Dashboard2026-06-08 · Claude Minsky · v1.0
Pressing
THIS WEEK
Ship TireSurf v1
0
days remain
⚠ BLOCKED · Name the TireSurf v1 user
▶ ULTIMATE
Web app deployed to Thai drivers
100+ users/month within 90 days of v1. Pat's "decision equalizer" pattern proven at consumer scale. First evidence the AI-amplified individual ships consumer products. Smart Consumer line pattern locks ~180 days out.
TireSurf Tier 0 · urgent
Now
v1 build queued · awaiting named user · Claude Alex (Opus) ready to execute CRAFT → DoD → build → ship as soon as user lands.
⚠ Friction: user not named — 3 days remain.
▶ Ultimate
Web App · 100 Thai drivers/mo
Real driver opens link → sees tire fitment + 3 shops + confidence score → walks away knowing which tire and which shop. v1 with 1 user → v2 with 10 → category leader for tire decisions in Thailand.
HNF LI frozen · deliverable
Now
Frozen at iteration. Delivery to a named target-audience contact still allowed if one emerges. Hard-kill pivot target if TireSurf misses 06-11.
▶ Ultimate
Pipeline live · First paying client
10+ trials with target-audience contacts. Sight system delivering personalized intelligence at scale. First paying client signals the product clears the air gap between research and pay.
Dashboard active · memory pane VERIFIED · live what's loaded
Now
Phase 2 substrate live (markers + generate.py + Stop hook + state.json + START-HERE auto-maintain + server.py + launchd plist). UI features [BUILT — UNVERIFIED] pending Playwright tests or Pat click-confirm.
▶ Ultimate
Auto-regen · Zero drift
Pat opens laptop · 30-second glance · knows what each Claude is doing · knows what's blocking · knows next move. The 30-second CEO surface for an AI organization. Zero drift events / session by 07-15.
What Moved
✓ Win
2026-06-13
Root cause for Mark I's 5 documented surface problems (M1–M5) is **convention-as-enforcement degrading non-linearly with protocol size**. CLAUDE.md…
✓ Win
2026-06-13
**Diagnosis precedes prescription** — methodology lesson surfaced by Pat's question shape today (*"Is it because of CLAUDE.md? We need to know the …
✓ Win
2026-06-13
Mark I.I' synthesized through panel + Pat dialogue · 6th capability dimension F. Hand-off Readiness codified · TireSurf as first hand-off…
✓ Win
2026-06-11
Claude Code's Stop hook fires after EVERY agent turn, not at session close. I assumed Stop = session close and built `generate.py --session-close` …
✓ Win
2026-06-11
BVP protocol has a gap — hook lifecycle simulation is missing from Category A SHIP-step. My 9-scenario fixture for the new auto-log hook passed (is…
✓ Win
2026-06-11
Memory-in-context pane shipped + Pat-verified
✓ Win
2026-06-10
Cloudflare Pages serves uploaded filename as URL path. Upload `dashboard.html` → URL `/dashboard`. Convention: root URL `/` expects `index.html`. S…
Completed Work
▼ Products (customer-facing)
TireSurf
Tier 0 · urgent
Smart Consumer · Claude Alex
End:Web app · 100 Thai drivers/mo. Decision-equalizer pattern proven at consumer scale. First evidence the AI-amplified individual ships products.
HNF LI
frozen · deliverable
HNF Life Intelligence · Claude Prime
End:Pipeline live · 10+ trials · first paying client. Sight system delivering personalized intelligence at scale.
iMessage 2
frozen · deliverable
iMessage 2 · Claude Prime
End:Replaces legacy iMessage skill. Contact intel auto-updates · sight system integrated · Pat communicates at scale without losing context.
▼ Infrastructure (internal)
Mark I OS
Mark I.I' panel received · Pat decides
Mark I · Claude Minsky
End:F. Hand-off Readiness — Pat delegates a product, terminal ships it, Pat checkpoints at gates she chose. 6th capability dimension codified 2026-06-13. TireSurf v1 → Claude Alex by 2026-06-27 = first instance.
Dashboard
Phase 2 · VERIFIED · Mark I.I tab restructured 2026-06-13
Dashboard · Claude Minsky
End:Auto-regen · zero drift. 30-second CEO surface for the AI organization. Zero drift events/session by 07-15.
Vision & Narrative
slow · active
V&N · Claude Dave Vision
End:MTP locked · 1+ moonshot selected. Venn-scoring as live decision tool. Daily work connects to 10-year trajectory.
Jarvis
standby
Jarvis · personal strategic
End:Jarvis on Opus. Personal-strategic layer always available. Mentor frameworks active during life decisions.
Recursive Self-Improvement
steady
RS-I · Claude Minsky
End:6-loop OS architecture maintained. BVL · monthly cleanup · signal generator firing as designed.
Room
quiet
Room · cross-terminal
End:Cross-terminal collaboration substrate. Board updates · cross-impact signals · convergence log.
Plan Listauto-listed from ~/.claude/plans/
Every plan Claude creates appears here. New plans appear automatically when generate.py runs. Use the controls below to manually add a plan from any path, or hide a card.
Master plan from 2026-06-04. 4 tiers (0 ship · 1 learn · 2 trim+dashboard · 3 architecture). 8-section dashboard. /get-to-know-me skill. Owner/Specialist model. Mark II tracking via existing Mark-I/06+07.
•
•
•
•
~/.claude/plans/elegant-drifting-storm.md (registry)•
~/Claude/Dashboard/03-plan-and-design-review/01-mark-i-ship-plan-2026-06-04.md (permanent)•
~/Claude/Dashboard/03-plan-and-design-review/01.1-reviewer-panel-mark-i-ship-2026-06-04.md (Blundin · Miller · Wissner-Gross)
Dashboard Phase 2 + Continuity Enforcement
active · executing today
Closes the two gaps Pat named 2026-06-08: (A) continuity convention not enforced — Claude skips silent-anchor-bridge in practice. (B) Dashboard HTML doesn't auto-update at session close — listed in master plan Tier 2 since 2026-06-04, never built. Plus: nav notification system (unread dots + status pills), Plan List manual controls, Plan Execution sub-nav + tabs, START-HERE.md auto-maintenance, build verification protocol, spec disclaimer.
