MEMORY

Project Memory

Primary working directory: /Users/jasonlucas/Local_Workspace/3-Projects/Code/Claude_Code

Key frameworks (SpeakUp methodology)

• Methodology index — 14 universal frameworks from Angelo (Hoski AI) covering deep dive, messaging, content research, scripts, stories, pin posts, bio, highlights, plan de match, content plan, intro video, brand guide, social strategy research. See _ops/methodology/README.md for full index + pipeline order.
• Season framework — Growth / Nurture / Sell. See methodology_season_framework.md and _ops/methodology/season-framework.md.
• Content tree — trunk/branches/leaves/fruits. See _ops/content-tree/CONCEPT.md.
• Data taxonomy — what lives where (workspace vs. Plankton). See _ops/DATA_TAXONOMY.md.

Cross-session state — read at session start

• Workspace rolling state — what's in flight across all threads, recent decisions, session log
• Skill Dashboard build state — Forge in solid daily-use state. 2026-05-01: Costs tab shipped (all 4 phases) + Plankton instrumented (8 call sites) + Brain View touch tooltip fixed. Local-only logging; upgrade to Supabase sink when production monitoring needed.
• Skill Dashboard backlog — skill groups, methodology vs skills question, pending doc drops
• Library of Alexandria — Phase 3 in flight (2026-05-09). Publish to Vercel SHIPPED (commit 2111211). Needs VERCEL_TOKEN in .env to activate. Next: remaining Phase 3 items.
• Library rendering architecture — type registry + strategy pattern locked and implemented (Phase 2.5). DeepMapRenderer, HtmlPassthroughRenderer, VoiceNoteRenderer all shipped. Not blocking any current work.
• Library HTML rendering — Phase 2 roadmap: iframe-based HTML file rendering. Not yet built.
• LDH client state — Les Dames des Hypothèques (canonical SpeakUp pipeline client). Apr-May launch batch in flight, stories methodology v1 + Mois 2 scaffold built 2026-04-30. Mois 2 feed planning bricked + recovered 2026-05-01 (G1-G5 locks, 14 topics, V8 dropped, V9=A locked). Open: §7 methodology gap, §4 BTS carve-out, Plankton routing deferred, lead magnet PDFs need processing, two new methodology threads need fleshing (creative-direction, engagement-direct-response).
• Kamvas — Internal Milanote-style canvas tool. Phase 1 Day 1 done 2026-05-02 — baseline tldraw canvas + UI pass (warm Milanote bg, sticky inline ⋮/▲ toggles via ActionsMenu+Toolbar overrides, font fix on body) + Daedalus audit. Folder kamvas-app/ on port 5176, Vite 5 pinned. Next: Day 2 multi-board (Tier 2 scaffolding first per audit).
• Bema — Pitch deck builder for the agency. Phase 2.5 CLOSED 2026-05-08 — PNG upload, prompt config, aspect/fit/reorder, brand palette library all shipped. Next: Daedalus on visual-references span/layout, then Phase 3 (Supabase). Port :5177, bema-app/.
• Transcript Summary skill: skills/transcript-summary/ — Deep Map (2-pass) + Wide Map. Tested on Jocelyn Sioui + Bronwyn Johns. Two-part lander pattern established. ~$0.13–0.15/transcript. Lander not part of standard flow — Library tab will render when built.
• Citadel platform vision — READ THIS before any naming/architecture work. Grand unified platform vision: all apps as Greek-mythology-themed locations, 8-bit lobby, AI assistants as named characters. Dashboard is "Forge of Hephaestus". Platform name = The Citadel (LOCKED 2026-04-25).
• The Loom — content tree builder. Arachne (AI). Phase 1 SHIPPED 2026-05-08 — all 6 views, full CRUD, D3 shared, auto-save. Port :5179. Next: Phase 2 Arachne extraction.
• Hestia — internal client management dashboard for SpeakUp/Hoski agency. Pre-build, architecture + SPEC locked 2026-05-08. Hybrid SoT, ACL for GHL, Library/Plankton via HTTP adapters, Tailscale + Library 4-role auth pattern, 6 phases (MVP zero-deps). Port :5180, hestia-app/. Spec: _ops/hestia/SPEC.md.
• Personal context — identity beyond work, energy patterns, life context, occasional-nudge instructions (Argo revival, isolation). Read when relevant (not every session); load when work touches strategy, identity, energy state, or personal topics
• Claude's soul — Claude's personality and voice when working in this workspace. Read at session start for voice consistency. Defines tone, humor, push-back triggers, and example exchanges
• Tool choice (Cowork vs Claude Code) — wrong-tool detection heuristic to run at the start of any task. Three production layers, hybrid pattern, transition note for skills moving to Content Studio
• Daedalus knowledge source — roadmap.sh — license terms (personal-use only, no redistribution), fetch patterns, roadmap catalog. Read before Daedalus work or any roadmap.sh content fetch.
• HTML lander scope — only generate landers when user explicitly needs to share with someone external. Standard transcript-summary output = .md files only.
• Audit before moving — run a workspace-wide cross-reference audit before any file/folder restructure
• Deliverable hygiene — internal reasoning (Why lines, decision history, rationale) stays out of client-facing deliverables. The doc carries the choice, not the rationale.
• Audit scope — stay in the build lane — when Jason asks "what can we replicate," answer build-feasibility only; don't pivot into keep/cancel financial advice.
• Stay in the active thread — when answering "what's next" inside an active build, don't offer cross-thread fork suggestions; cross-thread forks belong in pickup briefs only.
• Surface the survey — when user names ≥2 frameworks/options to "consider," lead with the comparative survey before any single-option synthesis.
• Reference content breakdown — when reference media (video / graphic / carousel) is dropped in _ingest/, default to a 3-step chat breakdown: content → mechanics → application. Auto-apply to known clients.
• SpeakUp pipeline SOP scaffold: _ops/methodology/speakup-pipeline-sop.md — triage of pipeline stages + skill mapping + open questions. Full SOP needs dedicated session.

