-ste100.org/): one
56
+ meaning for each word, active voice, short sentences, one idea in each sentence.
57
+ Use the project's terms from `GLOSSARY.md`, or the glossary that `GLOSSARY-MAP.md`
58
+ names, when the project has one. When the user says a question is not clear
59
+ ("wait, what?"), do not repeat it. Write it again with the context that was
60
+ missing. This follows `wait-what` in
61
+ [mattpocock/skills](https://github.com/mattpocock/skills/tree/main/skills/productivity/wait-what).
9
62
 
10
63
  The renderer is deterministic. Agents author a short view-data file, such as
11
64
  [`templates/artifact.json`](../templates/artifact.json), and link existing
@@ -32,9 +85,13 @@ review) is shown where they are working, and you always say its path:
32
85
  ## Decisions and view data
33
86
 
34
87
  A brief decision keeps its existing `id`, `status`, `question`, `answer`,
35
- `options`, and `affected` fields. Artifact views also support an optional
36
- `description` on an option and `requiresNote: true` when that option needs a
37
- short explanation. Every question also offers "Something else" (value `custom`,
88
+ `options`, and `affected` fields. A decision can also have `context` and `why`
89
+ (shown before the options), `details` (under "More detail") and
90
+ `recommendation` (the reason for the recommended option). An option can have
91
+ `description`, `example`, `tradeoff`, `recommended: true` (one option at most;
92
+ not `custom`) and `requiresNote: true` when that option needs a short
93
+ explanation. All of these are optional, and the page lists the recommended
94
+ option first. Every question also offers "Something else" (value `custom`,
38
95
  note required), so the person can always answer in their own words; do not
39
96
  declare it unless you want different wording. The canonical answer remains an
40
97
  option `value`, including `custom`. A saved answer keeps its note box: the
@@ -11,10 +11,11 @@ node .joycraft/runtime/completion.mjs scope --queue <queue-path> --spec <id>
11
11
  node .joycraft/runtime/completion.mjs snapshot --queue <queue-path> --spec <id> --out <subject.json>
12
12
  node .joycraft/runtime/completion.mjs finalize-review --queue <queue-path> --spec <id> --report <report.json>
13
13
  node .joycraft/runtime/completion.mjs accept-owner --queue <queue-path> --spec <id> --decision <owner-acceptance.json>
14
+ node .joycraft/runtime/completion.mjs accept-waiver --queue <queue-path> --spec <id> --waiver <feature-waiver.json> [--report <report.json>]
14
15
  node .joycraft/runtime/completion.mjs recover --queue <queue-path>
15
16
  ```
16
17
 
17
- `scope` checks a queue's declared change scope without writing a subject. `snapshot` does the same check first, then exclusively creates a new `--out` subject file without changing a status. It binds the implementation revision, queue/spec identity, criteria, and SHA-256 values for exact relevant inputs. The shared record shapes are defined in [.joycraft/contracts/types.d.ts](../contracts/types.d.ts). New subjects carry `binding.version: 1`: their queue projection contains the reviewed spec and its transitive dependencies, with completion metadata (`status`, `acceptedReport`, and `ownerAcceptance`) removed from those records. An unrelated queue record can change without staling that subject; a selected spec's other metadata and dependency edges remain bound. A spec binding preserves raw bytes except for its top-level `status:` value.
18
+ `scope` checks a queue's declared change scope without writing a subject. `snapshot` does the same check first, then exclusively creates a new `--out` subject file without changing a status. It binds the implementation revision, queue/spec identity, criteria, and SHA-256 values for exact relevant inputs. The shared record shapes are defined in [.joycraft/contracts/types.d.ts](../contracts/types.d.ts). New subjects carry `binding.version: 1`: their queue projection contains the reviewed spec and its transitive dependencies, with completion metadata (`status`, `acceptedReport`, `ownerAcceptance`, and `waiver`) removed from those records. An unrelated queue record can change without staling that subject; a selected spec's other metadata and dependency edges remain bound. A spec binding preserves raw bytes except for its top-level `status:` value.
18
19
 
19
20
  An optional `verification.changeScope` records an explicit full Git commit SHA in `baseRevision`; a branch name, `origin/main`, or inferred base is rejected. Changed verification inputs are captured from NUL-delimited Git status records, including staged, unstaged, rename, and deletion topology. Any changed path outside the verification inputs must be named exactly in `exclusions` with its reason. Store temporary packets and subjects in the project's ignored local-artifact home when it has one; otherwise add each exact untracked snapshot packet output to `exclusions`. `complete` filters only its exact report, receipt, owner-decision, and pending-journal paths while rechecking scope. A deliberate deletion needs a non-requirement verification input with `state: "absent"`; requirement inputs cannot be absent. A rename needs the new present input and the old non-requirement input declared absent.
20
21
 
@@ -86,9 +87,61 @@ verified, and the original failed or unverified report remains historical
86
87
  evidence. Archive may preserve this paired history only at session end; it does
87
88
  not rewrite the report or create a fresh acceptance.