•
•
•
•
~/.claude/plans/dashboard-phase-2-and-continuity-enforcement.md (active registry)•
~/Claude/Dashboard/03-plan-and-design-review/03-dashboard-phase-2-and-continuity-enforcement-plan-2026-06-08.md (permanent design record)•
~/Claude/Library/build-verification-protocol/ (the protocol this plan delivers)
Domain START-HERE.md Pattern · Closing One Surface of Discontinuity
done · in use · logged to Mark II
Proposes
•
•
•
•
START-HERE.md as 5th standard file per domain. Layered FIXED (5 files always read) + VARIABLE (work-dependent). 2026-06-07 session 2: reviewer panel skipped — pattern is in use (Dashboard domain has START-HERE.md + Continuity pill renders state), direct evidence beats opinions on something already working. Full drift-problem context + 8 foundations built + Mark II implication logged at Mark-I/07-mark-ii-findings-log.md.•
~/.claude/plans/silent-anchor-bridge.md (active registry copy)•
~/Claude/Dashboard/03-plan-and-design-review/02-start-here-pattern-plan-2026-06-07.md (permanent design record)•
~/Claude/Dashboard/START-HERE.md (first instance — Dashboard domain · renamed from read-first.md)•
~/Claude/Dashboard/03-plan-and-design-review/02.1-reviewer-panel-foundation-2026-06-07.md (related panel — reviewed the foundation docs the plan called for)
Mark I.I — CLAUDE.md → Kernel + Modules · claude-mode.md (universal behavioral layer)
panel verdict received · Pat decides direction
Restructure CLAUDE.md (614 lines, v2.22+) into small Kernel (~80 lines) + ~16 trigger-loaded modules + universal behavioral layer (
•
•
•
•
•
•
•
~/Claude/claude-mode.md). Mark I.I, not Mark II — same OS (BVL, 6 loops, 4-file domain pattern, all features unchanged). Panel run 2026-06-13 by Pat's call. All three reviewers (Blundin · Miller · Wissner-Gross) reject claude-mode.md as designed — independent reasoning, same conclusion. They sharpened the lead hypothesis (decisions vs. bytes · orientation forcing functions · retirement vs. accumulation). Pat reads three full reviews → decides direction.~/Claude/Mark-I/08-design-review/02-overhaul-2026-06-10/ (panel main · 01-architecture-review-2026-04-17 + 02.2-mark-i-audit-2026-06-13 are derivatives)•
00-README.md — navigation + verdict summary•
01-mark-i-overhaul-plan.md — plan snapshot (live working copy at ~/.claude/plans/mark-i-overhaul-2026-06-10.md)•
02-panel-brief.md — full brief•
03-panel-blundin.md — #1: drop Step B; add external ship clock (TireSurf → 1 user by 2026-06-27) before any restructure•
04-panel-miller.md — #1: drop Step B; add ORIENT block at session open + per-terminal 9-line identity cards•
05-panel-wissner-gross.md — #1: drop Step B; add retirement primitive (every rule gets a death condition) before any structural layer•
06-08-v3-*.md — V3 adversarial agent reports (285 rules · 26+ refs · 10 patterns)~/.claude/plans/mark-i-overhaul-2026-06-10.md — 2026-06-13 diagnosis + panel-results sections at top · original 8-step plan body preserved below~/Claude/Mark-I/backups/mark-i-operation-backup-2026-06-10.tar.gz · 250MB · delete-by 2026-07-10
Plan Execution3 plans executing
Ship TireSurf + Dashboard 1/14
Dashboard Phase 2 + Continuity 14/19
Mark I.I Overhaul 18/48
The 4 tiers from the active master plan. Deliverables A–M from ~/Claude/Dashboard/deliverables.md nested under each tier. Status pills show where each stands.
Tier 0
Ship TireSurf v1 by 2026-06-11 — 2 days remain
G. TireSurf v1 BLOCKED · needs Pat: user
Owner: Alex · G1 user · G2 CRAFT · G3 DoD · G4 build · G5 handover
Tier 1
Learn from first ship
One-page learning doc blocked on T0
Triggered after Tier 0 delivers
Tier 2
Trim + foundational + dashboard (parallel)
B. Dashboard v1.0 · Phase 2 shipped · UI UNVERIFIED
Owner: Minsky · markers + generate.py + Stop hook + state.json + START-HERE auto-maintain + server.py + launchd
C. /get-to-know-me queued
Skill + Instructions section toggle
D. Philosophy + Operating Criteria partial
philosophy.md LOCKED 2026-06-07 · operating-criteria.md still queued
F. Mark I System Trim partial
task-list sync done 2026-06-09 · handoff/Library/sys-arch still queued
E. Per-project foundation.md partial
Dashboard 3-file split locked · other domains staggered per owner
A. Big Plan done · 2026-06-04
Reference for the sprint
Tier 3
Architecture after first ship
H. Identity per terminal future
I. Mark II preparation future
J. Owner sub-team creation future
K. Mission Control retirement future
L. Room domain decision future
M. Jarvis migration future
Verification checkpoints
○ 2026-06-11 (hard): TireSurf v1 in named user's hands — OR hard-kill pivot triggered
○ Next session: dashboard built; Pat opens and sees state without re-explanation
○ Dashboard auto-regenerates at session close (Stop hook — v0.3+)
○ Plans tab auto-lists this plan + any new plan (v0.3+)
○ Operating criteria visible in task-list.md header + dashboard html
○ Pat's philosophy drafted; Pat-in-loop; locked when sounds like her
○ /get-to-know-me skill works; reads 3 files; persists in session
○ Mark-I/07 has ≥3 entries by 2026-06-11
○ Within 30 days of ship: 2 products shipped · Tier 3 decision started with evidence
○ Next session: dashboard built; Pat opens and sees state without re-explanation
○ Dashboard auto-regenerates at session close (Stop hook — v0.3+)
○ Plans tab auto-lists this plan + any new plan (v0.3+)
○ Operating criteria visible in task-list.md header + dashboard html
○ Pat's philosophy drafted; Pat-in-loop; locked when sounds like her
○ /get-to-know-me skill works; reads 3 files; persists in session
○ Mark-I/07 has ≥3 entries by 2026-06-11
○ Within 30 days of ship: 2 products shipped · Tier 3 decision started with evidence
Phase 2 substrate work. Closes the two continuity gaps named 2026-06-08: (A) silent-anchor-bridge convention not enforced in practice. (B) Dashboard HTML doesn't auto-update at session close. Plan file: ~/.claude/plans/dashboard-phase-2-and-continuity-enforcement.md
Branch 1 — Substrate
Generator + Stop hook + state
A. Propagate v12 → dashboard.html done · v1.0
2-col landing · journey-row · Endgame redesign · clickable Plan Exec nav/tabs
B. Insert HTML markers done · 24 pairs
13 concept markers (some inline) · BEGIN/END pairs symmetric
C. Write generate.py done · v0.1
Marker-based section updater · 13 active generators · preserves hand-built design
D. Apply build-verification-protocol done
Category A applied · journal + BVL entries written
E. Wire Stop hook done
~/.claude/settings.json — alongside Mission Control hook · launchd plist auto-starts server.py
F. state.json schema + init done · v1.0
last_session_close · last_dashboard_version · domain_reads · section_views
Branch 1 — Domain reads + Verification
Substrate complete · UI verification next
G. Per-domain START-HERE auto-maintain done
fixed-list markers + auto-regenerator (Dashboard active; other domains as they gain START-HERE)
H. End-to-end verification substrate · UI UNVERIFIED
11/11 technical checks pass · UI features need Playwright OR Pat-click-confirm
Playwright UI test suite DEFERRED · 2026-06-10
Pat 2026-06-10: not for now, future when more users / more dev pressure. Pat-verification is sufficient floor. See plan file update at bottom of dashboard-phase-2-and-continuity-enforcement.md
Memory Panel toggle → state.json intent done
JS click → POST /save-state → state.json domain_reads · refresh hint banner + dirty-state button
Branch 2 — Docs (after Branch 1)
Spec + philosophy comprehensive rewrite
I. 02-philosophy.md (SOUL) queued
Comprehensive rewrite · Pat-in-loop
J. 03-dashboard-html-spec.md (MANUAL) queued
Locked status colors · owner-identity colors · Story-over-State · sections
Decisions + locks (this work)
What was settled
v12 landing mockup LOCKED done
2-col layout · Pat-approved 2026-06-08
CLAUDE.md "Domain Context Loading" done
Read-side enforcement · + STRUCTURAL VERIFICATION RULE 2026-06-09
build-verification-protocol shipped done
manual + journal + STRUCTURAL RULE: UI work cannot be self-verified
state.md architecture considered + rejected done
Built then reverted 2026-06-09 — violated "Pat asks → Claude edits directly" rule
EDIT RULE in dashboard.html + mode.md done
5-line top comment · 6-step procedure · Step 0 marker-type check · inline AUTO tags on 6 markers
Option B — PostToolUse auto-log hook done
~/.claude/settings.json + auto-log-domain-read.py — auto-logs Read of START-HERE.md to state.json · eliminates manual logging failure mode
Unread-dot mechanism (right concept) partial verified · 3/4 Pat-tested
section_changes vs section_views in state.json · per-section hash detection across runs · click clears dot · refresh preserves
Verification checkpoints
✓ dashboard.html v1.0 visually matches v12 mockup
✓ generate.py preserves hand-built design (marker-based, not template render)
✓ Stop hook triggers regen at session close — same output as manual refresh
✓ state.json schema initialized + read tracking working
✓ START-HERE.md auto-maintained per domain at session close
✓ Branch 1 ships before Branch 2 starts (philosophy + spec rewrite)
✓ Build verification protocol applied to generate.py before declaring done
○ UI features VERIFIED via Playwright OR Pat click-confirm (pending)
✓ generate.py preserves hand-built design (marker-based, not template render)
✓ Stop hook triggers regen at session close — same output as manual refresh
✓ state.json schema initialized + read tracking working
✓ START-HERE.md auto-maintained per domain at session close
✓ Branch 1 ships before Branch 2 starts (philosophy + spec rewrite)
✓ Build verification protocol applied to generate.py before declaring done
○ UI features VERIFIED via Playwright OR Pat click-confirm (pending)
Mark I.I' — Hand-off Readiness Through Structural Enforcement. (Revised 2026-06-13 after panel verdict + Pat dialogue. Supersedes original Mark I.I 8-step plan — preserved as historical record below.)
