Reference¶
This section is the working memory annex for Copilot CLI sessions. Anything that is reference (look up on demand) rather than rule (must be followed every session) lives here, so the ~/.copilot/copilot-instructions.md file can stay focused on rules.
Two purposes share this folder (set 31 May 2026)
The portal serves two distinct uses:
- Atlas Brain Extended — Atlas's working-memory storehouse. Docs Atlas reaches for to make decisions, run a playbook, surface a rule's provenance, propose architecture, draft in Sush's voice. Atlas operates these; they're written for Atlas's retrieval patterns. Front door:
atlas-brain-extended.md. - Sush-authored sacred content — Sush's philosophy, voice library, founding principles (in
~/.copilot/plain-ai-*.md). Atlas READS but doesn't OWN. Edits proposed before commit.
Both coexist in this folder. The distinction is who edits, not where it lives. The 31 May 2026 trim cycle named what was already true — see trim-2026-05-cycle.md for the rationale + Rule #9 (Portal-First Retrieval) origin.
How this section is used
Future Copilot sessions read these pages as plain markdown via the filesystem, not over the web. The website is a rendered viewer for Sush. The canonical source of truth is C:\ssClawy\learning-docs\docs\reference\*.md.
copilot-instructions.md contains pointers to file paths in this folder, e.g.:
For the full deployment playbook, read
C:\ssClawy\learning-docs\docs\reference\deployment-playbook.md
The router at ~/.copilot/scripts/portal-router.ps1 reads cwd + recent file changes + prompt and recommends the right docs. Atlas runs it before any non-trivial response (per 🚨 Rule #9 in copilot-instructions.md).
What lives here¶
| Page | What's in it |
|---|---|
| Atlas Brain Extended (front door) | Atlas's working-memory taxonomy — grouped by retrieval situation. When Atlas needs memory grounding / proposes architecture / deploys / drafts content / integrates a tool / handles an incident — this is the per-situation table. Read this if you want to scan how Atlas thinks. Set 31 May 2026 during the trim cycle. |
| Trim 2026-05 cycle | May 2026 brain trim methodology + outcomes. Protected-content floor calc · brain regression test suite · per-critic synthesis · Y/X net-KB equation · Rule #9 (Portal-First Retrieval) origin · in-context manifest design. Read this for the canonical trim methodology future cycles should follow. |
| Triggers — keyword → learn-doc routing | Read this when you don't know which playbook to load. Maps keywords (Cosmos, Plain AI, deploy, etc.) and implicit triggers (about to touch a practice exam file → read SLA) to the playbook(s) you should read. Lifted out of copilot-instructions.md 22 May 2026 so the always-loaded soul file stays small. |
| Memory System Evolution | Honest evaluation of external memory-system patterns (notably obsidian-memory-for-ai v3.1) and the case for what we adopted (Promote-on-2+ rule, triggers.md), what we rejected (bi-temporal frontmatter, _inbox/, atomic-facts reorg), and why. Read first if Sush surfaces another memory-system idea — it explains what we already evaluated and the reasoning. |
| Architecture Gate + Data-First Playbook | The deep version of soul-file Rules #5 (Architecture Gate) and #6 (Data-First Sequence). The atlas-portfolio 4-day case study (3 failed attempts vs the 80-minute success). The probe-script pattern. The 12-step checklist for the next big tool/dashboard build. Read first when starting a new tool or dashboard, OR when Atlas catches itself skipping either rule. Set 25 May 2026 with Sush's express authority to interrupt execution. |
| Deployment Playbook | The 19-step pre-push checklist. CSS, JS, build, cache, smoke-test discipline. |
