spexcode 0.4.1 → 0.5.0

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 (111) hide show
  1. package/README.md +37 -17
  2. package/package.json +9 -5
  3. package/spec-cli/README.md +23 -14
  4. package/spec-cli/hooks/dispatch.sh +7 -5
  5. package/spec-cli/hooks/harness.sh +19 -8
  6. package/spec-cli/src/anchors.ts +110 -18
  7. package/spec-cli/src/claude-headless.ts +271 -0
  8. package/spec-cli/src/cli.ts +112 -38
  9. package/spec-cli/src/client.ts +42 -11
  10. package/spec-cli/src/doctor.ts +174 -36
  11. package/spec-cli/src/gateway-auth.ts +194 -0
  12. package/spec-cli/src/gateway-hub.ts +360 -0
  13. package/spec-cli/src/gateway.ts +27 -19
  14. package/spec-cli/src/git.ts +12 -5
  15. package/spec-cli/src/graph.ts +54 -43
  16. package/spec-cli/src/graphStream.ts +265 -34
  17. package/spec-cli/src/guide.ts +115 -56
  18. package/spec-cli/src/harness-select.ts +20 -6
  19. package/spec-cli/src/harness.ts +295 -39
  20. package/spec-cli/src/help.ts +68 -28
  21. package/spec-cli/src/host.ts +610 -0
  22. package/spec-cli/src/identity-presets.d.ts +13 -0
  23. package/spec-cli/src/identity-presets.js +128 -0
  24. package/spec-cli/src/index.ts +158 -87
  25. package/spec-cli/src/init.ts +53 -25
  26. package/spec-cli/src/layout.ts +6 -5
  27. package/spec-cli/src/lint.ts +25 -121
  28. package/spec-cli/src/login-page.ts +12 -5
  29. package/spec-cli/src/materialize.ts +39 -13
  30. package/spec-cli/src/mentions.ts +16 -11
  31. package/spec-cli/src/message-stream.ts +147 -0
  32. package/spec-cli/src/opencode-headless.ts +95 -0
  33. package/spec-cli/src/opencode.ts +145 -0
  34. package/spec-cli/src/pi-harness.ts +124 -0
  35. package/spec-cli/src/pi-headless.ts +195 -0
  36. package/spec-cli/src/project-identity.ts +73 -0
  37. package/spec-cli/src/pty-bridge.ts +321 -521
  38. package/spec-cli/src/pty-helper.mjs +86 -0
  39. package/spec-cli/src/reaper.ts +16 -8
  40. package/spec-cli/src/reviewSnapshot.ts +23 -0
  41. package/spec-cli/src/reviews.ts +347 -0
  42. package/spec-cli/src/session-timeline.ts +14 -0
  43. package/spec-cli/src/sessions.ts +211 -75
  44. package/spec-cli/src/shim-runtime.ts +169 -0
  45. package/spec-cli/src/slash-commands.ts +76 -0
  46. package/spec-cli/src/source-files.ts +79 -0
  47. package/spec-cli/src/specs.ts +12 -7
  48. package/spec-cli/src/supervise.ts +37 -15
  49. package/spec-cli/src/tree.ts +8 -8
  50. package/spec-cli/src/uninstall.ts +56 -23
  51. package/spec-cli/templates/hooks/prepare-commit-msg +32 -18
  52. package/spec-cli/templates/spec/project/.plugins/commands/extract/spec.md +24 -18
  53. package/spec-cli/templates/spec/project/.plugins/commands/regroup/spec.md +3 -3
  54. package/spec-cli/templates/spec/project/.plugins/commands/rename/spec.md +11 -0
  55. package/spec-cli/templates/spec/project/.plugins/commands/spec.md +22 -8
  56. package/spec-cli/templates/spec/project/.plugins/commands/supervisor/spec.md +1 -1
  57. package/spec-cli/templates/spec/project/.plugins/commands/tidy/spec.md +29 -11
  58. package/spec-cli/templates/spec/project/.plugins/core/idle/spec.md +1 -1
  59. package/spec-cli/templates/spec/project/.plugins/core/mark-active/spec.md +2 -0
  60. package/spec-cli/templates/spec/project/.plugins/core/session-fail/spec.md +1 -1
  61. package/spec-cli/templates/spec/project/.plugins/core/spec-first/spec-first.sh +28 -39
  62. package/spec-cli/templates/spec/project/.plugins/core/spec-first/spec.md +5 -3
  63. package/spec-cli/templates/spec/project/.plugins/core/spec-of-file/spec.md +1 -1
  64. package/spec-cli/templates/spec/project/.plugins/core/spec.md +2 -2
  65. package/spec-cli/templates/spec/project/.plugins/core/stop-gate/spec.md +7 -1
  66. package/spec-cli/templates/spec/project/.plugins/prompts/forge-link/spec.md +2 -2
  67. package/spec-cli/templates/spec/project/.plugins/prompts/reproduce-before-fix/spec.md +5 -5
  68. package/spec-cli/templates/spec/project/.plugins/prompts/spec.md +10 -6
  69. package/spec-cli/templates/spec/project/.plugins/skills/distill/spec.md +38 -50
  70. package/spec-cli/templates/spec/project/.plugins/skills/spec.md +9 -7
  71. package/spec-cli/templates/spec/project/.plugins/spec.md +17 -12
  72. package/spec-cli/templates/spec/project/spec.md +6 -5
  73. package/spec-cli/templates/spexcode.json +10 -2
  74. package/spec-dashboard/dist/assets/Dashboard-C_w_wdk5.js +27 -0
  75. package/spec-dashboard/dist/assets/EvalsPage-5_nfIYll.js +2 -0
  76. package/spec-dashboard/dist/assets/IssuesPage-By-u--95.js +1 -0
  77. package/spec-dashboard/dist/assets/MobileApp-CVEwjHr9.js +2 -0
  78. package/spec-dashboard/dist/assets/Modal-BqgvzMJD.js +1 -0
  79. package/spec-dashboard/dist/assets/PageScroll-B_dKCuXx.js +1 -0