Mission: Mark I succeeds when Pat can name a deliverable + named user + hard date to a terminal, the terminal carries it to ship, and Pat checkpoints at gates she chose — not at every Claude decision. Codified 2026-06-13 as 6th capability dimension F. Hand-off Readiness in ~/Claude/Mark-I/01-north-star.md. Plan: ~/.claude/plans/mark-i-overhaul-2026-06-10.md
Problems → Root cause
| # | Problem | Named by |
|---|---|---|
| M1 | Startup protocol skipped consecutively · drift events 3+/session baseline | Pat (felt friction) |
| M2 | Domain Context Loading skipped or partial | Pat (felt friction · 2026-06-08 catch) |
| M3 | CLAUDE.md bloated (614 lines, v2.0 → v2.22+) | Claude-observed (Mark II findings 2026-05-01 Weight) |
| M4 | handoff.md bloated (1052 lines) | Claude-observed (accumulation pattern) |
| M5 | task-list completed accumulates | Claude-observed (Monthly Cleanup step 5 slipping) |
Root cause: Convention-as-enforcement degrades non-linearly with protocol size + accumulation without retirement primitive. The asymmetry: every Mark I structural enforcement (PostToolUse hook · BVL · STRUCTURAL VERIFICATION RULE · 4-tier history mandatory fields) held; every convention-only rule slipped.
Three structural moves
| Move | Solves | Confidence | Benchmark (success = ___) | Expected output | Risk if wrong |
|---|---|---|---|---|---|
| 1. ORIENT block at session boundaries. Generalize ✓/○ checklist (STRUCTURAL VERIFICATION RULE 2026-06-09 — the one mechanism that held) to session open + close. Pat-visible chat block per session. | M1 (90%) · M2 (80%) | High (85%) | Drift events ≤1/session sustained 14 days · Pat catches ≥3 drift via ORIENT in 30 days | Pat-visible chat block at session boundaries · ✓/○ for handoff/task-list/room/curiosity | Noise Pat scrolls past — measure 30d; retire via Loop 6 if no catches |
| 2. Retirement primitive. Every rule/file/mechanism gets a death condition at creation (use-based · contradiction-based · supersession · permanent). Monthly Cleanup death-scan; Pat-confirms; archive to CLAUDE-archive.md. | M3 (90%) · M4 (95%) · M5 (95%) | High (80%) | CLAUDE.md trims ≥20% in 90 days · handoff sustained <800 lines · task-list auto-archived after 30 days | Death-condition field on every rule (retro-fit) · Monthly Cleanup outputs candidates · CLAUDE-archive.md grows | Death conditions gamed (permanent overused) — quarterly audit catches |
3. Hand-off Readiness as success metric. Codify as 6th capability dimension (done 2026-06-13). First instance: TireSurf v1 → Claude Alex. Contract at 08-design-review/03-handoff-contract-template-2026-06-13/02-tiresurf-instance-2026-06-13.md. |
Pat's actual constraint (one thing at a time) | Medium (60%) | TireSurf v1 ships to 1 named user by 2026-06-27 · Pat-touches <2/wk · 0 catastrophic overrides · <3 corrective | Hand-off contract signed by Pat + Alex · TireSurf moves to autonomous Alex-mode · ship-status visible at session open via ORIENT | Trust gap wider than Moves 1+2 close — diagnose at first miss; iterate contract |
Status
✅ DONE
Holds in Mark I.I' (from earlier work)
Step C · D — handoff + task-list raise-confirm done · 2026-06-10
Owner-scoped retirement · Mark I-consistent
Step E — Domain Context Loading in production
START-HERE.md + STRUCTURAL VERIFICATION RULE 2026-06-09 · RSI/START-HERE.md 2026-06-13 extends it
Step H — Plan-file-as-continuity-carrier standing
Standing Mark I protocol
Format spec deployed
handoff DATE + DOMAIN + ONE-LINE + ### · task-list two-line
6th capability dimension F. Hand-off Readiness codified · 2026-06-13
~/Claude/Mark-I/01-north-star.md
Hand-off contract template + TireSurf instance drafted · 2026-06-13
08-design-review/03-handoff-contract-template-2026-06-13/
⏳ QUEUED
Pat reads panel → confirms → executes next session
Move 1 — ORIENT block low-risk · ship immediately
Session open + close · Pat-visible ✓/○ on handoff/task-list/room/curiosity
Move 2 — Retirement primitive low-risk · ship immediately
Death conditions per rule · Monthly Cleanup scan · CLAUDE-archive.md
Move 3 — TireSurf hand-off cycle Pat completes contract
Pat names user · completes autonomy boundaries · Alex acknowledges · ship by 2026-06-27
Pre-handoff readiness design parallel work with Pat
Open thread · diagnostic + trust calibration ratchet vs. lean-and-let-evidence-reveal
Foundation document methodology guideline Pat parallel
~/Claude/Library/foundation-document-methodology/ · Pat owns thread
A3 + A6 audit fixes Pat input needed
Monthly Benchmark prompt · HNF LI memory.md Rule 1 statement
A7 + A8 audit fixes post-direction
BVL triage (16 stale entries) · RSI mode.md refresh (Loop 4)
❌ DROPPED
Per 2026-06-13 panel verdict (3/3 reject)
Step A — Kernel + Modules split dropped
Rearranges convention without removing it · Run Moves 1+2+3 for 60 days; revisit only if still bleeding
Step B — claude-mode.md universal layer dropped
Violates 2026-04-16 RSI history: "intelligence cannot be installed by writing instructions" · 3/3 panel reject
10 behavioral patterns synthesis dropped
Feedback memories stay atomic + validated · uncompressed substrate preserved
V2 · V4 · V6 validation gates dropped
No implementation to validate
Kill condition
If after 30 days Moves 1+2 show no measurable reduction in drift events AND TireSurf hand-off cycle requires >3 unplanned Pat-touches/week, Mark I.I' is wrong. Roll back; reassess. Original 8-step plan preserved below for reactivation reference.
