Consolidated from 7 memos, 2 audits, 3 prime-directive updates, 17 legacy todo items, and 40+ user-asks this session.
Priority tiers:
Source column = where the task originated.
---
| # | Task | Source | Effort |
|---|---|---|---|
| H1 | Anthropic billing → Extra Usage OFF (hard-caps Max at $200/mo, protects against cache-bug 10-20× inflation) | 24/7 Research memo | 30 sec |
| H2 | Decide Routines migration alternative — claude.ai/code/routines link not loading. Options: (A) stay on local scheduled-tasks MCP (what we use now), (B) AWS EventBridge + Lambda for API-based cron (needs Claude API key separate from Max), (C) cheap always-on box ($5/mo Hetzner VPS or $120 Raspberry Pi 5) | 24/7 Research memo + user ask 11:25 | 5 min call |
_Note: AWS SNS subscription confirmation — I can self-confirm via Gmail MCP per new CONFIRM-YOURSELF rule. Doing that next._
---
| # | Task | State | Source |
|---|---|---|---|
| N1 | Seductress 100-IDEAS memo (5.5-7.5K words, 10 buckets × 10 ideas) | SCOUT a413afbf15b4ad992 running | user 11:18 |
| N2 | Seductress-applied-to-Innerverse study (3.5-5K words) | SCOUT a618d77059c51b519 running | user 11:20 |
| N3 | Seductress practical build plan + moats (4.5-6K words) | SCOUT a06633dd29d05cc18 running | user 11:20 |
| N4 | Self-confirm AWS SNS subscription via Gmail MCP | 1 call away | new CONFIRM-YOURSELF rule |
| N5 | Silent Infinity bubbles redesign (R0191) — interactive images + feedback + tone indicators | partially started | user 11:22 |
---
| # | Task | Effort | Source |
|---|---|---|---|
| ~~D1~~ | ~~Bubbles redesign~~ SHIPPED R0191 (2026-04-23 11:08 — WCAG solid + tone-dot per chip) | | |
| D2 | Nano Banana image generation — BLOCKED on Gemini billing; swapped to SVG/CSS icons in R0191 | on hold | user 11:28 |
| ~~D3~~ | ~~Self-synthesis master plan~~ SHIPPED (SEDUCTRESS-MASTER-SYNTHESIS-2026-04-23.md) | | |
| D4 | Inbox-watch → claude-p queue pattern (fresh sessions per task, outbox/failed split) | 3-4h | 24/7 Research memo Step 5 |
| ~~D5~~ | ~~PreToolUse file-read injection filter~~ SHIPPED (titan-injection-scan.py, smoke-tested) | | |
| ~~D6~~ | ~~GTM digest email~~ SHIPPED (r-99410591090159947) | | |
| ~~D7~~ | ~~Day-7 return-rate metric~~ SHIPPED (nightly-pmf-d7 cron @ 6:50 AM + titan-innerverse-pmf-d7.py + CloudWatch Innerverse/PMF/Day7ReturnRate) | | |
| ~~D8~~ | ~~/me "Your week" verify~~ SHIPPED (smoke-tested 200 OK, Kimi finding stale) | | |
| ~~D9~~ | ~~Audio 15-22s cold load~~ SHIPPED R0195 (pre-warm noise samples in requestIdleCallback + shrink 5s→2s, first-tap-to-audio expected <500ms vs 15-22s) | | |
---
| # | Task | Effort | Source |
|---|---|---|---|
| W1 | T028 Session persistence — visible "saved" indicator + verify cross-browser hydration via DDB | 4h | Kimi audit R4 |
| W2 | R0175 wire classify_crisis to emit crisis_regex_miss CloudWatch metric (currently logs only, not metric) | 1h | CC Pattern 5 design memo |
| W3 | R0176 Skills System split — 38KB monolith → core (15KB) + 7 lazy skills (reflect/plan/somatic/parts/attachment/grief/work) | 2d | CC Pattern 4 design memo |
| W4 | T020-audit turn_weight_classifier in feedback_monitor (Advisor Tool pre-work) | 4h | CC audit 22:00 rec D |
| W5 | Voice TTS sentence-streaming (biggest UX win — TTFA 7.6s → ~2s) | 8-12h | Voice-CX memo top rec #1 |
| W6 | Voice SSML prosody — slow Ruth Neural to 93% rate, 600ms breaks after empathic acks, soft volume on emotional | 4-6h | Voice-CX memo #2 |
| W7 | Voice 1.5s silence after deep-disclosure turns (grief/fear/trauma classifier-triggered) | 4-6h | Voice-CX memo #3 |
| W8 | Substack essay #1 — Harnoor's byline, "contemplative AI" category claim | Harnoor writes | GTM memo Marketing #1 |
| W9 | Open Attio CRM + build pipeline stages + import HM Tech warm contacts | 2h | GTM memo Sales #1 |
| W10 | 1-page B2B pilot brief (PEPM, outcome metrics) → send to 10 warm HR contacts | 1d | GTM memo Sales #2 |
| W11 | Therapist program at $29/mo — outreach to 100 Psychology Today therapists (ACT/mindfulness filter) | 1w | GTM memo Sales #3 |