  80. package/spec-dashboard/dist/assets/ProjectsPage-RVP8AqK4.js +1 -0
  81. package/spec-dashboard/dist/assets/SessionInterface-6GBZ9nXN.css +32 -0
  82. package/spec-dashboard/dist/assets/SessionInterface-Bh3vq8SU.js +39 -0
  83. package/spec-dashboard/dist/assets/SessionWindow-BuJ5mzjC.js +13 -0
  84. package/spec-dashboard/dist/assets/Settings-B8KFocsz.js +1 -0
  85. package/spec-dashboard/dist/assets/TimelineChat-K0wdlweB.js +1 -0
  86. package/spec-dashboard/dist/assets/bindings-BC9vqpYU.js +1 -0
  87. package/spec-dashboard/dist/assets/index-BKaTHjmU.js +41 -0
  88. package/spec-dashboard/dist/assets/index-DcnCaBAC.css +1 -0
  89. package/spec-dashboard/dist/assets/terminalFont-DmTPNdbU.js +1 -0
  90. package/spec-dashboard/dist/index.html +10 -10
  91. package/spec-dashboard/src/reviewFilters.js +279 -0
  92. package/spec-dashboard/src/reviewQuery.js +174 -0
  93. package/spec-dashboard/src/session.js +167 -0
  94. package/spec-eval/src/cli.ts +49 -28
  95. package/spec-eval/src/humanok.ts +1 -1
  96. package/spec-eval/src/matrix.ts +693 -0
  97. package/spec-eval/src/sessioneval.ts +726 -66
  98. package/spec-cli/src/migrate-table.ts +0 -408
  99. package/spec-cli/src/migrate.ts +0 -386
  100. package/spec-dashboard/dist/assets/Dashboard-C_fGmOKK.js +0 -27
  101. package/spec-dashboard/dist/assets/EvalsPage-Cnr1s3bq.js +0 -2
  102. package/spec-dashboard/dist/assets/FoldToggle-x9gtO1OQ.js +0 -1
  103. package/spec-dashboard/dist/assets/IssuesPage-5f_vL-JV.js +0 -1
  104. package/spec-dashboard/dist/assets/MobileApp-DEO1jgGM.js +0 -1
  105. package/spec-dashboard/dist/assets/SessionInterface-CAlbMOFR.js +0 -66
  106. package/spec-dashboard/dist/assets/SessionInterface-DYP7pi_n.css +0 -32
  107. package/spec-dashboard/dist/assets/SessionWindow-JYbpPwNB.js +0 -13
  108. package/spec-dashboard/dist/assets/Settings-DKb5Ji_X.js +0 -1
  109. package/spec-dashboard/dist/assets/index-BQu-oJ8J.js +0 -41
  110. package/spec-dashboard/dist/assets/index-BbMkwuix.css +0 -1
  111. package/spec-dashboard/dist/assets/launch-BM9GgvkX.js +0 -6
@@ -0,0 +1,128 @@
1
+ // One browser-safe identity registry shared by backend validation and every dashboard projection.
2
+ // Geometry is data so the React renderer and favicon serializer cannot drift into separate drawings.
3
+
4
+ export const DEFAULT_PROJECT_ICON = 'spexcode'
5
+ export const DEFAULT_GATEWAY_ICON = 'gateway'
6
+
7
+ export const IDENTITY_PRESETS = Object.freeze([
8
+ {
9
+ id: 'spexcode', label: 'SpexCode', bg: '#166534', fg: '#f0fdf4',
10
+ shapes: [
11
+ { tag: 'circle', cx: 12, cy: 6.2, r: 2.2 },
12
+ { tag: 'circle', cx: 6.2, cy: 17.5, r: 2 },
13
+ { tag: 'circle', cx: 17.8, cy: 17.5, r: 2 },
14
+ { tag: 'path', d: 'M12 8.4v2.1c0 2.2-5.8 1.8-5.8 5' },
15
+ { tag: 'path', d: 'M12 10.5c0 2.2 5.8 1.8 5.8 5' },
16
+ ],
17
+ },
18
+ {
19
+ id: 'gateway', label: 'Gateway', bg: '#155e75', fg: '#ecfeff',
20
+ shapes: [
21
+ { tag: 'path', d: 'm12 3 8 4.2-8 4.2-8-4.2Z' },
22
+ { tag: 'path', d: 'm4 11.2 8 4.2 8-4.2' },
23
+ { tag: 'path', d: 'm4 15.2 8 4.2 8-4.2' },
24
+ ],
25
+ },
26
+ {
27
+ id: 'mdi:rocket-launch', label: 'Rocket', bg: '#9f1239', fg: '#fff1f2',
28
+ shapes: [
29
+ { tag: 'path', d: 'M14.5 5.2c2.2-2.2 4.8-2 5.3-1.8.2.5.4 3.1-1.8 5.3l-5.5 5.5-4.2-4.2Z' },
30
+ { tag: 'path', d: 'm11.2 6.8-4.1.6-2.5 2.5 4.1.7' },
31
+ { tag: 'path', d: 'm16.4 12-1 4.7-2.5 2.5-.7-4.1' },
32
+ { tag: 'circle', cx: 16.1, cy: 7.1, r: 1.2 },
33
+ { tag: 'path', d: 'M7.6 14.6c-2.2.4-3.4 1.6-3.6 3.8 2.2-.2 3.4-1.4 3.8-3.6' },
34
+ ],
35
+ },
36
+ {
37
+ id: 'compass', label: 'Compass', bg: '#1d4ed8', fg: '#eff6ff',
38
+ shapes: [
39
+ { tag: 'circle', cx: 12, cy: 12, r: 8.5 },
40
+ { tag: 'path', d: 'm15.2 8.8-1.8 4.6-4.6 1.8 1.8-4.6Z' },
41
+ ],
42
+ },
43
+ {
44
+ id: 'terminal', label: 'Terminal', bg: '#3f3f46', fg: '#fafafa',
45
+ shapes: [
46
+ { tag: 'rect', x: 3.5, y: 4.5, width: 17, height: 15, rx: 2 },
47
+ { tag: 'path', d: 'm7 9 3 3-3 3' },
48
+ { tag: 'path', d: 'M12.5 15H17' },
49
+ ],
50
+ },
51
+ {
52
+ id: 'package', label: 'Package', bg: '#6d28d9', fg: '#f5f3ff',
53
+ shapes: [
54
+ { tag: 'path', d: 'm12 3 8 4.5v9L12 21l-8-4.5v-9Z' },
55
+ { tag: 'path', d: 'm4.3 7.7 7.7 4.4 7.7-4.4' },
56
+ { tag: 'path', d: 'M12 12.1V21' },
57
+ ],
58
+ },
59
+ {
60
+ id: 'database', label: 'Database', bg: '#a16207', fg: '#fefce8',
61
+ shapes: [
62
+ { tag: 'ellipse', cx: 12, cy: 6, rx: 7.5, ry: 3 },
63
+ { tag: 'path', d: 'M4.5 6v6c0 1.7 3.4 3 7.5 3s7.5-1.3 7.5-3V6' },
64
+ { tag: 'path', d: 'M4.5 12v6c0 1.7 3.4 3 7.5 3s7.5-1.3 7.5-3v-6' },
65
+ ],
66
+ },
67
+ {
68
+ id: 'spark', label: 'Spark', bg: '#c2410c', fg: '#fff7ed',
69
+ shapes: [