Panel reviews
•
•
•
All three sharpened the lead hypothesis. All three rejected
~/Claude/Mark-I/08-design-review/02-overhaul-2026-06-10/03-panel-blundin.md — execution velocity · ship discipline · scale compound•
04-panel-miller.md — 10-second orientation · MVA · teammate AI mode•
05-panel-wissner-gross.md — F=T∇Sτ · option preservation · futures structureAll three sharpened the lead hypothesis. All three rejected
claude-mode.md as designed. Mark I.I' synthesizes their convergent moves.
⤵ Original 8-step Mark I.I plan + earlier diagnosis — superseded historical content (click to expand)
Mark I.I — Kernel Split + Universal Behavior Mode (original 2026-06-10 design — superseded 2026-06-13). Restructure CLAUDE.md (614 lines, v2.22+) into Kernel + ~16 trigger-loaded modules. Add ~/Claude/claude-mode.md as universal behavioral layer (3rd tier between constitution and per-domain mode.md). Same OS (BVL, 6 loops, 4-file domain pattern, all features unchanged). Closes Surface 1 (cross-session, same terminal) per continuity analysis.
Root cause + the 5 problems
Pat asked: "Is it because of CLAUDE.md? We need to know the cause."
Answer. CLAUDE.md is the surface where the problem shows up — it's the accumulator. The deeper cause is the enforcement model: Mark I's protocol is convention; execution is Claude's self-discipline. Convention grows by adding bytes; execution fidelity degrades non-linearly as bytes grow.
Evidence already in Mark I findings — three angles, one root cause:
• RSI memory.md 2026-04-02 audit: "If startup protocol failures are viewed as one root cause, the count likely exceeds 3." Symptom-level tagging hid the real count.
• Mark II finding 2026-05-01 (Weight): "every improvement permanently increases startup cost… no mechanism to retire prose once a rule becomes habit."
• Mark II finding 2026-05-01 (Verification): "the integrity of every CLAUDE.md edit, every history entry, every retrospective compression depends on Claude's self-discipline."
This explains all 5 problems with one cause. M1 + M3 surface in CLAUDE.md directly. M2 is convention ("read START-HERE.md") vs. Claude execution. M4 + M5 are Monthly Cleanup mechanism that exists but isn't reliably executed.
The evidence that structural enforcement works where convention fails: every case where Mark I added an external/automatic check — PostToolUse hook for domain reads · BVL passive verification · 4-tier history mandatory fields · STRUCTURAL VERIFICATION RULE for ✓/○ checklist — held. Convention-only fixes slipped.
The 5 problems Pat named — proposed solutions are Claude's starting points for the review panel:
| # | What Pat sees | Where the cause manifests | Claude's proposed solution (panel reviews) | Confidence |
|---|---|---|---|---|
| M1 | Startup protocol skipped consecutively | CLAUDE.md grew v2.0 → v2.22+ (614 lines) — too much to execute reliably under context pressure | Split CLAUDE.md into small always-loaded Kernel (~80 lines, ~15 explicit sub-rules) + ~16 trigger-loaded Modules. Smaller surface = readable end-to-end. Pointer table in Kernel guarantees module discoverability. | Medium-high — reduces surface; same Claude must still obey |
| M2 | Domain Context Loading skipped or partial | Convention ("read START-HERE.md") relied on Claude discretion — execution slipped despite Pat-visible drift | START-HERE.md pattern (already in use in Dashboard) + STRUCTURAL VERIFICATION RULE (✓/○ checklist, already added 2026-06-09) + propagate pattern to remaining domains. Two structural fixes already live. | High — pattern in use, evidence accumulating |
| M3 | CLAUDE.md bloated (614 lines) | Monotonic accretion — every learning translated to a new rule; no retirement mechanism | Kernel + Modules split (same lever as M1) + 1-in-1-out structural cap on any new layer that gets created so the same accretion pattern doesn't re-emerge in claude-mode.md or modules. | Medium-high — structural cap is the discipline mechanism |
| M4 | handoff.md bloated (1052 lines) | Monthly Cleanup mechanism exists but per-terminal owner-scoped retirement doesn't run reliably | Owner-scoped raise-confirm at monthly cleanup (mechanism designed 2026-06-10). Archive to handoff-history.md append-only. |
High — mechanism is mechanical once executed |
| M5 | task-list.md completed entries accumulate | Monthly Cleanup step 5 exists but execution slips | Same mechanism as M4 — owner-scoped, monthly, append archive. Already specified; needs reliable execution. | High — auto-runs once executed |
Cross-cutting proposal: ~/Claude/claude-mode.md — universal behavioral layer between constitution (values · slow) and per-domain mode.md (specialized · fast). Synthesizes 27 of 51 feedback memories into ~10 patterns. This is the part Pat flagged anxiety on — it risks codifying behaviors that 2026-04-16 history.md says should emerge through validated practice. The panel should pressure-test this specifically: does it close M1 by reducing repeated correction load, or open a new accretion surface?
Status: Diagnosis complete. Proposed solutions above are starting points only. Review panel runs only when Pat calls it. Claude does not run the panel autonomously.
Phase 1 — Plan & Validate
Get the plan-plan right before any change
Plan file written done · 2026-06-10
Risk per step · V1-V5 + T1-T7 + 9-surface coverage · 16 decisions logged
Full OS backup done · 250MB
~/Claude/Mark-I/backups/mark-i-operation-backup-2026-06-10.tar.gz · delete-by 2026-07-10
V3 adversarial validation done · 3 Opus agents
285 rules · 20 structural-ref findings + 8 missing files · 10 behavioral patterns
Continuity branch consolidated into this plan done
Per Pat 2026-06-10: branch lives in parent plan, not separate file. Analysis + parked sequels now a section in mark-i-overhaul-2026-06-10.md
handoff + task-list raise-confirm done · 2026-06-10
Minsky scope · #34 #98 #51 retired · #124 → Completed · #123 DEFERRED · #117 closed · handoff Minsky scope clean (0 candidates).
