feat(ai-partner): daily proactive prompt generation (#1465)

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s

[CraigBuckmaster]

Summary

Resolves phase 4, card #1465 of epic #1446. Generates one "Amicus noticed..." proactive prompt per user per day. Client-triggered, server-generated, client-cached — no background jobs, no battery tax, zero cost when the app isn't opened.

Server (ai-proxy)

• New POST /ai/daily-prompt endpoint with the same auth + rate-limit gating as /ai/chat.
• Haiku 4.5 call with a prompt-cacheable system block; user-specific signals (profile, recent chapters, date) live in the user message so the system block is a stable cache key.
• Strict response validation — the model must return valid JSON with non-empty prompt_text and seed_query. Fenced output is tolerated.

Client (app)

• Migration v17 adds the singleton amicus_daily_prompt_cache table.
• getDailyPrompt() service flow:
  1. Read cache → short-circuit if (date, profile_hash) still match.
  2. Otherwise call proxy and write-through.
  3. On offline / 5xx / bad shape → keep whatever stale cache exists; return null if none.
• useDailyPrompt() hook for the HomeScreen card (ai-partner: home screen card component #1466 will consume it).

Acceptance criteria

• Worker endpoint generates valid Haiku response
• Response always includes prompt_text, seed_query, cached_until
• Client cache hit returns same prompt within 24h (same date + hash)
• Cache invalidates when profile_hash changes
• Offline / 5xx gracefully falls back to stale cache or returns null
• Proxy enforces auth + rate limit on this endpoint too
• Unit tests cover: validation, cache hit, cache miss, profile change invalidation, offline fallback, wrong-shape response
• No any types; lint clean

Test plan

• ai-proxy: npm test — 66 tests passing (17 new in dailyPrompt.test.ts)
• app: npx jest src/services/amicus/__tests__/dailyPrompt.test.ts — 11 passing
• Full app suite: 3402 passing
• Coverage thresholds met: 80.41 / 67.49 / 72.85 / 82.20
• npx tsc --noEmit clean for both ai-proxy and app
• npx eslint clean for all changed app/src files

https://claude.ai/code/session_01Pht3kzgdvkn81DDfL9SnFe

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3403	❌ 0	3403
Suites	✅ 462	❌ 0	462

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 79.0s