70
+ { tag: 'path', d: 'm12 3 1.5 5.2L19 10l-5.5 1.8L12 17l-1.5-5.2L5 10l5.5-1.8Z' },
71
+ { tag: 'path', d: 'm18.5 15 .7 2.2 2.3.8-2.3.8-.7 2.2-.7-2.2-2.3-.8 2.3-.8Z' },
72
+ ],
73
+ },
74
+ ])
75
+
76
+ const BY_ID = new Map(IDENTITY_PRESETS.map((preset) => [preset.id, preset]))
77
+ const ICONIFY_ID = /^[a-z0-9-]+[:/][a-z0-9-]+$/i
78
+ const ALIASES = new Map([
79
+ ['rocket', 'mdi:rocket-launch'],
80
+ ['mdi/rocket-launch', 'mdi:rocket-launch'],
81
+ ['layers', 'gateway'],
82
+ ['default', 'spexcode'],
83
+ ])
84
+
85
+ export const IDENTITY_PRESET_IDS = Object.freeze(IDENTITY_PRESETS.map((preset) => preset.id))
86
+
87
+ export function resolvedIdentityIcon(value, fallback = DEFAULT_PROJECT_ICON) {
88
+ const raw = typeof value === 'string' ? value.trim() : ''
89
+ const id = ALIASES.get(raw) || raw
90
+ return id || fallback
91
+ }
92
+
93
+ export function identityPreset(value) {
94
+ const raw = typeof value === 'string' ? value.trim() : ''
95
+ return BY_ID.get(ALIASES.get(raw) || raw) || null
96
+ }
97
+
98
+ export function isIconifyIcon(value) {
99
+ return ICONIFY_ID.test(typeof value === 'string' ? value.trim() : '')
100
+ }
101
+
102
+ export function requireIdentityChoice(value) {
103
+ const raw = typeof value === 'string' ? value.trim() : ''
104
+ const id = ALIASES.get(raw) || raw
105
+ if (BY_ID.has(id)) return id
106
+ if (isIconifyIcon(id)) return id.replace('/', ':')
107
+ throw new Error(`unknown identity icon '${raw}' (choose a preset or Iconify prefix:name)`)
108
+ }
109
+
110
+ function attrs(shape) {
111
+ return Object.entries(shape).filter(([key]) => key !== 'tag')
112
+ .map(([key, value]) => `${key === 'className' ? 'class' : key}="${String(value).replaceAll('&', '&').replaceAll('"', '"')}"`).join(' ')
113
+ }
114
+
115
+ export function identitySvg(value, fallback = DEFAULT_PROJECT_ICON) {
116
+ const preset = identityPreset(value) || identityPreset(fallback)
117
+ const geometry = preset.shapes.map((shape) => `<${shape.tag} ${attrs(shape)}/>`).join('')
118
+ return `<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><rect x="1" y="1" width="22" height="22" rx="5" fill="${preset.bg}"/><g fill="none" stroke="${preset.fg}" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round">${geometry}</g></svg>`
119
+ }
120
+
121
+ export function identityFaviconHref(value, fallback = DEFAULT_PROJECT_ICON) {
122
+ const resolved = resolvedIdentityIcon(value, fallback)
123
+ if (identityPreset(resolved)) return `data:image/svg+xml,${encodeURIComponent(identitySvg(resolved, fallback))}`
124
+ if (/^https?:\/\//.test(resolved)) return resolved
125
+ if (isIconifyIcon(resolved)) return `https://api.iconify.design/${resolved.replace(':', '/')}.svg`
126
+ const glyph = `<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 100 100"><text x="50" y=".86em" font-size="82" text-anchor="middle">${resolved.replaceAll('&', '&amp;').replaceAll('<', '&lt;')}</text></svg>`
127
+ return `data:image/svg+xml,${encodeURIComponent(glyph)}`
128
+ }
@@ -13,20 +13,22 @@ import { resolveForgeHost } from '../../spec-forge/src/drivers.js'
13
13
  import { summarize } from './mentions.js'
14
14
  import { resolveLayout, mainBranch } from './layout.js'
15
15
  import { getBoardJson } from './graphCache.js'
16
- import { boardStream, notifyBoardChanged } from './graphStream.js'
16
+ import { boardStream, ensureBoardFileWatchers, notifyBoardChanged } from './graphStream.js'
17
17
  import { gitA, gitTry, repoRoot } from './git.js'
18
- import { newSession, listSessions, sendText, rawKey, stopSession, closeSession, resumeSession, mergeSession, reviewPayload, captureSessionResult, sessionPrompt, sessionGraph, registerWatch, deregisterWatch, renameSession, setSessionSort, superviseQueue } from './sessions.js'
18
+ import { listSessions, sendText, interruptSession, rawKey, stopSession, closeSession, resumeSession, mergeSession, reviewPayload, captureSessionResult, sessionPrompt, sessionGraph, registerWatch, deregisterWatch, renameSession, setSessionSort, sessionCreateRequest, superviseQueue, TMUX_SOCK } from './sessions.js'
19
19
  import { superviseTimeline, readTimeline } from './session-timeline.js'
20
- import { defaultHarness, HARNESSES, launcherList, launcherDefault } from './harness.js'
20
+ import { readSessionMessages, sessionMessageStream } from './message-stream.js'
21
+ import { defaultHarness, HARNESSES, dashboardLauncherList, launcherDefault } from './harness.js'
21
22
  import { evalTimeline, readBlobByHash } from '../../spec-eval/src/evaltab.js'
22
23
  import { putBlob } from '../../spec-eval/src/cache.js'
23
- import { evalNodes } from '../../spec-eval/src/scenarios.js'
24
24