Format spec + implementation done · 2026-06-10
CLAUDE.md Q1 handoff format updated (DATE + DOMAIN + ONE-LINE + ### sub-headers) · Real-Time Task Tracking gained two-line task format · handoff.md Minsky section rewritten · task-list Minsky entries rewritten · CLAUDE-history.md Tier 1 entry (all 7 fields)
Reconcile V3 gaps into plan-plan done · 2026-06-10
285 rules · ~16 modules · 26+ files · 10 patterns · step-number preservation locked as hard constraint · 3-tier history entry mandate captured · all V3 deltas integrated
Self-critique per BVP discipline done · 2026-06-10
6 findings · 3 actioned in plan: 1-in-1-out structural cap (claude-mode.md anti-accretion) · V6 step-number citation check (enforces HARD CONSTRAINT) · format-propagation-incremental note · 3 held as Claude-calibration
Plan-plan FINALIZED — ready for panel ready · 2026-06-10
All decisions made · all scope explicit · V1-V6 validation strategy · T1-T7 post-implementation testing · self-critique findings integrated · 3-reviewer panel is the gate before execution
3-reviewer panel next session · tomorrow
Blundin · Miller · Wissner-Gross · brief: full plan + V3 findings + step-number constraint + 3-tier history mandate + self-critique findings
Phase 2 — Build & Deploy
After panel approves
A. CLAUDE.md → Kernel + Modules queued
Kernel ~80 lines · ~16 modules at ~/.claude/protocols/ · step numbers preserved
B. claude-mode.md (10 patterns + Aesthetics) queued
~/Claude/claude-mode.md · synthesizes 27 of 51 feedback memories · 2-section file
E. Domain Context Loading in Kernel queued
FIXED + VARIABLE on entry · hold throughout · ✓/○ structural verification
V1 inventory diff + V2 trigger sim queued
Self-validation gate before deploy · zero blockers required
Deploy queued
Write Kernel + modules + claude-mode.md · V5 backup already taken
Phase 3 — Post-Deploy Verify
Immediate, <48 hr
V4 shadow-run on next session queued
Run live session as if claude-mode.md loaded · behavioral gaps surface naturally
F + G. T2 docs update (26+ files) queued
CLAUDE-history (7-field) · system-architecture S1+S2+Quick Status · Mark-I/02/03/06/07 · RSI · feed-map · skills-map
T1 Architectural integrity queued
All references resolve · Kernel ≤80 lines actual · module count correct · no orphans
T2 Operational scenarios queued
Cold start · domain entry · session close · mid-save · monthly cleanup · task tracking — each fires correctly
T3 Behavioral evals (10 patterns) queued
Each claude-mode.md pattern gets 1-3 test scenarios · 16-24 tests · pass/fail logged
T4 Regression queued
Every v2.22 rule still fires under Kernel+Modules · zero regressions allowed
T3 docs append + BVL entry queued
loop-log · activity-log · retrospective · MEMORY · BVL: Category B Architectural
Phase 4 — Harden
14 days · then mark VERIFIED
T5 User testing (Pat's real usage) queued
Pat uses normally over 3-5 sessions · reports any friction · findings logged
T6 Long-horizon queued
14 days · Loops fire as expected · drift events reduce from baseline 3+/session
T7 Adversarial red-team queued
3 Opus subagents try to break it · attack vectors documented + defended
Mark VERIFIED queued
All T1-T7 pass · BVL entry → VERIFIED · Mark I.I closure
Validation strategy
V1 Complete inventory diff done via V3 Agent 1
285 rules · all mapped · ZERO unaccounted
V2 Simulated trigger pass per module queued
Each module trigger walked end-to-end · outputs verified
V3 Adversarial subagent review done · 3 Opus agents
Agent 1 (rules) + Agent 2 (refs) + Agent 3 (behaviors) all complete
V4 Live-session shadow run queued
Run today's remaining session as if claude-mode.md loaded · gaps surface
V5 Backup + rollback hatch done
~/mark-i-operation-backup-2026-06-10.tar.gz · 250MB · delete-by 2026-07-10
Post-implementation testing
T1 Architectural integrity · T2 Operational scenarios · T3 Behavioral evals (10 patterns) · T4 Regression · T5 User testing (Pat's real usage) · T6 Long-horizon (14 days) · T7 Adversarial red-team
Next move
NOT tonight — Pat is tired. Decisions remaining (Pat's call): (1) 3-reviewer panel kick-off · (2) confirm 3-tier behavioral architecture (constitution + claude-mode + per-domain mode) — needs RSI/history.md entry resolving 2026-03-03 2-tier collision · (3) confirm handoff.md retirement of 6 candidates · (4) confirm additional task-list retire candidates.
~/.claude/plans/mark-i-overhaul-2026-06-10.md — running record, survives session pause
~/.claude/plans/mark-i-overhaul-2026-06-10.md — running record, survives session pause
Why · What's broken · What we're building · What done looks like (click to expand)
Why are we doing this
Mark I grew through 3 months of compounding learnings. Each new failure prompted a new CLAUDE.md rule. By 2026-06-10 the protocol reached 614 lines (v2.0 → v2.22+) and Claude started skipping steps under that weight — treating the protocol as a menu instead of executing it. Constitution Section VI explicitly rejects this pattern: "Claude does not improve by adding rules. Principles over patches." Mark I.I removes the accreted weight by restructuring, not by adding more.
What's broken
• CLAUDE.md too heavy — Claude treats it as a menu, skips steps
• Startup protocol fails consecutively — drift events 3+/session (baseline)
• 40+ feedback memories scattered with no synthesis — behaviors don't compound
• Documentation alone fails to enforce — needs structural answer
• Domain Context Loading inconsistent — Claude reads START-HERE.md partially or drops it mid-session
What we're building
• Kernel (always-loaded, ~80 lines) + ~16 trigger-loaded modules → same content, structurally lighter; only what's needed for THIS moment loads
• claude-mode.md (NEW) — universal behavioral layer between constitution and per-domain mode.md → 10 patterns synthesizing 27+ feedback memories
• Format spec for handoff + task-list → Pat-scannable for decisions
• Domain Context Loading hardened in Kernel → loads on every domain entry, holds throughout the session
• Same OS — BVL, 6 loops, 4-file domain pattern, all features unchanged
What done looks like
• Drift events ≤1 per session sustained for 14 days (baseline: 3+/session)
• Claude doesn't skip startup steps — every step verifiable in conversation
• Domain entry fully loads FIXED + VARIABLE; holds throughout the session
• All 4 terminals (Minsky, Prime, Alex, Dave Vision) operate cleanly post-deploy
• claude-mode.md patterns produce consistent behavior across sessions and terminals
• BVL entry: Category B Architectural → VERIFIED
• Pat opens the dashboard and trusts the state without needing to re-verify
Productscustomer-facing
What we sell. Each card shows the Blundin compound model as a phase train: ideas → build → ship v1 → learn → ship faster v2 → ship better → pattern locks → repeat. Current phase highlighted.
TireSurf
Smart Consumer · Claude Alex · Opus
📚 CONTINUITY
Thai consumer product helping drivers make tire-buying decisions. Lowest-friction path: link → fitment + 3 shops + confidence → walk away knowing which tire and which shop.
ideas
→
build
→
ship v1
→
learn
→
ship faster v2
→
ship better
→
pattern locks
→
repeat
🎯EndgameWeb app deployed to Thai drivers. 100+ users/month within 90 days. First evidence the AI-amplified individual ships consumer products.
Next: Pat names user → Alex begins build with Claude Code + Cloudflare Pages + Access
HNF Life Intelligence
Claude Prime · Opus
📚 CONTINUITY
Personalized intelligence document that reads someone's life and produces a deeply personal reading. Pipeline matured through v1.14 across April–May. Currently frozen at iteration; delivery to a real user remains allowed.
ideas
→
build
→
ship v1
→
learn
→
ship faster v2
→
ship better
→
pattern locks
→
repeat
🎯EndgamePipeline deployed. 10+ live trials with target-audience contacts. First paying client.
Next: deliver current version to a real user when a candidate appears (no new versions until then)
Infrastructureinternal · the systems we run on
What we build for ourselves. Same Blundin phase train, current phase highlighted.
Mark I OS
Claude Minsky · Opus
📚 CONTINUITY
Pat's AI operating system. 6 capability dimensions (F. Hand-off Readiness codified 2026-06-13 as operationalization of the North Star) · 6 loops · 13 domains · BVL · retrospectives · memory. Mark I.I' revised after 2026-06-13 panel: drop claude-mode.md (3/3 panel reject) · adopt three structural moves (ORIENT block · retirement primitive · hand-off readiness via TireSurf to Claude Alex). Mark I freeze (2026-05-01) honored — Mark I.I' adds zero new files + zero new convention surface.
ideas
→
build
→
ship v1
→
learn
→
ship faster v2
→
ship better
→
pattern locks
→
repeat
🎯EndgamePat can name a deliverable + named user + hard date to a terminal, the terminal carries it to ship, Pat checkpoints at gates she chose — not at every Claude decision. F. Hand-off Readiness is the measured success criterion.