| Phase D Playbook (guided question-bank cleanup) | Read first when picking up Phase D / mashup cleanup / cert QA work on guided.aguidetocloud.com. State, priorities (P1 prevention hook is highest ROI), canonical per-cert workflow, triage rules per heuristic (H1/H2/H3/H4 + FP categories), discovered_bugs persistence model (git-tracked JSON ↔ session SQLite), self-learning protocol (build a scanner when you see a bug class 2+ times), stop criteria, resume one-liner. Self-updating doc — bumped at the end of every Phase D session. Born 3 Jun 2026 in session 9e5cc23f after 3 commits + parallel session 89327d9c shipped scanner v2 + B-Lite strip on juniper-jncis-ent. |
| Parallel-Safe Git Rules | The M-flag detection, worktree pattern, the May 2026 incident that drove the rules. |
| Voice & Tone | Sush's core values, the brag-allergy, phrasing fingerprints. How to write as him without flattening him. |
| Zen Design System Quickref | One-page lookup for tokens, guardrails, and the parity-file map between Hugo and Astro. Now includes the hub Zen pattern + voice library + Layer 2 cert-landing pattern. |
| Blog Notebook System | The Field Notebook layout for /blog/. Architecture, design tokens, frontmatter conventions, shortcodes, the "handwritten over emoji" principle, deployment history (8 May 2026), known issues. Read first if picking up blog work cold. |
| Blog Notebook Follow-ups Prompt | Self-contained starter prompt for the next session that tackles the 3 flagged items from the 8 May SME audit (JSON-LD nested-quote bug, SEO content lengths, Clarity CSP decision). |
| Hugo Build Wrapper | Why never to run bare hugo. The wrapper script, why builds take ~100s, and the lock-file recovery flow. |
| Tool Integration Checklist | The 12 mandatory touch points when adding a new free tool to the public site. |
| Production Incident Log | Every production bug + the guardrail that was added because of it. The "growing test suite" principle. |
| Site Audit & Cleanup Log (May 2026) | Working memory for the periodic site cleanup audit. What's been deleted, what's kept-but-redundant (and why), what's pending. The Cloudflare _redirects gotcha lives here too. |
| May 2026 Zen Shipping Log | Full chronological log of the Layer 1 + Layer 2 Zen redesign, OG/SEO push, paid-product loop fix. Read this first if you're picking up Layer 2 or Practice page work. |
| Shift Playbook | How to extend, audit, and deploy shift.aguidetocloud.com. Voice rules, Three Dials framework, deploy gotchas, audit personas, deferred backlog. |
| Plain AI Playbook | How to extend, audit, and deploy plainai.aguidetocloud.com. Content model, visual rules, the cache trap, the pivot history (AI for grandma → Plain AI), and the wrangler-on-arm64 workaround. |
| Agentic Planet Playbook | How to extend, audit, and deploy agents.aguidetocloud.com — the cockpit field-guide for techies building with agents. Origin story, the 14-pillar IA (minus 13/14), four iteration rounds, full pending-improvements roadmap, gotchas. Read first if picking up the project cold. |
| Claw Planet Playbook | How to extend, audit, and deploy claw.aguidetocloud.com — the plain-English study reference for OpenClaw. Origin story, the 10-§-section IA, locked visual decisions (B3 reference layout, OpenClaw red, Bracket Mark logo), voice guardrails, verification states, batch-by-batch workflow. Read first if picking up Claw Planet cold. Updated 14 May 2026 with the v0b multi-vendor expansion admonition — Claw now covers OpenClaw + Anthropic + OpenAI + Google + Microsoft. |
| Planet Pivot Playbook | Meta-playbook for repurposing an existing planet when its scope no longer matches what Sush wants to learn out loud about. Captured 14 May 2026 from the Claw v0a → v0b multi-vendor expansion. Six patterns: (A) multi-turn requirements gathering ("creative-juice arc") · (B) when to call the rubber-duck · (C) sourced-seed content lifecycle · (D) Cat A vs Cat B problem (dev infrastructure vs hosted product) · (E) Microsoft slice + data classification discipline · (F) Batch 0 migration phases in safe order. Includes anti-patterns, lessons learned, full file map, and a checklist for the next planet pivot. Read first if Sush proposes a new planet OR a scope expansion of an existing one. |