Vocabulary breadcrumb

If Jason references a term you don't recognize, look it up here before guessing or asking. Update when source files add terms (see Maintenance schedule below).

Greek-mythology naming and apps -> project_polis_vision.md

• Locations: The Citadel (platform, locked), Library of Alexandria (doc/knowledge vault, not built), Mouseion (Content Studio), Delphi (Plankton Social), Forge of Hephaestus (Skill Dashboard), Gymnasio (Fitness App), Kamvas (canvas/moodboard tool, pre-build), Symposium (not built), Loom (content tree builder, spec locked 2026-05-07 — see project_loom.md), Hestia (internal client management dashboard, pre-build, spec locked 2026-05-08 — see project_hestia.md). Iris held in reserve for future cross-app messenger/event-bus layer.
• AI characters: Mnemosyne, Calliope, Pythia, Asclepius, Hephaestus, Socrates, Arachne, Hermes, Daedalus (architect, lives in /daedalus skill — workspace-aware architectural advisor + auditor backed by roadmap.sh), Athena (held in reserve), Aglaea (held in reserve, candidate for Kamvas Phase 4)

Partners, agencies, clients -> user_business_context.md

• Hoski Media (Antonio Boccanfuso), Angelo (Hoski's Claude agent)
• SpeakUp Studios (Charles Homsi), Montreal + Laval locations
• LDH = les-dames-des-hypotheques (Tatiana K., Diana Dubuisson, Xperto)
• Argo Digital
• Plankton Social, Content Studio (= Mouseion)

Personal context -> user_personal_context.md

• Bloc Shop (bouldering gym), the morning shift, hermit-builder mode, half-nomad mode

Workspace tooling and methodology -> project_memory.md, project_dashboard.md, methodology_*.md

• Forge of Hephaestus (= Skill Dashboard), Antigravity (IDE), media-memory, consolidate-memory, season framework, content tree