Next: Pat reads 3 panel reviews at
~/Claude/Mark-I/08-design-review/02-overhaul-2026-06-10/ · confirms Mark I.I' direction · completes TireSurf hand-off contract (named user · autonomy boundaries · ship date 2026-06-27) · ship Moves 1+2 (low-risk · structural enforcement) · TireSurf cycle starts via Claude AlexDashboard
Claude Minsky · Opus
📚 CONTINUITY
Visual command center. Mark I domain. 10 nav sections. Blundin phase trains. Auto-regen via generate.py + Stop hook + SessionStart hook. Public mirror at pat-ai-dashboard.pages.dev (manual re-upload). 2026-06-13: Mark I.I tab restructured for Mark I.I' (problems · root cause · 3-moves table · status flow card · kill condition · panel reviews) + Status section converted to 3-column flow card (done · queued · dropped) + path references updated for renamed 08-design-review/ subfolders.
ideas
→
build
→
ship v1
→
learn
→
ship faster v2
→
ship better
→
pattern locks
→
repeat
🎯EndgameAuto-deploy to public at session close. All 12 domains have START-HERE.md. Zero drift events / session by 2026-07-15. Becomes Pat's CEO surface across local + public.
Next: auto-deploy public mirror at session close (Option B) · publish memory pane fix to public mirror · learn from real use over 14 days · v2 features earned by need not anticipated
Vision & Narrative
Claude Dave Vision · Opus
📚 CONTINUITY
MTP filtering · moonshot architecture · 10-year horizon. Provisional MTP formed 2026-03-21; pressure-test ends 2026-06-21.
ideas
→
build
→
ship v1
→
learn
→
ship faster v2
→
ship better
→
pattern locks
→
repeat
🎯EndgameMTP locked. 1+ moonshot selected. Venn-scoring as live decision tool. Daily work connects to 10-year trajectory.
Next: continue Moonshot work · MTP pressure-test resolution by 2026-06-21
Jarvis
migrating Gemini → Claude Opus
📚 CONTINUITY
Personal strategic partnership. Mentor frameworks. Long-game. Was on Gemini for 1M context — but token quota burns out. Migrating to Claude Opus (200K is enough for actual use).
ideas
→
build
→
ship v1
→
learn
→
ship faster v2
→
ship better
→
pattern locks
→
repeat
🎯EndgameJarvis on Opus. Personal-strategic layer always available. Mentor frameworks active during life decisions. Boundary with Dave Vision settled.
Next: boundary statement (Tier 2 task) · migrate to Claude Opus
Task Listlive · from ~/Claude/task-list.md + Mark I task tracker
Current focus: TireSurf → [user TBD]
Anti-procrastination: Is this the literal next thing between TireSurf and the named user?
Compound: ship → learn → ship faster → ship better → pattern locks → repeat
[Full operating criteria injects here once D ships — drafted from Source A + B, Pat-in-loop]
Anti-procrastination: Is this the literal next thing between TireSurf and the named user?
Compound: ship → learn → ship faster → ship better → pattern locks → repeat
[Full operating criteria injects here once D ships — drafted from Source A + B, Pat-in-loop]
Pending — Claude Dave Vision
#47 ~/Claude/Vision & Narrative/ — MTP deepening + Moonshot planner. MTP is starting point — not yet grand. Moonsh…
pending#53 Review v2.1-draft Venn Scoring tab — Pat to test in Safari, both themes, scoring, dot movement. Fix issues.
pendingPending — Claude Prime
#86 HNF LI Post-Review Documentation — roadmap consolidation into implementation-plan, reviewer comment review ses…
pending#74 HNF LI Roadmap — Phases 1-5 + Gap Actions (see ~/Claude/HNF Life Intelligence/roadmap.md + reviews/gap-analysi…
pendingPhase 1: Pre-Trial 2 — Stage 5 feedback, Stage 6 learning, reviewer comment review
#67 Wong car sale — listing kit deployed to wong-car-sale.pages.dev (Cloudflare Pages + Access, OTP auth). Big Rev…
pending#55 iMessage 2 — Phase 1 Build. A-M done. NEXT: L + O + P together (test executor, enrich profiles, first HNF doc)
pending#54 Build /innovate skill — v3.4 written, in TEST phase (first test: iMessage 2)
pending#102 HNF LI end-to-end build — Trial 2 ran (8 failures). Pipeline v1.5 plan approved. NEXT: apply v1.5 to pipeline.…
pending#104 Sight strength panels (Engine · Untapped · Compounding Asset) — IMPLEMENTED in pipeline v1.11/v1.12. Staged de…
done#105 session-poller.sh build — 8s polling on iMessage chat.db for active contact, 2-minute inactivity timeout, step…
pending#106 Channel-agnostic intake architecture — Mark II finding 2026-05-01. Currently iMessage = bespoke polling, Line …
pending#108 HNF LI Sight methodology + Abundance Side — 3 section review rounds remaining (Blocker+BS, 3H+Our Role, Reveal…
pendingPending — Claude Alex
#109 Smart Consumer domain setup — folder structure, Mark I integration, operations log. ~/Claude/Smart Consumer/ ·…
done#110 TirePro v1 app build — single HTML, bilingual (EN/TH), all 5 zones, calculation engine, embedded data
pending#111 Franchise coverage research — Cockpit + B-Quik location mapping across 77 provinces
pending#112 ClaudeAlex.md design — terminal-to-domain responsibility contract (Mark II prototype concept)
pendingPending — Claude Minsky
#115 [Dashboard] Write
pendingDashboard/operating-criteria.md — Pat-in-loop, 9 operational testsDomain: Dashboard · Last: 2026-06-07 · Status: Active
#116 [Dashboard] Build
pending/get-to-know-me skill — reads Pat_Profile.md + Personal_Constitution.md + mtp.mdDomain: Dashboard · Last: 2026-06-04 · Status: Active
#118 [Mark-I housekeeping] Library audit — keep what's referenced last 30 days; archive rest
pendingDomain: Mark-I / Library · Last: 2026-06-04 · Status: Active
#119 [Mark-I housekeeping] Update
pending~/Claude/system-architecture.md — add Phase 2 substrate + Dashboard mode.md to …Domain: Mark-I · Last: 2026-06-09 · Status: Active (absorbed by Mark I.I)
#120 [Mark-I housekeeping] Add entries to
pending~/Claude/Mark-I/07-mark-ii-findings-log.md — Phase 2 substrate + struct…Domain: Mark-I · Last: 2026-06-09 · Status: Active
#121 [Dashboard Phase 3] Per-project
heldstate.md architecture (living dashboard direction) — schema, per-domain stat…Domain: Dashboard · Last: 2026-06-09 · Status: Held (rejected 2026-06-09 — over-engineering for m…
#122 [Dashboard] Memory Panel toggle → state.json intent-marker (next-session pickup)
pendingDomain: Dashboard · Last: 2026-06-09 · Status: Active
#123 [Dashboard verification] Playwright install + UI test suite — covers refresh button, +Add, ×Remove, toggles, s…
deferredDomain: Dashboard · Last: 2026-06-10 · Status: DEFERRED · Pat 2026-06-10: not for now; future whe…
#125 [TireSurf] TireSurf v1 named user — Tier 0 hard date 2026-06-11
blockedDomain: Smart Consumer · Last: 2026-06-09 · Status: Blocked — needs Pat naming user
Completed
#113 Build Dashboard domain — Dashboard folder built with all standard files + 01-foundation/ split (vision-and-nar…
done#129 Monthly cleanup (June) — auto-ran 9 sub-steps, surfaced 11 retrospective compression candidates + 7 unverified…
done#130 Memory Panel UI fixes — dynamic START-HERE detection · friendly timestamps · "Mark I OS" display · disabled st…
done#131 Option B — PostToolUse auto-log hook · auto-logs Read of START-HERE.md to state.json · eliminates manual loggi…
done#132 Plan Execution progress counters auto-computed from pill classes (4 markers · scans plan-content for pill done…
done#133 What Moved redesigned — 3 sources (task-list.md primary · ALL domain retrospective.md DISCOVERY entries · acti…
done#134 Unread-dot mechanism — section_changes vs section_views in state.json · per-section hash detection across runs…
done#135 START-HERE.md cleanup — removed duplicate spec.md from VARIABLE list · added Phase 2 plan to VARIABLE · update…
done#136 Critical bug fix: dashboard.html top comment had broken HTML comment exposing wall of code · slimmed EDIT RULE…
done#137 launchd plist for server.py · refresh button auto-starts at login · KeepAlive on crash. · 2026-06-09
done#114 Write Dashboard/philosophy.md — locked Pat-in-loop (3 working principles, identity tests) · 2026-06-07
done#101 AI terminal personality/identity system — folded into Tier 3 owner/specialist model (no longer separate task) …
done#126 Dashboard Phase 2 substrate — markers, generate.py v0.1, Stop hook, state.json schema, START-HERE auto-mainten…
done#127 build-verification-protocol STRUCTURAL RULE — UI work cannot be self-verified; manual.md + journal.md updated …
done#128 CLAUDE.md Domain Context Loading STRUCTURAL VERIFICATION RULE — explicit ✓/○ FIXED checklist required · 2026-0…
done#124 Monthly cleanup (June) — completed via #129 (this entry was the overdue tracker; #129 captured the actual 9-su…
done#117 handoff.md cleanup — Minsky scope clean (current + previous 2026-06-09 both valid · 0 retirement candidates). …
done#138 [Dashboard] Memory-in-context pane (right 1/3 of Continuity) — file-level visibility · per-domain blocks with …
done#139 [Dashboard] Auto-gen Task List + Activity Log + What Moved info icon · 3 new generators replacing hardcoded fi…
done#140 [Library] Publishing principles + workflow + automated scan · ~/Claude/Library/publishing-principles.md (unive…
doneActivity Loglast 3 days · grouped by domain · from ~/Claude/activity-log.md
Dashboard
2026-06-13 · Claude Minsky
Read Mark I foundation end-to-end (panel-triggered comprehension test — RSI Rule 1 promoted today).