| Claw v0b Follow-ups Prompt | Self-contained kickoff for the next session that picks up Claw v0b after Batch 0 shipped (14 May 2026). What's live, what's queued (Batch 0b URL migration + Batch A Anthropic content), required reading, working pattern reminders, SQL todo state, full file map. |
| Cosmos Philosophy | The metaphor that governs how the universe expands. Universal laws, planet-internal laws, inter-planet contracts. Read first before any new-planet work. |
| Cosmos Nav Playbook | How the cross-planet navigation rail works across all 5 worlds. Schema, file map, per-planet renderings, how to add a new planet, SLA guardrails. Built 5 May 2026. |
| Cosmos Config Manifest | The [cosmos] block — newsletter URL, feedback link, parent site. How to swap newsletter platform in 5 file edits, no code. Brainstorm of what else could use this pattern. |
| Cosmos Audit & Resume Doc | The universe-wide coordination map — 8-planet status table, locked decisions, the 10-rule GUARDRAIL for every future cosmos session, doc structure standard, new-planet onboarding checklist, resume pointer. Now also houses the Cosmos Atlas (Planet #8) physics gotchas section — moonScale ≠ scaleFactor on iPhone, sun-occlusion fade trap in alt-lenses, focus-scroll trap on programmatic card open, audience lens col 0 overlap with ambient player, screensaver fade per-element control, smooth-card-open layered fix. Read first when picking up universe-wide work cold OR when touching cosmos-atlas/src/scripts/cosmos.ts, atlas.json, orbit logic, lenses, or interaction physics. Grew R3 (12 May 2026). |
| Brain Bar Lessons | What worked + what to do differently from the first planet build. Patterns to reuse for the next planets. |
| Memory System Architecture | One-page Mermaid map of the 6 memory tiers (constitution · reference · journal · annex · archive · session-store SQL). When each tier gets read, what triggers loads, what NOT to add. Read this if asked to "explain the memory system" — answer in seconds instead of composing from scratch. Built 10 May 2026. |
| Skills Strategy | The agentskills.io spec digested + our build list + the four meta-gates (voice / quality / last-used / meta-review). Captures: existing skills audit (14 found at ~/.copilot/skills/, mostly unused), why they don't trigger (description fit, not infrastructure), Category B priority list (voice-review + connect-classify P0), Category C script-wrap list (gsc/stripe/realtime weekly trio P0), Atlas extensions to the metadata field (last_used, voice_critical, surface), the skills-meta self-policing skill, build order, and the open questions for Sush. Read first when picking up any skill work — new build, retire decision, eval design, or "why isn't this skill triggering?". Built 21 May 2026 in session 09a04d64. |
| SEO Sweep Phase 2 Prompt | Earth-only sweep, ✅ SHIPPED Mon 11 May 2026. 845 failing → 50 failing across all Tier 1+2 surfaces. 818 page rewrites in 7 commits. 8 sibling scripts in aguidetocloud-revamp/scripts/. Full results table inside. |
| SEO Sweep Phase 3 Prompt | Cosmos-wide SEO sweep — Phase 3 hand-off. Per-planet audit (Brain Bar, Guided, Shift, Plain AI, Cosmos Atlas, Agentic Planet, Claw Planet), live homepage state, SLA touch-points, recommended commit order (lowest risk → SLA-touched last). ~100-170 pages to rewrite across 7 planets. Read first when picking up cosmos SEO work cold. |
| Stripe Payment Playbook | Everything Copilot needs to know about the aguidetocloud.com Stripe setup. What's possible via API, what isn't (Radar Rules API doesn't exist; Cash App/Klarna/Affirm aren't region-available), what's already shipped (11 May 2026 baseline: 3 commits to checkout.ts), the wallet discovery (Apple/Google/Link work under card automatically), the Payment Method Configuration landscape, the pre-test pattern for checkout.ts changes, fraud cluster handling via Radar value lists, common pitfalls. Read first if picking up Stripe / paid product work cold. Built 11 May 2026 alongside ~/.copilot/scripts/stripe-weekly.py. |
