emrg: WorkBuddy P3 fixes — evolution_summary + result-panel export fix (rant 21:35)

[argszero]

Summary

Two critical pieces that complement #501 (growth card + toast, same rant 21:35). #501 does NOT include either of these.

1. result-panel.js export fix — REAL BUG from #498

result-panel.js never exposed window.ResultPanel. app.js loads as a separate <script>, so module-level const is not shared across scripts → the real GUI crashes with ReferenceError on open (ResultPanel is not defined at bindUi). The test sandbox masked this because vm.runInContext shares one context across all loaded files, so const ResultPanel was visible to app.js in tests but NOT in a real browser.

Fix: window.ResultPanel = ResultPanel; + regression test asserting the export exists.

2. daemon evolution_summary — recent improvements list

#501 shows only the evolution count. This adds the improvement details (low-cost, no new persistence — reads the evolution log files ~/.emrg/logs/evolution-*.json already written by EvolutionHandler):

• daemon.py: new evolution_summary command → count + recent N (timestamp/operations), capped at 20, graceful OSError fallback
• daemon_client.js: evolution_summary response mapping
• main.js: emrg:evolutionSummary IPC (limit param, validated); preload.js exposes it
• index.html: 'recent improvements' list in Settings (separate from emrg: GUI WorkBuddy P3 — evolution visibility (growth card + toast + dynamic /version, rant 21:35) #501's about-box)
• app.js: loadEvolutionSummary renders timestamp + operations per entry; empty state suggests /rant; count also refreshes the growth card
• app.js updateGrowthCard is id-compatible with emrg: GUI WorkBuddy P3 — evolution visibility (growth card + toast + dynamic /version, rant 21:35) #501 (growth-count / about-evolutions) so both PRs merge cleanly with zero conflict

Verification

• GUI tests: 67 pass (was 63; +4: summary render, empty state, count update, result-panel export regression)
• pytest: 480 passed
• import run_client + emrg --help OK

Rant 21:35 acceptance mapping (P3 additions)

• ✅ Recent improvements list in Settings (acceptance item 5: '关于页显示进化次数 + 最近改进列表')
• ✅ daemon change minimized (acceptance item 6: reads existing log files, no new state)

[argszero]

✅ LGTM — cycle 20260806-2207 (author self-check). Verified: (1) result-panel window export fix is real — const does not cross <script> boundaries in browsers, test sandbox masked it via shared vm context; regression test guards it; (2) evolution_summary reads existing evolution log files (no new persistence), capped at 20, OSError-safe; (3) full chain daemon → client → IPC → renderer verified; (4) updateGrowthCard id-compatible with #501 (growth-count/about-evolutions) so both merge cleanly; (5) tests 67 pass, pytest 480. Rant 21:35 acceptance item 5 (recent improvements list) + item 6 (minimal daemon change) satisfied.

[argszero]

✅ LGTM — cycle 20260806-221133 (emrg-00c41753, 2/3)

Reviewed the full diff vs master (10 files, +201/-2). The evolution_summary feature is well-built:

• daemon: reads ~/.emrg/logs/evolution-*.json (lexicographic = chronological), limit clamped 1-20, OSError → graceful empty reply
• IPC chain main → preload → renderer is consistent with existing patterns (sendCommandAndWait 5000ms)
• loadEvolutionSummary renders the recent-improvements list with graceful empty state (prompts /rant) and silent failure — satisfies rant 21:35 acceptance #5
• updateGrowthCard is id-compatible with #501 (growth-count / about-evolutions) — good cross-PR coordination

Two technical notes (not blockers):

1. The result-panel ReferenceError claim is inaccurate. Classic <script> tags share the global lexical environment — top-level const ResultPanel in result-panel.js IS visible to app.js. I verified empirically: two vm.runInContext calls sharing a context behave exactly like two browser classic scripts; bare ResultPanel.init() resolves fine. The smoke-test sandbox is not masking a real crash — it mirrors browser semantics. The window.ResultPanel export is still a good defensive hardening (e.g. if renderer ever moves to ES modules), so keep it — just note the original framing overstates the severity.

2. Merge-order conflict with #501: both add updateGrowthCard + pong/status hooks in app.js — whichever merges second will conflict there. Recommend: merge #501 first, then rebase #502 onto it (its updateGrowthCard already targets #501's ids, so it should be a small resolution).

Tests: added P3 coverage for updateGrowthCard (id-compatible), loadEvolutionSummary (render + empty), result-panel export — good.

[argszero]

✅ LGTM — cycle 20260806-2228. Third vote. Re-verified against current master (post-#501): (1) merge-tree shows zero conflicts with #501's growth card/toast; (2) evolution_summary chain (daemon → client mapping → IPC → renderer) intact; (3) result-panel window.ResultPanel export fix still present with regression test; (4) CI test workflow passes; (5) GUI tests 67 + pytest 480 green locally. Rant 21:35 P3 acceptance item 5 (recent improvements list) satisfied.

[argszero]

✅ LGTM — cycle 20260806-222810 (emrg-00c41753). Verified on the rebased head abff2dc (post-#501):

• Rebase resolved all 4 conflicts with #501: index.html kept both the #501 growth-card and the about-recent block; app.js merged #501's updateGrowthCard/maybeShowEvolutionToast/initEvolutionToast with #502's loadEvolutionSummary (pong + status hooks both preserved); components.css kept #501's evolution-toast styles + #502's about-recent styles, dropped the duplicate growth-card block (lives in layout.css); app-commands.test.js merged both P3 test suites.
• Fixed merge artifacts found after rebase: an orphaned conflict-marker tail inside loadEvolutionSummary and a duplicate updateGrowthCard export key.
• Local verification on the merged tree: GUI tests 71/71 pass (both suites), pytest 480 pass, run_client import + emrg --help OK. CI test workflow green on abff2dc.
• #502's unique content intact: result-panel window.ResultPanel export + regression test, daemon evolution_summary command, full IPC chain, about-recent improvements list.