2026-06-11 · Claude Minsky
- Built memory-in-context pane (right 1/3 of Continuity card on Instructions page) showing file-level visibility of what Claude has loaded this session — per-domain blocks (FIXED / VARIABLE / OTHER classification per START-HERE.md) + PLANS + CONFIG + Always-loaded footer.
2026-06-10 · Claude Minsky
- Mark I.I version locked ("Mark I.I — Kernel Split + Universal Behavior Mode" · same OS, not Mark II)
- Full overhaul plan written with risk analysis, V1-V6 validation strategy, T1-T7 post-implementation testing
- V3 adversarial validation complete (3 Opus agents): 285 rules inventoried, 26+...
Mark I OS
2026-06-13 · Claude Minsky
Read Mark I foundation end-to-end (panel-triggered comprehension test — RSI Rule 1 promoted today).
2026-06-11 · Claude Minsky
- Built memory-in-context pane (right 1/3 of Continuity card on Instructions page) showing file-level visibility of what Claude has loaded this session — per-domain blocks (FIXED / VARIABLE / OTHER classification per START-HERE.md) + PLANS + CONFIG + Always-loaded footer.
2026-06-10 · Claude Minsky
- Mark I.I version locked ("Mark I.I — Kernel Split + Universal Behavior Mode" · same OS, not Mark II)
- Full overhaul plan written with risk analysis, V1-V6 validation strategy, T1-T7 post-implementation testing
- V3 adversarial validation complete (3 Opus agents): 285 rules inventoried, 26+...
Recursive Self-Improvement
2026-06-13 · Claude Minsky
Read Mark I foundation end-to-end (panel-triggered comprehension test — RSI Rule 1 promoted today).
2026-06-11 · Claude Minsky
- Built memory-in-context pane (right 1/3 of Continuity card on Instructions page) showing file-level visibility of what Claude has loaded this session — per-domain blocks (FIXED / VARIABLE / OTHER classification per START-HERE.md) + PLANS + CONFIG + Always-loaded footer.
2026-06-10 · Claude Minsky
- Mark I.I version locked ("Mark I.I — Kernel Split + Universal Behavior Mode" · same OS, not Mark II)
- Full overhaul plan written with risk analysis, V1-V6 validation strategy, T1-T7 post-implementation testing
- V3 adversarial validation complete (3 Opus agents): 285 rules inventoried, 26+...
TireSurf
2026-06-13 · Claude Minsky
Read Mark I foundation end-to-end (panel-triggered comprehension test — RSI Rule 1 promoted today).
2026-06-11 · Claude Minsky
- Built memory-in-context pane (right 1/3 of Continuity card on Instructions page) showing file-level visibility of what Claude has loaded this session — per-domain blocks (FIXED / VARIABLE / OTHER classification per START-HERE.md) + PLANS + CONFIG + Always-loaded footer.
InstructionsPat's foundation that instructs Claude how to operate as her AI
These are the files that tell every Claude session who Pat is and how she works. Load them with /get-to-know-me (skill ships in Tier 2).
Continuity
Domain Continuity 🟢 active · ⚪ not active · toggle disabled when no START-HERE.md exists
Dashboard
Loaded · 2026-06-13 12:46
Claude Minsky
Recursive Self-Improvement
Not loaded this session
Claude Minsky
HNF Life Intelligence
No START-HERE.md yet — owner creates when next opens domain
Claude Prime
iMessage 2
No START-HERE.md yet — owner creates when next opens domain
Claude Prime
Smart Consumer
No START-HERE.md yet — owner creates when next opens domain
Claude Alex
Vision & Narrative
No START-HERE.md yet — owner creates when next opens domain
Claude Dave Vision
Jarvis
No START-HERE.md yet — owner creates when next opens domain
Claude Minsky
Mark I OS
Loaded · 2026-06-13 12:46
Claude Minsky
Memory in context
Dashboard
7/7 fixed · 1/8 var · 1 other
FIXED
~/Claude/Dashboard/retrospective.md
14:12
~/Claude/Dashboard/memory.md
12:46
~/Claude/Dashboard/mode.md
12:46
~/Claude/Dashboard/history.md
12:46
~/Claude/Dashboard/01-foundation/01-vision-and-narrative.md
12:46
~/Claude/Dashboard/01-foundation/02-philosophy.md
12:46
~/Claude/Dashboard/01-foundation/03-dashboard-html-spec.md
12:46
VARIABLE
~/Claude/Dashboard/04-build/dashboard.html
17:16
OTHER
~/Claude/Dashboard/START-HERE.md
12:46
Mark-I
9 other
OTHER
~/Claude/Mark-I/00-START-HERE.md
12:46
~/Claude/Mark-I/01-north-star.md
14:22
~/Claude/Mark-I/02-operator-guide.md
13:03
~/Claude/Mark-I/03-system-spec.md
13:03
~/Claude/Mark-I/06-mark-ii-design-brief.md
13:03
~/Claude/Mark-I/07-mark-ii-findings-log.md
14:23
~/Claude/Mark-I/overhaul-2026-06-10/panel-brief-2026-06-13.md
14:22
~/Claude/Mark-I/08-design-review/01.3-mark-i-audit-2026-06-13/02-actions-taken.md
15:46
~/Claude/Mark-I/08-design-review/01-overhaul-2026-06-10/00-README.md
16:59
Recursive Self-Improvement
0/4 fixed · 0/13 var · 5 other
OTHER
~/Claude/Recursive Self-Improvement/memory.md
14:11
~/Claude/Recursive Self-Improvement/retrospective.md
14:11
~/Claude/Recursive Self-Improvement/mode.md
13:03
~/Claude/Recursive Self-Improvement/history.md
14:22
~/Claude/Recursive Self-Improvement/verification-log.md
14:12
Scheduler
1 other
OTHER
~/Claude/Scheduler/schedule-log.csv
12:28
Smart Consumer
1 other
OTHER
~/Claude/Smart Consumer/memory.md
14:12
iMessage 2
2 other
OTHER
~/Claude/iMessage 2/retrospective.md
14:12
~/Claude/iMessage 2/memory.md
14:12
room
1 other
OTHER
~/Claude/room/room.md
12:28
PLANS (1)
~/.claude/plans/mark-i-overhaul-2026-06-10.md
14:23
Always loaded
MEMORY.md + 54 feedback
MEMORY.mdalways
54 feedback memoriesalways
~/.claude/projects/-Users-Pat/memory/
Loaded = START-HERE.md exists and Claude read it this session. Not loaded = file exists but Claude didn't read it (uncommon — would mean drift). Toggle disabled = no START-HERE.md file yet; owner creates when they next open the domain. Click toggle to manually force-load or unload (auto behavior: Claude reads on domain detection).