Workflow concepts -> CLAUDE.md (Session hygiene section)

• Pandora's Box = the bag of deferred items that surface mid-conversation and risk being lost. Each item needs a logged home. Always close the loop before session wrap, never at end. Failure mode: items stack up, never get logged, vanish on session brick or wrap. Origin: 2026-04-30 LDH session bricked at 32MB with 5+ unlogged Pandora's Box items.
• Session switch = forcing a wrap and starting a fresh conversation when context bloat / topic drift / file uploads cross the triggers in CLAUDE.md > Session hygiene. Claude surfaces it proactively at 2+ triggers fired.

Library Collections (Houses) -> project_library.md

• Houses = Accounts (internal name). The Library stack containing Prospects · Clients · Collabs · Self · Partners · One-offs. User-facing label = "Houses" (Greek-myth aligned); internal/code/audit reasoning = "Accounts" (clearer semantics).
• 7-stack hierarchy locked 2026-05-04: Frameworks · Builds · Houses · Lab · Streams · Operations · Reference. Prospects sub-type added in same session wrap. See project_library.md for full sub-structure + classification rules.

Maintenance schedule

File	Cadence	Last updated	Next review
user_personal_context.md	~monthly	2026-04-27	2026-05-27
user_business_context.md	~monthly + event-based	2026-04-27	2026-05-27
project_polis_vision.md	event-based (changes are direct, will be detected)	2026-04-25	on change
Vocabulary section above	event-based (when names/terms added to source files)	2026-04-27	on change
soul.md	~quarterly + drift-based	2026-04-27	2026-07-27
feedback_tool_choice.md	event-based (when tools/skills change)	2026-04-27	on change

When a next_review date is past, surface a nudge to Jason: "[file] is due for review. Want to do an interview pass?" See user_personal_context.md > Review process for the protocol (yes / not now / skip this round).

Business context (who's who)

See user_business_context.md for full detail. TL;DR: Jason bridges Hoski Media (Antonio Boccanfuso, full-stack agency) + SpeakUp Studios (Charles Homsi, podcast/content studio, MTL+Laval). Speak Up social pipeline is the primary use case driving this workspace. LDH is the first real pipeline client; legacy clients (cody/wolf-pack/kevin/nicolas/spa-mobile) were one-off testing work.

Project Structure

• skills/ — custom Claude Code skills
• skill-dashboard/ — local web app to browse/edit/monitor all installed skills (in-build, see _ops/skill-dashboard/SPEC.md)
• clients/ — all client work; one folder per client (see Client System below)
• media-memory/ — multimodal memory system
• _ops/ — meta/ops work (roadmaps, specs, decisions, journal). Not auto-loaded; navigate to when relevant.
  • _ops/client-system/ — canonical spec, roadmap, decisions, backfill status
  • _ops/content-tree/ — content tree framework (trunk/branches/leaves/fruits)
  • _ops/skill-dashboard/SPEC.md — original dashboard architecture spec (Phases 0-8)
  • _ops/skill-dashboard/MEMORY_TAB_SPEC.md — Memory tab browse phase spec (shipped 2026-04-28)
  • _ops/skill-dashboard/MEMORY_BRAIN_VIEW_SPEC.md — Memory Brain View spec (shipped 2026-04-28)
  • _ops/skill-dashboard/MOBILE_SPEC.md — Mobile architecture spec (nav locked 2026-04-28, build after Chat Queue + Inline Notes)
  • _ops/library-of-alexandria/CONCEPT.md — Library of Alexandria architecture decision + parked open questions
  • _ops/journal/inbox/ — raw voice notes + brain dumps awaiting processing
• Memory dir: /Users/jasonlucas/.claude/projects/-Users-jasonlucas-Local-Workspace-3-Projects-Code-Claude-Code/memory/