| Blog Cover Prompt Template | Reusable recipe for OG / LinkedIn cover images for /blog/ posts. Locked palette · medium · composition rules · mood. Reusable prompt template + 3 example scene prompts (Copilot CLI / monthly updates / licensing). Modelled on Darren Johns's "Teaching an Old Dog New Tricks" article cover (4 May 2026). Paste-into-Designer-or-ChatGPT workflow; the "Future: programmatic generation" section is now real and points to og-image-system.md. Read first when writing a new blog post and need to generate its cover image manually. Built 11 May 2026, updated 12 May 2026. |
| OG Image System | The locked two-tier OG cover system for aguidetocloud.com. Tier 1a — cert study guide + practice exam ✅ SHIPPED 13 May 2026 PM — V3 programmatic SVG (dark #0F0F10 BG + Inter ExtraBold cert code + indigo accent + family-coloured band + ink-lotus + wordmark), 149 study guide covers + 125 live practice exam covers, Microsoft 5-family palette (AZ peach · MS pink · AI lavender · SC periwinkle · MB teal) + DEFAULT neutral for AWS/Cisco/CompTIA/GCP/etc. Generator at aguidetocloud-revamp/scripts/og-generator-cert/, in-place atomic replacement (no baseof.html change). Practice mode reads guided/src/content/certs/*.toml as strictly READ-ONLY. Tier 1b — blog covers ✅ SHIPPED 13 May 2026 PM — V3-blog "B2 Editorial-Light" — warm-white #FAFAF8 BG + Inter ExtraBold headline left + tiny line-art glyph corner-right (calendar / compare / layers / list) + lotus+wordmark bottom-left. 18 covers live at static/images/og/blog/, generator at scripts/og-generator-blog/. 🔴 Universal format lock — JPG @ q=85 with mozjpeg encoder + 4:4:4 chroma subsampling + stripped metadata (§ 6.11.3) — apply to ALL OG generators (Node V3 cert, V3-blog, legacy Python next time touched). Yields ~22-31 KB per cover, sharp text edges on coloured accents. Tier 2 Azure Foundry gpt-image-2 watercolour path is ARCHIVED. Peer-set audit (Linear/Vercel/Stripe/Cloudflare/MS Learn — 11 reference covers in session-state), v1→v4 iteration history, family-palette lock, adaptive font-sizing rules, frontmatter contract (og_headline + og_glyph), in-place vs /v3/ route trade-off, guided-repo read-only contract, pre-existing SEO guardrail follow-up. 🆕 § 6.16 (19 May 2026) — operational hardening documents the two-generator clobber trap (Python clobbers V3-blog when both write to same path), silent og_glyph fallback, and the three new check-seo-lengths.ps1 checks (strict-fail on missing og_headline · warn on invalid glyph · warn on OG >50 KB legacy-format heuristic). Read first if picking up OG image work cold OR if asked to "make a cover image". Built 12 May 2026, V3 locked 12 May PM, V3-blog SHIPPED 13 May PM, V3 cert+practice SHIPPED 13 May PM, ops hardening 19 May. |
| Cosmos Intelligence Playbook | The cosmos-wide live counter + private analytics dashboard. GA4 single property + cosmos_planet event-scoped custom dim → public pill on every planet via cosmos-bar + gated 🌌 Cosmos tab in Command Centre + nightly summary worker with KV stale-while-revalidate. Full architecture, auth model (CC SHA-256 hash + plaintext Bearer + constant-time compare), file map across 8 repos, NZT week math, GA4 gotchas, signals engine, Phase B roadmap (outcome events, journey, GSC, LLM memo), 13 gotchas-with-cost. Read first if extending cosmos analytics, adding a new planet to the counter, or touching /api/cosmos-summary or /api/stats?realtime=cosmos. Built 12 May 2026 — two rubber-duck passes + one SME code review + 14 commits across 8 repos. |
| LinkedIn Advisor Playbook | Thought-advisor mode — turning customer sessions into LinkedIn posts that push existing blogs without feeling promotional. The formula (hook → context → reframe → visual → link → 2 hashtags), three hook variants (customer-question · honest-admission · T3-pattern), voice guardrails (extends Voice & Tone), sizing rules (~80 words), visual rules per topic shape, cadence (1/fortnight, trigger-based not date-based), the trigger system (what I scan for at end of each customer-session journal entry), a queue of blog posts ready to push with their trigger conditions, posted log, and what NOT to post. Read at the end of every customer or T3 session to scan for a moment worth posting. Built 12 May 2026 alongside the brand-kit decision-tree post. |