  import { fileHumanReading } from '../../spec-eval/src/filing.js'
25
25
  import { fileHumanOk } from '../../spec-eval/src/humanok.js'
26
- import { buildExportModel, renderExportHtml, buildSessionEvals } from '../../spec-eval/src/sessioneval.js'
26
+ import { buildExportModel, renderExportHtml, SessionEvalUnavailableError } from '../../spec-eval/src/sessioneval.js'
27
27
  import { saveUpload, MAX_UPLOAD_BYTES } from './uploads.js'
28
- import { attachViewer, detachViewer, resizeBridge, forwardWheel, superviseBridges, type Viewer } from './pty-bridge.js'
28
+ import { attachViewer, detachViewer, resizeBridge, hideViewer, forwardInput, superviseBridges, type Viewer } from './pty-bridge.js'
29
29
  import { installProcessGuards } from './resilience.js'
30
+ import { resolveProjectIdentity } from './project-identity.js'
31
+ import { evalDetailReview, evalsReview, issuesReview } from './reviews.js'
30
32
 
31
33
  // last-resort net: an unforeseen async throw (e.g. a worktree vanishing mid-read during a worker
32
34
  // self-merge) is logged and the server KEEPS SERVING instead of exiting and dropping the public port.
@@ -34,12 +36,34 @@ installProcessGuards()
34
36
 
35
37
  const app = new Hono()
36
38
  app.use('/api/*', cors())
39
+ app.onError((error, c) => {
40
+ if (error instanceof SessionEvalUnavailableError) return c.json({ error: error.message }, 503)
41
+ console.error(error)
42
+ return c.text('Internal Server Error', 500)
43
+ })
37
44
  const { injectWebSocket, upgradeWebSocket } = createNodeWebSocket({ app })
38
45
 
39
46
  app.get('/', (c) => c.text('spec-cli — GET /api/graph · /api/specs · /api/specs/:id/history · /api/settings · /api/sessions · /api/slash-commands'))
40
47
  // the supervisor's readiness gate (supervise.ts): a bare git-free 200 so a booting child reports ready the
41
48
  // instant Hono is listening. Not under /api/* — loopback-only (supervisor→child), no CORS needed.
42
49
  app.get('/health', (c) => c.text('ok'))
50
+ // @@@ instance identity - who THIS backend is: the serve generation's instanceId (minted by the supervisor,
51
+ // constant across zero-downtime reloads, handed down via env) and the project root it serves. This is the
52
+ // answer the host gateway ([[host-gateway]]) compares an endpoint record against before proxying to it — a
53
+ // recycled port serving another project or a stale record fails the match instead of being routed to. Git-free
54
+ // after the first memoized resolution; a self-run child (no supervisor) answers instanceId:null, which no
55
+ // record claims, so it is simply not hosted.
56
+ const instanceStartedAt = new Date().toISOString()
57
+ app.get('/api/instance', (c) => {
58
+ const root = repoRoot()
59
+ return c.json({
60
+ instanceId: process.env.SPEXCODE_INSTANCE_ID ?? null,
61
+ root,
62
+ identity: resolveProjectIdentity(root, root),
63
+ pid: process.pid,
64
+ startedAt: instanceStartedAt,
65
+ })
66
+ })
43
67
  // the assembled graph (merged tree + overlay + sessions) — the dashboard's single source. Same data
44
68
  // as `spex graph --json`; the frontend only adds x/y pixels on top. Freshness is PUSH-first ([[graph-stream]]): the
45
69
  // dashboard reloads on a `/api/graph/stream` event, not a tight poll, so the route is a conditional-request
@@ -53,6 +77,7 @@ app.get('/health', (c) => c.text('ok'))
53
77
  // is bounded by [[graph-cache]]'s own build watchdog, so the next poll retries a fresh build.
54
78
  const BOARD_TIMEOUT_MS = Number(process.env.SPEXCODE_BOARD_TIMEOUT_MS || 20000)
55
79
  app.get('/api/graph', etag(), async (c) => {
80
+ ensureBoardFileWatchers()
56
81
  const timeout = Symbol('timeout')
57
82
  const json = await Promise.race([getBoardJson(), new Promise<typeof timeout>((r) => setTimeout(() => r(timeout), BOARD_TIMEOUT_MS))])
58
83
  if (json === timeout) return c.json({ error: 'graph build timed out' }, 503)
@@ -64,23 +89,9 @@ app.get('/api/graph/stream', (c) => boardStream(c))
64
89
  app.get('/api/specs', async (c) => c.json(await loadSpecs()))
65
90
  // the search corpus ([[graph-lean]]): a filesystem-only {id,title,path,desc,body} for every node, NO git. The
66
91
  // board omits `body` to stay lean, so the search palette fetches this ONCE when it opens (cached client-side)
67
- // to rank nodes over their prose — off the board's hot poll. A literal segment, before the `:id` routes.
68
- // Scenario prose rides the same corpus: the board's `scenarios` fold is slim ({name, tags, test}), so a measurable
69
- // node's row here carries its declared scenarios' description/expected (+ test reference and per-scenario code) — one fetch
70
- // serves both the palette's scenario plane and the focus-panel preview.