| W12 | Canon B2B voice-agent Lambda scaffold (copy innerverse-mirror handler, strip Silent Infinity prompts) | 3d | Seductress-B2B brief |
| W13 | Landing page canon.ai (domain check) or placeholder | 1d | Seductress-B2B brief |
---
| # | Task | Effort | Source |
|---|---|---|---|
| S1 | Silent Infinity prompt-eval dataset — 50 canonical conversations × 10-dim rubric, nightly Haiku, alert on >10% drop | 1w | GTM memo QA #1 |
| S2 | Adversarial crisis-path test matrix in pytest — zero tolerance false negatives | 3d | GTM memo QA #2 |
| S3 | CloudWatch Synthetic E2E voice-turn test every 5 min in prod | 2d | GTM memo QA #3 |
| S4 | axe-core in Playwright CI — WCAG AA fail-build | 2d | GTM memo QA #4 |
| S5 | 8 JTBD user interviews via Respondent.io | 2w | GTM memo User-Testing #1 |
| S6 | Dscout 4-week diary study (n=30) + PHQ-9 pre/post | 4w | GTM memo User-Testing #4 |
| S7 | Vanta SOC 2 Type I kickoff | 3-6mo | GTM memo Sales #4 |
| S8 | 6 SEO landing pages — "AI for overwhelm", "AI for grief", "AI that listens without advice", "contemplative AI", "AI alternative to therapy", "AI for burnout" | 1w | GTM memo Marketing #2 |
| S9 | Session recap Twitter/X cards + gift-a-session at checkout (R0173 supplies the data) | 1w | GTM memo Marketing #5 |
| S10 | Coordinated Product Hunt + Show HN + Reddit dual-launch | 2w | GTM memo Marketing #4 |
---
| # | Task | Source |
|---|---|---|
| L1 | T016 Graduated conversation compaction in conversation_store.py (already partially in-file, needs wiring) | prior todo |
| L2 | Email-buddy channel adapter (reuse memory layer, 1 day) | prior todo |
| L3 | Voice M2 WebSocket + Nova Sonic scaffolding (2-3 days) | prior todo |
| L4 | 1-hour Bedrock prompt-cache TTL investigation | CC audit T019 follow-up |
| L5 | P5 Nano Banana — BLOCKED on Gemini billing | prior blocker |
| L6 | Kimi R5 clinical advisor hire (decision: pre-launch gate vs post-launch?) | Kimi audit R5 |
| L7 | Kimi R6 PHQ-9/GAD-7 optional self-assessment | Kimi audit R6 |
| L8 | Kimi R7 real-world connection prompts | Kimi audit R7 |
| L9 | Kimi R8 usage-pattern awareness (anti-dependency) | Kimi audit R8 |
| L10 | Kimi R9 multi-language | Kimi audit R9 |
| L11 | Kimi R10 offline breathing/grounding | Kimi audit R10 |
| L12 | Kimi R11 SEO discoverability (partially covered in W-tier, remaining: crisis-query landing pages) | Kimi audit R11 |
| L13 | Kimi R12 open crisis-protocol documentation | Kimi audit R12 |
| L14 | Higher-ed pilot at $3/enrolled student/year | GTM memo Sales #5 |
| L15 | Monitoring: k6 spike test (0→100 concurrent) before launches | GTM memo QA #5 |
| L16 | Watchdog task-output GC cron — nightly cleanup of 0-byte .output files >7d | from swarm consolidation analysis |
---
R0172 Correction-as-Memory v2 · R0173 Session Recap · T014 memory block → late-message ($600/mo savings) · T020 Makefile deploy-smoke · R0175 classify_crisis fn · R0175b handler wire · R0178/R0179 voice gap tuning · R0180 Waiting Words animation · R0181 voice bubble labels below · R0182 Polly SSML rate 107% · R0183 friendly voice error map · R0185 reduce-motion accessibility · R0186 bubble auto-send · R0187 send-box declutter + mic removal · R0188 warmup + 1GB Lambda + mem-fetch 250ms · R0189 PMF metrics (SessionDepth + 1-tap rating) · R0190 delete-all + retention · CC audit cycle 22:00 UTC · 24/7 Agentic Watchdog cron · PreToolUse Bash guardrail (32 patterns) · PostCompact hook wired · AWS Budget $50/day · SNS cost-alarms · CloudWatch deadman alarm · .gitignore secrets hardened · Nightly Report Writer cron · Seductress Word doc on Desktop · Inbox poller+queue · Agent Stack dark email template
---
1. MODULARITY (2026-04-21)
2. PARALLEL-WORK (2026-04-21)
3. NEVER-STOP-CONTINUE-EMAIL-ONLY (2026-04-22)
4. NEVER-STOP-STRICTER-NO-WAITING (2026-04-23 morning)
5. CONFIRM-YOURSELF-NO-USER-WAITING (2026-04-23 midday)
6. LEARN-FROM-CLAUDE every 6h (2026-04-22)
7. TITAN-DRAFTS-AGENT-STACK-DARK-TEMPLATE (2026-04-22)
---