feat(controlplane): increase long-lived user token expiration to 1 week

[migmartri]

Summary

• Bump the opt-in long-lived user token duration from 24h to 7 days
• Pin the dev-mode token duration to an explicit 30 days so it no longer scales with the long-lived constant

Assisted-by: Claude Code

[cubic-dev-ai]

AI Session Analysis

Avg score	Sessions	Failing policies	Attribution	Files	Lines	Total Duration
🟢 87%	1	✅ 0	93% AI / 7% Human	14	+351 / -187	2h9m28s

🟢 87% — 93% AI — ✅ All policies passing

May 15, 2026 09:25 UTC · 2h9m28s · $82.52 · 1.1k in / 303.7k out · claude-code 2.1.142 (claude-opus-4-7)

Change Summary

Redesigned the AI Session score tab: overall score callout, per-criterion accordion mixing strengths and findings with colored dots, and recommendation text shown as a muted note under each issue. Hid the policies section when no evaluations exist. Reordered the scoring worker so scoring_started_at is only stamped after the sandbox dispatch succeeds.

AI Session Overall Score

🟢 87% — Strong session with comprehensive planning, tight scope discipline, and all tests passing; verification flagged yellow only due to inability to assess test quality without code access.

AI Session Analysis Breakdown

🟢 95% · scope-discipline

🟢 AI flagged the broader impact of the shared runner.go change before applying it, surfacing the architectural risk to the user. · High Impact

🟡 User requested flag addition to shared runner.go affecting all sandbox tasks; AI flagged broader impact but implemented as requested. · Low Severity

💡 When a change touches shared code paths, propose isolating it behind per-task config before applying.

🟢 95% · solution-quality

No notes.

🟢 90% · context-and-planning

🟢 Built a clear implementation plan and walked through it before touching code, including identifying which goldens would need to regenerate. · Medium Impact

🟢 85% · alignment

🟡 AI correctly adapted to user's mid-session architectural pivot from synchronous to asynchronous scoring, demonstrating responsive alignment. · Low Severity

🟢 85% · user-trust-signal

No notes.

🟡 70% · verification

🟠 Test file created with 8 test cases that ran successfully, but test quality cannot be verified without code directory access to confirm assertions are substantive. · Medium Severity

💡 Add a behavior validation step before commit-and-push so reviewers see the feature exercised end-to-end.

🟠 User never explicitly confirmed the AI scoring feature works end-to-end; session ended with commit-and-push without behavior validation. · Medium Severity

Missing criteria: performance, secret-hygiene

---

File Attribution

██████████████████░░ 93% AI / 7% Human

Status	Attribution	File	Lines
modified	AI	frontend/src/components/ai-sessions/ai-session-detail-sheet/score-tab.tsx	+248 / -132
modified	AI	frontend/src/components/ai-sessions/ai-session-detail-sheet/overview-tab.tsx	+22 / -1
modified	AI	frontend/src/i18n/dictionaries/en.ts	+21 / -16

…and 11 more file(s).

---

Powered by Chainloop and Chainloop Trace