71
- app.get('/api/specs/lite', (c) => {
72
- const scByNode = new Map(evalNodes(repoRoot()).map((y) => [y.id, y.scenarios]))
73
- return c.json(loadSpecsLite().map((row) => {
74
- const sc = scByNode.get(row.id)
75
- return sc?.length
76
- ? { ...row, scenarios: sc.map((s) => ({
77
- name: s.name, description: s.description, expected: s.expected,
78
- ...(s.test ? { test: s.test } : {}),
79
- ...(s.code?.length ? { code: s.code } : {}),
80
- })) }
81
- : row
82
- }))
83
- })
92
+ // to rank nodes over their prose — off the board's hot poll. Review rows, including scenarios, come only
93
+ // from their paged endpoints and cannot be reconstructed from this corpus.
94
+ app.get('/api/specs/lite', (c) => c.json(loadSpecsLite()))
84
95
  // one node's body + parsed parts ([[graph-lean]]): the board no longer ships either, so the detail view
85
96
  // fetches this when a node opens. 404 for an unknown id.
86
97
  app.get('/api/specs/:id/content', (c) => {
@@ -136,9 +147,11 @@ app.post('/api/specs/:id/evals/ok', async (c) => {
136
147
  // else the bytes with a sniffed MIME and an immutable cache header (the name IS the content hash).
137
148
  // HTTP Range is honored — a <video> can only SEEK when the server answers byte ranges (a browser clamps
138
149
  // currentTime to the seekable window, which stays [0,0] without them); one general mechanism at the
139
- // transport, so every evidence kind streams the same way.
150
+ // transport, so every evidence kind streams the same way. A trailing `.<ext>` on the hash is IGNORED
151
+ // decoration for third-party markdown renderers (GitLab/GitHub only emit a <video> player when the URL
152
+ // ends in a video extension); the served bytes and MIME stay the stored ones — a wrong suffix never lies.
140
153
  app.get('/api/evidence/:hash', (c) => {
141
- const r = readBlobByHash(c.req.param('hash'))
154
+ const r = readBlobByHash(c.req.param('hash').replace(/\.[a-z0-9]+$/i, ''))
142
155
  if (!r.ok) return c.text(r.message, r.reason === 'invalid' ? 400 : 404)
143
156
  const total = r.bytes.length
144
157
  const base = { 'Content-Type': r.mime, 'Cache-Control': 'public, max-age=31536000, immutable', 'Accept-Ranges': 'bytes' }
@@ -163,17 +176,23 @@ app.post('/api/evidence', async (c) => {
163
176
  })
164
177
  // the SETTINGS read surface — one route for everything spexcode.json / spexcode.local.json resolves to:
165
178
  // `layout` (resolveLayout()'s main/worktrees/branch shape — the write-guard's project-identity probe reads
166
- // `.layout.main`) and the named launcher profiles ([[launcher-select]]) the New-Session form's dropdown
167
- // offers — `{ name, harness }` only (the `cmd` is a host secret, never shipped to the browser) — plus the
168
- // configured `default` NAME so the dropdown pre-selects the SAME launcher a bare `spex session new` uses
169
- // (the CLI/config default), instead of the alphabetically-first one. Missing defaultLauncher is returned
170
- // as an actionable config error, not hidden by falling through to the built-in `claude` launcher.
179
+ // `.layout.main`) and the dashboard-visible launcher profiles ([[launcher-visibility]]) the New-Session picker
180
+ // offers — `{ name, harness, cmd, headless }`: the cmd is read-only display data for the picker (the dashboard sits
181
+ // behind the gateway auth; the browser can read but never edit config) plus the configured `default` NAME
182
+ // so the picker pre-selects the SAME launcher a bare `spex session new` uses when that row is visible, else
183
+ // its first visible row rather than a hidden headless default,
184
+ // Missing defaultLauncher is returned as an actionable config error, not hidden by falling through to the
185
+ // built-in `claude` launcher.
186
+ // `tmuxSocket` is the `-L <name>` label our private tmux server runs under (a backend fact, env-overridable),
187
+ // so the row's attach modal ([[attach-menu]]) can offer the RAW `tmux -L <socket> attach -t <id>` fallback
188
+ // beside the blessed `spex session attach` command — the frontend never hardcodes the socket.
171
189
  app.get('/api/settings', async (c) => c.json({
172
190
  layout: await resolveLayout(),
173
- launchers: launcherList().map(({ name, harness }) => ({ name, harness })),
191
+ launchers: dashboardLauncherList(),
192
+ tmuxSocket: TMUX_SOCK,
174
193
  ...launcherDefault(),
175
194
  }))
176
- // the `surface: command` plugin-root nodes (built/active only) for the new-session `/` dropdown — each with
195
+ // the `surface: command` plugin-root nodes (built/active only) for new-session and live-inbox `/` dropdowns — each with
177
196
  // its prompt `body` ({{targets}} placeholder), `kind`, and folder `dir` + co-located `files`. surface is a
178
197
  // frontmatter field, not a dir (specs.ts loadSurface); `surface: system` siblings are gathered elsewhere.
179
198
  // `?surface=review` lists the review-track presets instead ([[review-commands]] — the eval detail's
@@ -183,12 +202,24 @@ app.get('/api/plugins', (c) => c.json(c.req.query('surface') === 'review' ? load
183
202
  // (local threads + the resident forge slice), the SAME mergedIssues() the CLI drain reads, verbatim
184
203
  // (the dashboard computes nothing over it: no re-sort, no salience ranking). The `enabled` flag mirrors
185
204
  // the issues-workflow on/off switch so the frontend hides the view when the feature is OFF.
186
- app.get('/api/issues', etag(), (c) =>
187
- c.json({
188
- enabled: issuesEnabled(),
189
- stores: issueStores(),
190
- issues: mergedIssues({ host: resolveForgeHost(), state: residentForgeState() }, loadSpecsLite().map((s) => s.id)),
191
- }))
205
+ app.get('/api/issues', etag(), async (c) => c.json(await issuesReview(c.req.query('q'), c.req.query('page'))))
206
+ // Evals uses the identical paged-review response. `scope:` inside q selects the worktree source; without
207
+ // it the source is the current cached board. Filtering/counts always precede the one 25-row slice.
208
+ app.get('/api/evals', etag(), async (c) => {
209
+ ensureBoardFileWatchers()
210
+ const page = await evalsReview(c.req.query('q'), c.req.query('page'), { view: c.req.query('view') })
211
+ return page ? c.json(page) : c.json({ error: 'no such review source' }, 404)
212
+ })
213
+ // ONE bounded detail response for both source roots: the selected scenario's complete A/B history and at
214
+ // most five lightweight neighbors. It never serializes another scenario's history or the scoped model.