88
89
 
90
+ ## Feature waiver
91
+
92
+ Review defaults are strict for every feature. A `FeatureWaiver` loosens them for
93
+ one feature's queue, only in the user's own words from that feature's
94
+ interview. The agent never proposes or asks about one, and the next feature
95
+ starts from the defaults. Record the user's words in the note of a settled brief
96
+ decision, then write the feature's one waiver file, `waiver.json` beside
97
+ `brief.md` (never a verification input), before any review runs:
98
+
99
+ ```json
100
+ {
101
+ "schemaVersion": 1,
102
+ "kind": "feature-waiver",
103
+ "queuePath": "docs/features/example/specs/.joycraft-spec-queue.json",
104
+ "covers": ["no-review"],
105
+ "instruction": "quick janky prototype, skip review on this",
106
+ "decision": "skip-review",
107
+ "source": { "path": "docs/features/example/brief.md", "sha256": "<SHA-256 of the brief>" },
108
+ "createdAt": "2026-11-02T12:00:00.000Z"
109
+ }
110
+ ```
111
+
112
+ `covers` is a nonempty subset of `no-review` and `tool-break`. `decision` is the
113
+ id of the brief decision that holds the user's words. `accept-waiver` refuses a
114
+ waiver for another queue, at another path, with unknown fields or covers, an
115
+ empty instruction, a `source` that is not this feature's `brief.md`, a brief hash
116
+ that differs, or a `decision` that is missing, not `clarified`/`assigned` with an
117
+ answer, or whose `note` or `answer` does not contain the instruction as whole
118
+ words (at least three; whitespace and line wraps are normalized).
119
+
120
+ - **Skip means review never runs** (decision `block-review-dispatch`).
121
+ `dispatch-preflight.mjs` refuses a reviewer for a queue whose `waiver.json`
122
+ covers `no-review`, in the worker tree or the coordinator tree, and refuses
123
+ while that file is unreadable. So no review, and no `FAIL`, exists for a spec
124
+ that `no-review` completes. Without `--report`, `no-review` completes an
125
+ `in-review` spec. As an extra guard it still refuses when a review report for
126
+ the spec is visible in the feature folder or `.joycraft/local/packets/`.
127
+ - With `--report`, the report must be current and independent (validated as for
128
+ `accept-owner`), `UNVERIFIED`, and every non-`PASS` observation must carry
129
+ `cause: "tool-break"`; the waiver must cover `tool-break`. The report keeps
130
+ its verdict; evidence errors are returned as diagnostics. A tool-break-only
131
+ waiver does not block reviewer dispatch.
132
+ - A `FAIL` verdict or any `FAIL` observation never completes, whatever `covers`
133
+ says. A `PASS` report uses `finalize-review`.
134
+
135
+ Limit: a review that ran before `waiver.json` existed is caught only by the
136
+ folder guard. The waiver comes from the interview, so it exists before dispatch.
137
+ On success the queue entry records `waiver: { path, sha256 }`, plus
138
+ `review: { path, sha256 }` for a report-backed completion, mutually exclusive
139
+ with `acceptedReport` and `ownerAcceptance`. A waived spec counts as `done` for
140
+ dependents. Report it as `done, waived by the user: "<instruction>"`.
141
+
89
142
  ## Evidence and reporting
90
143
 
91
- Evidence records observations rather than promises. Command and test observations retain their actual argv, exit code, and unfiltered output receipt. Source review supports structural claims only. A required visual claim needs an inspected render, route, state, viewport, interactions, and observed outcomes. Record unavailable work as `UNVERIFIED`, failures as `FAIL`, and limits explicitly.
144
+ Evidence records observations rather than promises. Command and test observations retain their actual argv, exit code, and unfiltered output receipt. Source review supports structural claims only. A required visual claim needs an inspected render, route, state, viewport, interactions, and observed outcomes. Record a check that should have run and did not (a tool break, an app that did not start) as `UNVERIFIED`, failures as `FAIL`, and limits explicitly. A claim that needs evidence beyond the recorded tool's `canShow`, or a package the user skipped, is an agreed limit, not `UNVERIFIED`; a check within `canShow` that could not run because the tool broke is `UNVERIFIED` with `cause: 'tool-break'`.
92
145
 
93
146
  Report the command result, reviewer identity, receipt paths, actual observations, and limitations. A green suite or an implementation commit alone is not accepted completion.
94
147
 
@@ -10,6 +10,9 @@ For a mutable worker or stateful runtime check, first apply the conditional
10
10
  worktree or a resource lease. A `read-only` workspace keeps declared source
11
11
  inputs frozen; it may still write declared review reports and receipts outside
12
12
  those inputs. Cache-mutating runtime work needs separate mutable state.
13
+ Reviewer preflight refuses a queue whose feature `waiver.json` covers
14
+ `no-review` ([completion guidance](completion.md#feature-waiver)); complete
15
+ those specs with `accept-waiver` instead.
13
16
 
14
17
  ## Contents
15
18
 
@@ -93,6 +96,11 @@ more, alone, with a fresh packet path. If it fails again, stop that spec's
93
96
  review and tell the user its printed reason; never continue as if the review
94
97
  started. Do not retry a `Task startup stopped:` refusal unchanged; fix its cause.
95
98
 
99
+ A worker that holds a spec for a tool break reports the spec, the tool, the
100
+ cause and the repair it tried. The coordinator tells the user at the first hold,
101
+ in the conversation, while other workers keep running; it does not wait for the
102
+ session end. The session-end summary lists every held spec again.
103
+
96
104
  The return is `ready: true` for an available valid startup path, including a
97
105
  hooks-disabled/manual path. It is not a cryptographic attestation. Missing human
98
106
  prompt history for an enforced startup is explicit
@@ -158,6 +166,10 @@ decisions. Without a selector, the packet conservatively carries every current
158
166
  answered decision from that brief. The packet retains its whole-brief reference
159
167
  so workers can read each decision in context. Its inputs include any current
160
168
  coupled counterpart required to review the requested behavior.
169
+ When `AGENTS.md` has a `## Review loop` section or this machine has
170
+ `.joycraft/local/review-loop.json`, the packet carries `reviewLoop`:
171
+ `{ path, section, local? }`, the shared section plus, when present,
172
+ `local: { path, record }` with this machine's host tools.
161
173
 
162
174
  ## Bind the hook source view
163
175