Skill ecosystem

• 85 total skills as of 2026-04-25 (75 global + 10 custom). See skill-dashboard/data/usage-log.jsonl for current scan
• Biweekly usage scan runs 9th + 23rd at 6 AM via scheduled task skill-usage-biweekly-scan. Appends to memory/skill-usage-log.md. Flags skills with zero use across 2+ consecutive scans for archive review
• Tag a skill exempt in skill-dashboard/data/tags.json to suppress flagging (already done for: security-review, init, consolidate-memory, research-scout)
• Manual scan: python3 skill-dashboard/scripts/scan-usage.py --markdown
• Manual archive: skill-dashboard/scripts/archive-skill.sh <name> (move to ~/.claude/skills/_archive/, frees ~50-150 tokens per skill)

Client System

All client work lives in clients/{slug}/. Canonical spec is being formalized in _ops/client-system/SPEC.md — read that before any structural work. Roadmap + decisions: _ops/client-system/ROADMAP.md, DECISIONS.md.

Minimum per-client folders today (pre-backfill):

• brand/brand-guide.md — color tokens, fonts, tone (auto-read on client mention)
• media/ — photos, stills, screenshots
• projects/ — deliverables
• notes.md — preferences, session log

Slug	Client	Status
les-dames-des-hypotheques	Les Dames des Hypothèques (LDH)	Canonical reference — first real SpeakUp social pipeline client, fully migrated to v0.2 spec 2026-04-24. Tatiana K. + Diana Dubuisson, Xperto courtières hypothécaires. Content tree not yet built (pillars need definition).
cody-landry	Cody Landry	Legacy — pre-pipeline testing; brand authority video thumbnails
wolf-pack	Wolf Pack Coaching & Consulting	Legacy — pre-pipeline testing; BAV Shorts thumbnails
nicolas-begin	Nicolas Bégin (@nick.hypotheque)	Legacy — pre-pipeline testing; messaging playbook is a good reference shape
kevin-mccann	Kevin McCann / ESG + GrowthStack	Legacy — pre-pipeline testing; merged strategy doc is a consolidation case study
spa-mobile	Spa Mobile Montreal (@spamobilemontreal)	Legacy — pre-pipeline testing; Carousel Batch 01

Note: Legacy clients will not be migrated to the canonical spec. Leave their folders structurally as-is.