215
+ app.get('/api/evals/detail', etag(), async (c) => {
216
+ ensureBoardFileWatchers()
217
+ const node = c.req.query('node')?.trim()
218
+ const scenario = c.req.query('scenario')?.trim()
219
+ if (!node || !scenario) return c.json({ error: 'node and scenario are required' }, 400)
220
+ const detail = await evalDetailReview(node, scenario, c.req.query('scope')?.trim() || null)
221
+ return detail ? c.json(detail) : c.json({ error: 'no such review source' }, 404)
222
+ })
192
223
  // the single-thread read ([[issues]]) behind `spex issue show <id>` — the SAME findIssue lookup, from the
193
224
  // resident forge slice (instant view, background reconcile — the list route's freshness contract). A local
194
225
  // id, or a forge id (`<host>#<n>`); unknown → 404 (eval-remark threads are not issues, so they 404 here too).
@@ -330,7 +361,7 @@ app.post('/api/remarks/:action{resolve|retract}', async (c) => {
330
361
  return c.json({ error: String((e as Error).message || e) }, 400)
331
362
  }
332
363
  })
333
- // the dashboard input's `/` dropdown — computed by the launcher's HARNESS adapter the same way that harness
364
+ // the harness slice of the dashboard input's `/` dropdown — computed by the launcher's HARNESS adapter the same way that harness
334
365
  // computes its own `/` menu ([[harness-adapter]]). The client passes `?harness=<id>` for the ACTIVE session,
335
366
  // so a codex tab gets CODEX's menu, not the default's; unknown/absent → default. Insert-only on the client.
336
367
  app.get('/api/slash-commands', (c) => {
@@ -370,18 +401,9 @@ app.post('/api/sessions/edges/unwatch', async (c) => {
370
401
  return c.json({ ok }, ok ? 200 : 404)
371
402
  })
372
403
  app.post('/api/sessions', async (c) => {
373
- const body = await c.req.json().catch(() => ({}))
374
- const prompt = typeof body?.prompt === 'string' ? body.prompt : ''
375
- if (!prompt.trim()) return c.json({ error: 'empty prompt' }, 400)
376
- if (typeof body?.harness === 'string') return c.json({ error: 'harness is not a create-session input; use launcher' }, 400)
377
- // the named launcher ([[launcher-select]]) — fixes the session's harness AND its persisted launch command.
378
- const launcher = typeof body?.launcher === 'string' && body.launcher.trim() ? body.launcher.trim() : undefined
379
- // parent = the spawning session's id, resolved by the CALLER (createSession) in its own process and passed
380
- // through here ([[session-nesting]]); the browser's New Session omits it → a top-level session.
381
- const parent = typeof body?.parent === 'string' && body.parent.trim() ? body.parent.trim() : null
382
- try {
383
- return c.json(await newSession(typeof body?.node === 'string' ? body.node : null, prompt, parent, launcher), 201)
384
- } catch (e) { return c.json({ error: String((e as Error).message || e) }, 400) } // unknown launcher id → 400, not a 500
404
+ const body = await c.req.json().catch(() => null)
405
+ const result = await sessionCreateRequest(body)
406
+ return result.status === 201 ? c.json(result.session, 201) : c.json({ error: result.error }, 400)
385
407
  })
386
408
  // one server-side merge bundle (ahead/dirty/diff(merge-base)/gates/proposal) for the manager cockpit;
387
409
  // dashboard and `spex session review` are thin callers. 404 for an unknown id. See [[manager-cockpit]].
@@ -389,18 +411,14 @@ app.get('/api/sessions/:id/review', async (c) => {
389
411
  const r = await reviewPayload(c.req.param('id'))
390
412
  return r ? c.json(r) : c.json({ error: 'no such session' }, 404)
391
413
  })
392
- // the ONE session eval read ([[session-eval]]): default = the lean interactive model — worktree-rooted
393
- // rows only, no diff enrichment, no inlined bytes (evidence streams lazily from /api/evidence), each
394
- // reading carrying `inSession` so the tab leads with what THIS session measured. `?format=html` = the
395
- // EXPORT artifact: one self-contained HTML (diff + gates + evidence inlined as data-URIs) for
396
- // CI/share/bare-browser — a heavier REPRESENTATION of the same read, not a second route. 404 unknown id.
414
+ // The self-contained HTML is the sole full-model transport exception. Interactive rows, including the CLI,
415
+ // use /api/evals pages; a bare request fails loudly rather than reopening a hidden full JSON path.
397
416
  app.get('/api/sessions/:id/evals', async (c) => {
398
417
  if (c.req.query('format') === 'html') {
399
418
  const m = await buildExportModel(c.req.param('id'))
400
419
  return m ? c.html(renderExportHtml(m)) : c.text('no such session', 404)
401
420
  }
402
- const m = await buildSessionEvals(c.req.param('id'))
403
- return m ? c.json(m) : c.json({ error: 'no such session' }, 404)
421
+ return c.json({ error: 'interactive eval rows use /api/evals pagination; use ?format=html only for export' }, 400)
404
422
  })
405
423
  // the session's live pane as text (one-shot snapshot) for a backend client (`spex session show --capture`). Empty and fail
406
424
  // stay distinct: an empty pane is 200 with empty body; unknown id → 404, offline (no live pane) → 409, error → 502.
@@ -419,6 +437,13 @@ app.get('/api/sessions/:id/timeline', (c) => {
419
437
  const r = readTimeline(c.req.param('id'), Number.isFinite(limit) && limit > 0 ? limit : undefined)
420
438
  return r ? c.json(r) : c.json({ error: 'no such session' }, 404)
421
439
  })
440
+ // An adapter may expose its native event log as the console's optional full-process drill-down. REST establishes
441
+ // the complete ordered snapshot + byte cursor; SSE follows appends from that cursor (or Last-Event-ID on reconnect).