| Realtime Counter Playbook | The live "people in cosmos" pill + Site Analytics tile + tool-counter + CC "Live Now" — architecture, file map, operational runbook. Captures the 4-incident history (13/14/16 May 2026) and the permanent fix shipped 16 May (commit 6a9f6b7b, Option A+ "decouple public realtime from GA4 quota via scheduled KV refresh"). Architecture: GHA cron → handleRealtimeRefresh → realtime:active + realtime:pages KV keys → public handlers read KV only (fail closed, never fall back to GA4). GA4 calls/hour: O(visitors × poll_rate × N_pops) → constant ~60. Includes a 10-row "the pill is broken again" troubleshooting table, the ~/.copilot/secrets/guided-admin-password ≠ CC admin password gotcha, and the cross-quota-dimension lesson (per-hour vs per-day GA4 limits, different reset windows). Read first if the live counter shows nothing OR before touching functions/api/stats.js realtime paths OR .github/workflows/realtime-refresh.yml. Built 16 May 2026. |
| MSFT Gold Explore Queue | Curated triage of 20 internal mcaps-microsoft repos discovered in the 21 May 2026 second-pass scan (817 repos scanned, 195 name-matches). Tiered: 5 install-this-week (mcp-gateway, MSX-MCP, iq-core, MSX-Milestones-Skill, clawpilot-seismic-bridge) · 4 try-and-compare (CSA-Sherpa-Agent vs Connect tracking, Frontier-Fast-Start vs Frontier SE, copilot-account-research, MSXelerate) · 4 ANZ-relevant learn-from (belux-clawpilot as anz-clawpilot template, uspro-se-toolbox governance pattern, clawpilot-skills marketplace pattern, Cooking-with-Cowork inner-source framework) · 10 know-it-exists. Includes install order, why-now for each, decision criteria for the try-and-compare set, and a draft spec for PAC's future "Scout" tab (weekly auto-scan + LLM-scored digest — none of the existing internal tools do this, so it's PAC's defensible differentiator). Read first when picking install priorities for Sush's CLI/MCP fleet, or when designing PAC's Scout tab. Built 21 May 2026. |
| Personal Atlas Cockpit Spec | Self-contained starter prompt for Phase 0 of Sush's purely personal mission-control app — menubar/tray + full window, "Today" popover + tabbed deep-work window, 4 visual directions locked (TARS Console / Paper Cockpit / Linear Dense / Mission Departure Board), full PII allowed (machine-local + OneDrive Corp), MSFT-tenanted Azure OpenAI only, Tauri preferred over Electron. NOT to be confused with Frontier SE (org-wide) or anz-clawpilot (parked ANZ-team scaffold). HARD RULE: Phase 0 = static HTML mockups ONLY, no code, no MCP, no real APIs, no git init. Includes mock data set with real names (BNZ Yahya, Tam, Riki, Daniel, Spark, ASB), pickup pointers, and the 5 open questions the next session must ask before mocking. Read first when picking up Personal Atlas Cockpit work cold.** Built 21 May 2026 in session 3386ffee. |
| PAC Architecture & Conventions | Reference playbook for PAC v0.2.1 (22 May 2026 cut). Why MSAL was ripped out (CA-blocked twice); the WorkIQ MCP + Copilot-CLI---acp internal-first replacement; process model + IPC surface; the 22 May sidebar layout regression story and the 3 QA gates that lock the fix; env-scrub discipline; ACP read-only tool defaults; full QA gate catalogue (24 gates at cut); v0.2.2 candidates. Read first when resuming any PAC work, especially before touching main process, IPC, or sidebar/layout code. |
