feat: capture_work — braindump → decomposed issues → filed in gitea (P4)

The last big agent capability. In the Capture screen, a rough braindump runs real
big-model decomposition into a small, estimated issue set; you review/edit the
labels and approve, and the issues are opened in gitea. This is the one place the
big model earns its keep (docs/agent-tools.md).

core (@commitea/core):
- capture-work: PROPOSE_ISSUES_TOOL + CAPTURE_SYSTEM; captureWork(complete, dump)
  forces a single structured decomposition and returns validated issues; parseCaptureArgs
  drops blank titles + invalid est/p labels. ProposedIssue / CaptureProposal.
- gitea client: createIssue({title, body?, labelIds?}) → POST /issues, normalized.

app:
- model bridge model:capture runs captureWork on the (loaded) big model.
- gitea bridge gitea:createIssues opens each approved issue with its est/* + p/*
  labels (reusing the #41 label-id resolver — zero-pollution, no invented labels).
- Capture screen: when a model is configured, "Brew tickets" runs real capture and
  "Approve all" files the set; otherwise the scripted demo interview runs. Fixed a
  race — the brew handler re-checks model status at click time so a configured
  model never falls into the scripted path before status resolves.

Verified: 108 core tests green (7 capture + createIssue added), desktop typecheck
clean, 14 fixture e2e green. Gated live e2e against gemma-4-26b: the auth braindump
→ 3 real tickets ("Resolve token refresh + session staleness" est/3d p/1, "Fix
webhook double-firing" est/2d p/2, "Write auth setup docs" est/1d p/3), reviewable
and editable; Discard so the test files nothing (createIssue POST is unit-tested).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

packages/core/src/agent/capture-work.ts

@@ -0,0 +1,109 @@
+/**
+ * capture_work — braindump → a small set of concrete issues. This is the one
+ * place the big model earns its keep (decomposition + estimate negotiation).
+ * It returns a *proposal*; nothing is filed until the human approves it in the
+ * Capture tray and it goes through the create-issue write path. Pure
+ * orchestration over an injected `complete` — stubbable, so it's testable offline.
+ */
+
+import {
+  type EstimateLabel,
+  ESTIMATE_LABELS,
+  type PriorityLabel,
+  PRIORITY_LABELS,
+} from '../labels/label-schema.js'
+import type { ChatMessage, CompletionResult, ToolDecl } from './chat-client.js'
+
+export interface ProposedIssue {
+  title: string
+  body: string
+  estimate?: EstimateLabel
+  priority?: PriorityLabel
+}
+
+export interface CaptureProposal {
+  issues: ProposedIssue[]
+  /** One-line schedule impact, if the model offered one. */
+  consequence?: string
+}
+
+export const PROPOSE_ISSUES_TOOL: ToolDecl = {
+  name: 'propose_issues',
+  description: 'Return the decomposed issue set for a braindump. Call this exactly once.',
+  parameters: {
+    type: 'object',
+    properties: {
+      issues: {
+        type: 'array',
+        items: {
+          type: 'object',
+          properties: {
+            title: { type: 'string', description: 'a clear imperative title' },
+            body: { type: 'string', description: 'one or two lines of detail' },
+            estimate: { type: 'string', enum: ['est/1d', 'est/2d', 'est/3d', 'est/5d', 'est/8d'] },
+            priority: { type: 'string', enum: ['p/1', 'p/2', 'p/3', 'p/4'] },
+          },
+          required: ['title'],
+        },
+      },
+      consequence: { type: 'string', description: 'one-line note on the schedule impact' },
+    },
+    required: ['issues'],
+  },
+}
+
+export const CAPTURE_SYSTEM = [
+  'You are Reginald, decomposing a rough braindump into a small set of concrete Gitea issues.',
+  'Call propose_issues exactly once. Split genuinely separate work; merge trivially-coupled work; invent no scope.',
+  'Each issue gets an imperative title, a one-line body, an estimate (est/1d…8d) and a priority (p/1…4).',
+  'Estimate honestly — a "quick" task is rarely one day. Keep the set tight; three good issues beat eight vague ones.',
+].join(' ')
+
+function isEstimate(v: unknown): v is EstimateLabel {
+  return typeof v === 'string' && (ESTIMATE_LABELS as readonly string[]).includes(v)
+}
+function isPriority(v: unknown): v is PriorityLabel {
+  return typeof v === 'string' && (PRIORITY_LABELS as readonly string[]).includes(v)
+}
+
+/** Coerce the model's raw propose_issues args into a validated proposal. */
+export function parseCaptureArgs(args: unknown): CaptureProposal {
+  const a = (args ?? {}) as { issues?: unknown[]; consequence?: unknown }
+  const issues: ProposedIssue[] = []
+  for (const raw of Array.isArray(a.issues) ? a.issues : []) {
+    const r = (raw ?? {}) as Record<string, unknown>
+    const title = typeof r.title === 'string' ? r.title.trim() : ''
+    if (!title) continue
+    issues.push({
+      title,
+      body: typeof r.body === 'string' ? r.body : '',
+      estimate: isEstimate(r.estimate) ? r.estimate : undefined,
+      priority: isPriority(r.priority) ? r.priority : undefined,
+    })
+  }
+  return { issues, consequence: typeof a.consequence === 'string' ? a.consequence : undefined }
+}
+
+/**
+ * Run one decomposition turn. Forces the model to answer via propose_issues and
+ * returns the validated set. An empty set means the model declined to structure it.
+ */
+export async function captureWork(
+  complete: (messages: ChatMessage[], tools?: ToolDecl[]) => Promise<CompletionResult>,
+  braindump: string,
+): Promise<CaptureProposal> {
+  const res = await complete(
+    [
+      { role: 'system', content: CAPTURE_SYSTEM },
+      { role: 'user', content: braindump },
+    ],
+    [PROPOSE_ISSUES_TOOL],
+  )
+  const call = res.toolCalls.find((t) => t.name === 'propose_issues')
+  if (!call) return { issues: [] }
+  try {
+    return parseCaptureArgs(JSON.parse(call.arguments || '{}'))
+  } catch {
+    return { issues: [] }
+  }
+}