442
+ app.get('/api/sessions/:id/messages', (c) => {
443
+ const r = readSessionMessages(c.req.param('id'))
444
+ return r ? c.json(r) : c.json({ error: 'no such session' }, 404)
445
+ })
446
+ app.get('/api/sessions/:id/messages/stream', (c) => sessionMessageStream(c))
422
447
  // the session RECORD detail (`spex session show`): the board row (status · node · branch · launcher · …)
423
448
  // plus the full originating prompt (the row itself carries only the preview). One id-addressed read backs
424
449
  // the CLI's show; 404 for an unknown id.
@@ -445,44 +470,88 @@ app.post('/api/sessions/:id/merge', async (c) => {
445
470
  return c.json(r, r.dispatched ? 200 : 409)
446
471
  })
447
472
 
448
- // one WS over a shared tmux control-mode client (pty-bridge): server→client = raw pane bytes (binary); the
449
- // view takes no keyboard input, so client→server is only a text control frame — {t:'resize',cols,rows} or
450
- // {t:'wheel',…}. The bridge resolves the wheel against tmux pane state: copy-mode repaint for normal panes,
451
- // SGR mouse report injection for mouse-owning TUIs. A real tmux client, so the first paint is one coherent
452
- // frame and live bytes arrive as events.
473
+ // one WS owns one native tmux client (pty-bridge): server→client = that client's rendered PTY bytes (binary);
474
+ // client→server text controls resize, visibility, and xterm-native input (which carries the mouse/wheel
475
+ // SGR reports xterm produces natively in mouse-report mode). Server→client text commits a completed resize immediately before its binary tmux transaction;
476
+ // hiding starts that viewer's bounded helper release without closing the warm socket. tmux itself resolves wheel input between
477
+ // copy-mode and a mouse-owning TUI. The bridge never splices capture-pane state into this stream.
478
+ // keep-alive ping cadence for the terminal socket — the server half of [[reconnect]]'s heartbeat contract,
479
+ // and the contract's ONE primitive number: the client mirrors it (SERVER_PING_MS in the dashboard's
480
+ // resilientSocket.js, pinned by its test) and DERIVES its silence deadline (2.5×) from it.
481
+ // A healthy link is guaranteed inbound traffic every PING window, so the client may presume an OPEN socket
482
+ // silent past its derived window dead. The same tick also sends a WebSocket protocol ping; browsers answer its
483
+ // pong below JavaScript, so a backend reload stays compatible with a tab running the previous frontend bundle.
484
+ // The server owns that mirror deadline and detaches the viewer itself when a half-open link never reports close.
485
+ // Terminal pixels remain binary frames, so heartbeat controls never enter xterm.
486
+ const TERM_PING_MS = 10000
487
+ const TERM_DEAD_MS = 2.5 * TERM_PING_MS
453
488
  app.get('/api/sessions/:id/socket', upgradeWebSocket((c) => {
454
489
  const id = c.req.param('id') as string
455
- // the size-first handshake: a client that already knows its pane size carries it as ?cols=&rows= so the
456
- // first frame is drawn at the true size. Absent/garbage → undefined, and the bridge falls back to prewarm.
457
- const qc = Number(c.req.query('cols')), qr = Number(c.req.query('rows'))
458
- const initialSize = qc > 0 && qr > 0 ? { cols: qc, rows: qr } : undefined
459
490
  let viewer: Viewer | null = null
491
+ let ping: ReturnType<typeof setInterval> | undefined
492
+ let pongDeadline: ReturnType<typeof setTimeout> | undefined
493
+ let cleaned = false
494
+ const disarmPongDeadline = () => { if (pongDeadline) clearTimeout(pongDeadline); pongDeadline = undefined }
495
+ let armPongDeadline = () => {}
496
+ const cleanup = () => {
497
+ if (cleaned) return
498
+ cleaned = true
499
+ if (ping) clearInterval(ping)
500
+ disarmPongDeadline()
501
+ if (viewer) detachViewer(id, viewer)
502
+ viewer = null
503
+ }
460
504
  return {
461
505
  onOpen(_evt, ws) {
462
- viewer = { send: (buf) => { try { ws.send(Uint8Array.from(buf)) } catch { /* viewer gone */ } } }
463
- if (!attachViewer(id, viewer, initialSize)) { try { ws.close() } catch { /* already closed */ } }
506
+ viewer = {
507
+ send: (buf) => { try { ws.send(Uint8Array.from(buf)) } catch { /* viewer gone */ } },
508
+ commitSize: (cols, rows) => { try { ws.send(JSON.stringify({ t: 'resize-commit', cols, rows })) } catch { /* viewer gone */ } },
509
+ }
510
+ attachViewer(id, viewer)
511
+ armPongDeadline = () => {
512
+ if (cleaned) return
513
+ disarmPongDeadline()
514
+ pongDeadline = setTimeout(() => {
515
+ cleanup()
516
+ try { ws.close() } catch { /* cleanup already detached the dead viewer */ }
517
+ }, TERM_DEAD_MS)
518
+ pongDeadline.unref()
519
+ }
520
+ // `raw` is @hono/node-ws's real ws.WebSocket. Protocol pong is intentionally the server-side liveness
521
+ // signal: unlike an application text reply, every browser generation answers it automatically.
522
+ ws.raw.on('pong', armPongDeadline)
523
+ armPongDeadline()
524
+ ping = setInterval(() => {
525
+ try { ws.raw.ping() } catch { /* viewer gone; onClose reaps */ }
526
+ try { ws.send('ping') } catch { /* client dead-man still needs observable inbound traffic */ }
527
+ }, TERM_PING_MS)
464
528
  },
465
529
  onMessage(evt) {
466
530
  if (!viewer) return
467
531
  const data = evt.data
468
- // no keyboard input: the only client→server messages are the resize frame and the wheel frame. Binary
469
- // is ignored; pane navigation stays inside the tmux bridge instead of becoming browser scroll state.
532
+ // Binary input is ignored; JSON keeps terminal input distinct from binary pane output while preserving
533
+ // xterm's ordered string exactly. The bridge accepts input only from this viewer's visible claim.