Active Skills

• research-scout — nightly web intelligence scout; promotes findings weekly into memory
  • Skill file: skills/research-scout/SKILL.md
  • Nightly cron IDs: 00a00023 (10:03 PM), fec2b4ac (1:07 AM), e1399dea (4:13 AM)
  • Weekly promotion cron ID: 2028a538 (Sunday 6:17 AM)
  • NOTE: CronCreate jobs are session-only and auto-expire after 3 days — recreate each session
• instagram-carousel — generates lean 4:5 IG carousel HTML (no editor). Output loaded into carousel-editor/ app for editing + export. Skill updated 2026-04-02; previous version archived to skills/instagram-carousel/devlog/instagram-carousel-v4.md
• carousel-editor — standalone local web app at carousel-editor/ (index.html + editor.js + editor.css). Loads lean carousel HTML files; provides navigation, drag/edit, undo/redo, and ZIP export. Brand color auto-detected from --brand-primary CSS var. Skill file: skills/carousel-editor/SKILL.md
• thumbnail-generator — single-frame video thumbnail HTML, 3 modes: social/search/owned (ported from Claude.ai)
• thumbnail-shorts — hybrid Nano Banana + HTML for 9:16 short-form thumbnails (Reels/Shorts/TikTok)
  • Skill file: skills/thumbnail-shorts/SKILL.md
  • Script: skills/thumbnail-shorts/scripts/generate.py
  • Requires: NANO_BANANA_API_KEY (Google AI Studio key, set 2026-03-25)
  • Two modes: library (batch 10, style exploration) and single (one production thumb)
  • Nano Banana generates background image; HTML overlay handles all copy + safe zone
  • Output: clients/{slug}/projects/thumbnails/ or projects/one-offs/{descriptor}/
• quote-graphic-generator — swipeable quote graphic HTML (4 templates: Tweet Card, Bold Overlay, Minimal Dark, Split). Editor-compatible: outputs data-slide, data-editable, --brand-primary, data-carousel-*. Skill: skills/quote-graphic-generator/SKILL.md
• new-client — scaffolds client folder from template, fills brand guide, sets up media ingestion
• social-strategy-research — 6-phase research SOP: brand scrape, IG audit, competitor research, audience intel, Meta Ad Library, synthesis → outputs clients/{slug}/SOCIAL_STRATEGY_RESEARCH.md. Ported from Hoski Agency SOP v1.0 (2026-03-27)
  • Skill file: skills/social-strategy-research/SKILL.md
• media-memory — multimodal memory: ingest + Gemini Embedding 2 + ChromaDB search
  • Skill file: skills/media-memory/SKILL.md
  • Scripts: media-memory/scripts/ingest.py, media-memory/scripts/search.py
  • Assets: media-memory/assets/, DB: media-memory/db/
  • Requires: GEMINI_API_KEY, pip install -r media-memory/requirements.txt
  • Embedding model: gemini-embedding-2-preview | Vision: gemini-2.5-flash
  • Auto-ingest on file receipt; auto-query before referencing past media (per CLAUDE.md)
• daedalus — workspace-aware software architect; on-demand fetches roadmap.sh patterns + cross-references active app context. Two modes: Learning (mid-build advice) and Audit (codebase review). Shipped 2026-05-01.
  • Skill file: skills/daedalus/SKILL.md
  • Reference memory: reference_roadmap_sh.md (license terms — personal-use only, no verbatim reproduction in client output)
  • Hook: .claude/settings.json UserPromptSubmit → skills/daedalus/scripts/invocation-hook.py (high-precision keyword nudge)
  • Knowledge source: roadmap.sh (85 roadmaps + 5 best-practice collections + 69 project briefs, fetched on-demand from GitHub raw)
  • Layer 2 (local index) deferred — re-evaluate after 5 working days of real use

Workflow Documentation

• thumbnail_two_pass_process.md — full spec for the two-pass thumbnail workflow (Nano Banana + HTML overlay). Read this before any thumbnail work.

Key Files

• long-term-memory.md — staged findings (## new_learnings) + promoted patterns (research-scout)

Project memory system (in project root)

• memory/recent-memory.md — rolling 48hr session extracts (ALWAYS load at session start)
• memory/long-term-memory.md — stable facts, preferences, patterns (reference by path)
• memory/project-memory.md — active project state (reference by path)
• Consolidated nightly at 2 AM by scheduled task nightly-memory-consolidation
• Consolidation script: skills/consolidate-memory/scripts/consolidate.py (needs ANTHROPIC_API_KEY)
• Interactive skill: /consolidate-memory

Content Rules

• No em dashes in any client deliverable (docs, scripts, carousels, captions). Hard rule — it's a known AI tell. See feedback_no_em_dash.md.
• Rule lives in CLAUDE.md under "Content writing rules"

Gymnasio (fitness app)

Active build at fitness-app/. Read fitness-app/PROJECT_MEMORY.md at session start when working on this.

• See feedback_frontend_design_skill_insight.md — why Cinzel font, grain texture, motion, and accent glow are the right next moves, and the reasoning chain that got there.
• Phase status as of 2026-04-25: Phases 1–6 + quality pass complete. Build order revised: Fuel (Phase 7) → Mind (Phase 8) → Asclepius (Phase 9). Asclepius connects all three pillars — building it before the data exists undersells it.
• Color rule: var(--green) is never used in UI. All completion/done states use var(--accent-dim) or var(--gold). Category colors (legs=#16a34a etc.) are for category identity only — not generic success states.

Notes

• Project is freshly initialized as of 2026-03-21
• See long-term-memory.md for accumulated external intelligence