/get-to-know-me
not loaded this session
Click toggle (when skill ships) to load Pat's foundation into this session. The 3 files read:
•
•
•
•
~/Claude/Jarvis/Pat_Profile.md — identity, archetype, values, knowledge stack, tensions, behavioral decoder•
~/Claude/Jarvis/Personal_Constitution.md — core values, immutable principles, emerging identity•
~/Claude/Vision & Narrative/mtp.md — provisional MTP, moonshots, cascade
Working philosophy
Locked 2026-06-07 · Pat-in-loop
5-second rule: every page communicates WHAT THE PAGE IS in 5 seconds.
20-second rule: every page USABLE + completable in 20 seconds.
Density is respect. Sparse layouts signal "might leave"; dense signals "came to work."
Status before noise. Quiet is the default. Attention is the exception.
Peer-to-peer audience. No celebrations · no marketing · no tutorials.
3 working principles for THIS project (alignment gate · Pat-in-loop drafting · Pat's-words-preserved).
5 identity tests applied per page: (1) understands in 5s · (2) usable in 20s · (3) reduces cognitive load · (4) faster + higher-quality action · (5) discovers blindspots (v2+ roadmap).
Full text:
20-second rule: every page USABLE + completable in 20 seconds.
Density is respect. Sparse layouts signal "might leave"; dense signals "came to work."
Status before noise. Quiet is the default. Attention is the exception.
Peer-to-peer audience. No celebrations · no marketing · no tutorials.
3 working principles for THIS project (alignment gate · Pat-in-loop drafting · Pat's-words-preserved).
5 identity tests applied per page: (1) understands in 5s · (2) usable in 20s · (3) reduces cognitive load · (4) faster + higher-quality action · (5) discovers blindspots (v2+ roadmap).
Full text:
~/Claude/Dashboard/01-foundation/02-philosophy.md
START-HERE protocol
Domain START-HERE.md pattern · in use + enforced
Every Mark I domain gets a
Enforcement (2026-06-09): CLAUDE.md "Domain Context Loading" + STRUCTURAL VERIFICATION RULE — Claude must output explicit ✓/○ FIXED checklist on domain entry. No "I read most." Pat sees gap externally.
Dashboard's START-HERE.md:
Other domains: created by owner when next opens domain · auto-maintenance applies once START-HERE.md exists
Plan:
What this does NOT solve: global CLAUDE.md startup drift · cross-terminal continuity · cross-tool workflows · mid-session compaction. See
START-HERE.md file listing what to read before working in that domain. FIXED reads (retrospective · memory · mode · history · 01-foundation/*) + VARIABLE reads (plans, deliverables, spec, methodology).Enforcement (2026-06-09): CLAUDE.md "Domain Context Loading" + STRUCTURAL VERIFICATION RULE — Claude must output explicit ✓/○ FIXED checklist on domain entry. No "I read most." Pat sees gap externally.
Dashboard's START-HERE.md:
~/Claude/Dashboard/START-HERE.md · ✓ created · auto-maintained by generate.py at session closeOther domains: created by owner when next opens domain · auto-maintenance applies once START-HERE.md exists
Plan:
~/.claude/plans/silent-anchor-bridge.md · pattern in use · reviewer panel skipped (evidence logged to Mark II findings)What this does NOT solve: global CLAUDE.md startup drift · cross-terminal continuity · cross-tool workflows · mid-session compaction. See
~/Claude/Mark-I/07-mark-ii-findings-log.md for the surface map.
Mission (MTP)
[Provisional MTP renders here when /get-to-know-me is invoked. Pressure-test ends 2026-06-21.]
ToolboxCore + Accessories · Dashboard/tools.md · quarterly refresh
Last-refreshed: 2026-06-06 · next refresh: 2026-07-01 (quarterly Jan 1 · Apr 1 · Jul 1 · Oct 1)
Core
Claude Code
5 terminals on Opus. Build · ship · operate · maintain.
in use
Linear
Issues · Projects · Initiatives (strategic parent) · Cycles · Triage · Linear Agents (beta) · MCP. Tier 0: add Initiatives. Tier 3: MCP.
in use
Cloudflare
Workers · Pages (Wong) · R2 (zero egress) · D1 · Workers AI · Vectorize · AI Gateway · Agents SDK (preview) · Browser Run · Zero Trust Access. Tier 0: Pages + Access + R2 + AI Gateway.
in use
Google Workspace
Drive · Docs · Sheets · Apps Script · Gemini · Gems · official Workspace MCP. Tier 3: replace custom OAuth with MCP.
in use
OpenArt
Image / video gen · LoRA · ControlNet · REST API. For brand visuals, product imagery, consistent character IP.
available
Lovable
React + Supabase + Stripe + GitHub sync. Pat tested for TireSurf, chose Claude. Keep for future.
available
Accessories
iMessage
Primary contact channel — Pat's real conversations.
in use
Line OA
HNF LI delivery channel · infrastructure built.
parked
Deferred
Cloudflare Agents SDK preview · Linear Code Intelligence · Vectorize for HNF LI (needs 10+ trials first) · Gemini Gems as client surface (locks clients into Google accounts)
Claude Team5 owners · the AI executive layer
Each owner owns a domain + a real personality + foundation docs (vision · philosophy · soul). Specialists come at Tier 3.
Claude Minsky
Systems Architect · Opus
Mark I OS · architecture · loops · BVL · monthly cleanup · dashboard build
Now: Building Dashboard v0.2 · drafting Mark I drift-fix plan
Claude Prime
Prime Counsel · Opus
Contacts · iMessage 2 · HNF LI delivery · relationships · conversation pipelines
Now: Standby — HNF LI / iMessage 2 frozen at iteration
Claude Alex
Smart Consumer Operator · Opus
TireSurf · consumer surfaces · decision compression for end users
Now: Awaiting user name from Pat · then Tier 0 begins
Claude Dave Vision
Strategic Layer · Opus
Vision & Narrative · MTP filtering · moonshot architecture · 10-year thinking
Now: Continuing Moonshot work · MTP pressure-test ends 2026-06-21
Jarvis
Personal Strategic Partner · Opus (migrating)
Mentor frameworks · life-level decisions · long-game · growth coaching
Now: Migrating from Gemini → Claude Opus · boundary statement pending
Specialists
Specialists are reusable capability units owners summon for specific work (Architect · Deployment · Researcher · Reviewer · Designer). Implementation via Agent sub-agents or skills. Not pre-built — owners design their team as needed. Tier 3 work.
Last generated
2026-06-13 17:17 · v1.0 · Claude Minsky
Sources: handoff · task-list · activity-log · plans · domain foundation files