470
534
  if (typeof data === 'string') {
535
+ if (data === 'pong') {
536
+ armPongDeadline()
537
+ return
538
+ }
471
539
  try {
472
540
  const m = JSON.parse(data)
473
- if (m?.t === 'resize') resizeBridge(id, Number(m.cols), Number(m.rows), !!m.full)
474
- else if (m?.t === 'wheel') forwardWheel(id, !!m.up, Number(m.col), Number(m.row), Number(m.ticks))
541
+ if (m?.t === 'resize') resizeBridge(id, viewer, Number(m.cols), Number(m.rows))
542
+ else if (m?.t === 'visible' && m.visible === false) hideViewer(id, viewer)
543
+ else if (m?.t === 'input' && typeof m.data === 'string') forwardInput(id, viewer, m.data)
475
544
  } catch { /* ignore */ }
476
545
  }
477
546
  },
478
- onClose() { if (viewer) detachViewer(id, viewer) },
547
+ onClose() { cleanup() },
479
548
  }
480
549
  }))
481
550
  // ONE input route, `kind` the discriminator — the transport split is an implementation fact, not API surface.
482
- // kind:"text" (the docked ❯ line, `spex session send`, the server-side merge dispatch) injects a whole prompt
551
+ // kind:"text" (Command Box, `spex session send`, the server-side merge dispatch) injects a whole prompt
483
552
  // through the rendezvous control socket — socket-only + fail-loud: a prompt the agent doesn't confirm
484
553
  // accepting returns 502 with the reason (never a silent 200), so a dead dispatch is seen, not a false success.
485
- // kind:"keys" is the LAST-RESORT raw face (`send --keys`, the dashboard's type mode): an ORDERED BATCH of
554
+ // kind:"keys" is the LAST-RESORT raw face (`spex session send --keys`): an ORDERED BATCH of
486
555
  // nav-mode key tokens over tmux send-keys, delivered in array order so tap order survives
487
556
  // ([[nav-mode-key-ordering]]); unstable by nature — callers try a plain text send first. An unknown kind is a
488
557
  // loud 400, never a guessed channel.
@@ -506,6 +575,10 @@ app.post('/api/sessions/:id/input', async (c) => {
506
575
  // soft stop: kill the agent's tmux + socket but KEEP the worktree (resumable). Distinct from close, which
507
576
  // removes the worktree. {ok:false} = no such session.
508
577
  app.post('/api/sessions/:id/stop', async (c) => c.json({ ok: await stopSession(c.req.param('id')) }))
578
+ app.post('/api/sessions/:id/interrupt', async (c) => {
579
+ const result = await interruptSession(c.req.param('id'))
580
+ return c.json(result, result.ok ? 200 : 502)
581
+ })
509
582
  app.post('/api/sessions/:id/close', async (c) => c.json({ ok: await closeSession(c.req.param('id')) }))
510
583
  // set (or clear, with a blank) a session's display-name override; persists to the session's global record
511
584
  // (`session.json`) so it survives a restart. Unknown id → 404. That record sits INSIDE the watched store, but
@@ -530,23 +603,21 @@ app.post('/api/sessions/:id/sort', async (c) => {
530
603
  const port = Number(process.env.PORT || 8787)
531
604
  // @@@ server-side connection reaping ([[spec-cli]]) - abandoned connections must die SERVER-SIDE, or they
532
605
  // pile up and wedge the backend (135 leaked conns once starved :8787 into looking dead — the cascade that
533
- // triggered the mass-restore incident, since every client-side timeout-kill leaks one). The `serverOptions`
534
- // timeouts below are kept (harmless), but they are NOT the mechanism: MEASURED (eval
535
- // server-reaps-abandoned-connections), Node's `headersTimeout`/`requestTimeout` do NOT reap an INCOMPLETE
536
- // request via the connectionsCheckingInterval sweep a slow-loris survives indefinitely; only
537
- // keepAliveTimeout (idle-between-requests) ever fires. So the real reaper is the explicit socket-level
538
- // `installConnectionReaper` below (see reaper.ts): a per-socket deadline that reaps a slow-loris / idle
539
- // keep-alive but exempts an ACTIVE WS/SSE stream (board-stream, terminal socket) for as long as it streams.
606
+ // triggered the mass-restore incident, since every client-side timeout-kill leaks one). The ONE mechanism is
607
+ // the socket-level `installConnectionReaper` below (reaper.ts): a per-socket deadline that reaps a
608
+ // slow-loris / idle keep-alive but exempts an ACTIVE WS/SSE stream (board-stream, terminal socket) for as
609
+ // long as it streams. Deliberately NO `serverOptions` timeouts here: they were measured to be not harmless
610
+ // but a second mechanism racing the reaper (issue #65 a 20s headersTimeout won at default config and
611
+ // silently capped SPEXCODE_REAP_HEADER_MS); the install disables Node's overlapping timeouts so the
612
+ // deadlines have a single owner.
540
613
  // @@@ loopback bind ([[public-mode]]) - this child is NEVER the internet face: the supervisor (and in public
541
614
  // mode the gateway) fronts it, and dials it only via 127.0.0.1. Binding loopback is what makes "loopback is
542
615
  // the trust boundary" true — without a hostname Node binds all interfaces and the child is reachable from
543
616
  // the LAN with no password, bypassing the gate entirely (measured: eval auth-boundary).
544
- const server = serve({ fetch: app.fetch, port, hostname: '127.0.0.1', serverOptions: {
545
- keepAliveTimeout: 10000, headersTimeout: 20000, requestTimeout: 60000, connectionsCheckingInterval: 10000,
546
- } })
617
+ const server = serve({ fetch: app.fetch, port, hostname: '127.0.0.1' })
547
618
  installConnectionReaper(server as unknown as HttpServer)
548
619
  injectWebSocket(server)
549
- superviseBridges() // keep a warm tmux client per live session, so opening a tab is instant
620
+ superviseBridges() // restore visible helpers after failure; their viewer subscriptions survive replacement
550
621
  superviseQueue() // launch queued sessions as slots free (catches agent-authored proposals/crashes the server never sees directly)
551
622
  superviseTimeline() // record authored-lifecycle transitions to each session's durable timeline ([[session-timeline]])
552
623
  console.log(`spec-cli serving .spec (from git) on http://localhost:${port}`)