refactor(commands): move commands-side rendering out of src/daemon and retire the doctor progress flag - #2349

Completion evidence

Zero edges from src/commands / src/mcp into src/daemon

Query over pnpm depgraph output (node ids drop the src/ prefix; kind 1 is type-only, which the condition excludes):

const g = JSON.parse(fs.readFileSync(process.argv[2], 'utf8'));
const KIND = ['value', 'type', 'dynamic'];
const rows = g.edges
  .map(([from, to, kind]) => ({ from: g.nodes[from].id, to: g.nodes[to].id, kind: KIND[kind] }))
  .filter((e) =>
    (e.from.startsWith('commands/') || e.from.startsWith('mcp/')) &&
    e.to.startsWith('daemon/') && e.kind !== 'type');

27a97ee: value+dynamic edges src/commands|src/mcp -> src/daemon: 5
  dynamic  src/commands/capture/output.ts -> src/daemon/result-serialization.ts
  dynamic  src/commands/management/output.ts -> src/daemon/client/doctor-progress.ts
  dynamic  src/commands/management/output.ts -> src/daemon/handlers/doctor-output.ts
  dynamic  src/commands/management/output.ts -> src/daemon/result-serialization.ts
  dynamic  src/mcp/tool-result.ts -> src/daemon/result-serialization.ts

e4bc1ea: value+dynamic edges src/commands|src/mcp -> src/daemon: 0

(5 file edges = the 11 import() sites the issue inventories; the graph dedupes repeats between the same pair.)

Eager-closure budgets

scripts/__tests__/eager-closure-budgets.test.ts: 442 passed. Measured over all 225 entries the gate covers, before vs. after: no entry changed; src/cli.ts stays at 380 modules (daemon/client/doctor-progress.ts leaves the closure, commands/command-progress.ts enters it).

Measured delta for the two src/commands edges kept lazy: making them static takes src/cli.ts 380 → 384 (commands/output/result-serialization.ts, core/doctor-output.ts, core/status-markers.ts, core/deploy-result-target.ts), which the no-growth rule rejects — so the lazy edges stay and no budget was raised. The src/mcp/tool-result.ts edge is under no measured entry's closure, so it became static.

Doctor CLI output, three modes

Both sides drive readDaemonSocketProgressResponse and doctorCliOutput exactly as their own CLI wires them (main: no sink, the client renders and sets the global flag; here: the CLI's createStderrCommandProgressSink plus its state). Captured stderr + text + data, diff empty:

--- mode 1 progress streamed to stderr ---
stderr:
✓ agent-device: agent-device 0.17.9 using /tmp/agent-device
! device: No booted device.
text:
Doctor: pass
No blockers found.
--- mode 2 no progress ---
stderr:
text:
Doctor: pass
✓ agent-device: agent-device 0.17.9 using /tmp/agent-device
! device: No booted device.
  hint: Boot a simulator.
--- mode 3 custom sink ---
stderr:
text:
Doctor: pass
✓ agent-device: agent-device 0.17.9 using /tmp/agent-device
! device: No booted device.
  hint: Boot a simulator.
custom sink events:[{"type":"command",…},{"type":"command",…}]

Mode 1 and 2 also get an end-to-end CLI pin (src/__tests__/cli-network.test.ts), verified failing with the suppression branch disabled; mode 3 is pinned at both seams (command-progress.test.ts, management/output.test.ts).

Rename-only proof

git diff -M90% --stat origin/main...HEAD — 30 files, 407 insertions, 231 deletions:

 src/{daemon/handlers => core}/doctor-output.ts     |   0
 src/{daemon/handlers => core}/status-markers.ts    |   0
 src/{daemon => commands/output}/result-serialization.ts      |  21 +---
 src/{daemon/__tests__ => commands/output}/snapshot-serialization.test.ts |   2 +-

The two pure moves are content-identical. result-serialization.ts loses only the two resolver functions extracted to core/deploy-result-target.ts; its test file splits the same way (the 132-line delete / 102-line add pair), so both halves carry their existing assertions unchanged.

Generated by Claude Code

Size Report

Startup median (7 runs, lower is better):

Clean review at 29f3527. The shared helpers and serializers have clear owners, and the doctor progress state is local to each CLI run. Streamed output avoids duplicates; custom sinks still get the full result. The tests cover both streamed and unstreamed CLI output, and the old global flag is gone.

Ready for human review. Remaining CI is still running, so this is not yet a merge-ready claim.