| PAC v0.8 Second-Brain | PAC v0.8 architecture (22 May 2026 overnight cut). Triage/Calendar/Threads three-tab workspace over a flat-JSON store fed by workiq.cmd. Mermaid diagram of the data flow. Three Architecture Gates (bridge approach · store backend · draft delivery — the last being the hardest: m365_create_draft doesn't exist in CLI surface, so Option C ships voice-aware drafts to ~/.copilot/drafts/*.md and swaps to Outlook Drafts in v0.9 behind the same IPC contract). Threading heuristic (Jaro-Winkler + 3× prefix normalisation). 168 QA gates total at cut, 86 unit tests across ranker/threads/draft. Bundle 396.6 KB JS / 88.5 KB CSS. Read first when picking up PAC v0.9 (the Outlook-drafts-for-real path, auto-refresh timer, CSS budget breach guard) or extending Triage / Calendar / Threads. Lessons from the overnight build at the bottom. |
| M365 + Clawpilot Playbook | The canonical answer for wiring M365 (email / chat / calendar) into Atlas without building a new app. Set 27 May 2026 from Issy peer-AI knowledge transfer. The auth recipe (MSAL acquireTokenInteractive in embedded browser, scopes against /me user-delegated — no admin consent wall): Mail.Read/Send/ReadWrite, Chat.Read/ReadWrite, People.Read, Calendars.ReadWrite. Direct Graph vs WorkIQ routing (Graph for nouns, WorkIQ for verbs). Skills pattern (markdown files, LLM is the renderer). MSX probe-before-render skill pattern. Path C Helm + Clawpilot connector (Clawpilot owns M365, Helm sidebar asks Clawpilot via ACP when customer context demands). Failure-mode matrix mapping every PAC + atlas-portfolio bump to the alternative. Execution checklist for the next session that acts on this. Read first when wiring M365 into anything — this is the canonical answer. |
| atlas-portfolio Architecture | Successor to PAC + Atlas CC (both paused 2026-05-24). Static HTML + node script + msx-mcp stdio subprocess — third time's the charm. Why Path B (static HTML, NO Electron) won the architecture gate; the az login auth pattern (NO MSAL) lifted from Frontier SE / msx-mcp; the canonical MCP SDK Client subprocess pattern; right-anchored markdown parsing (defensive default — opp names contain pipes); VPN profile gotcha (MSFT-AzVPN-Manual not MSFTVPN-Manual for prod MSX IP allowlist); upstream msx-mcp Node 24 + Windows execFile patch (PR #431 on mcaps-microsoft/msx-mcp); M365 filter heuristic; snapshot schema v2 with per-customer opps embedded; PAC Atlas Console v3 design DNA (Inter + JetBrains Mono, #fbbf24 amber, #06b6d4 cyan, typewriter // XX · UPPERCASE section labels, tilde ~ markers). Two standing rules: NEVER load mock data (real MSX only) · scope is M365 / Copilot / Agent 365 only (NO ACR / NO MACC). v0.1 shipped 4 phases in 5 hours on 2026-05-24. Read first when resuming any atlas-portfolio work. |
| Clawpilot Embed Research | Parked 2026-05-27. v2.2.0 attempted, reverted to v2.1.10 baseline. Phase 0 source-read findings (no external API, mini-mode architecture, single-instance lock, Teams Relay path, identity model); decompiled app.asar location; what's possible vs not for external integration; why rc1 SetParent+positioning failed (multi-monitor + DPI + Z-order race + diag-logging blind spot); why rc2 Path B "side-by-side launcher" felt insufficient (Sush's emotional test failed: visible seam, two-windows pretending to be one); 7 concrete recommendations for the next attempt (read MS-internal gim-home/m source first; run a hardware-specific SetParent spike; investigate inverse "Clawpilot embeds HELM" direction; consider Teams app embedding; fork as last resort); artifacts preserved in session 591b6a54-1de9-4542-a77a-079ba239d967/files/ including the extracted asar (~505 MB), the working PowerShell bridge code, and the Playwright self-test. Read first when picking up Clawpilot embed work cold. |
| atlas-portfolio Data Pipeline | The operational counterpart to atlas-portfolio-architecture.md. What gets scraped, in what order, into what file, and how to verify the result. Source-of-truth map covering MSX (msx-mcp → snapshot-customers---c360 → PRU/AVD/agreements) + Lynx home (one-off, tenant map) + Lynx /reports/your-tenants (portfolio rollup) + Lynx per-tenant page (per-app MAU/WAU/DAU) + Lynx admin (cohort search OR Admin Configs tab — both kept, loader prefers tab). Execution DAG, dependency table, weekly tracker template Atlas copies into the session journal each refresh. Concrete QA sentinels (ASB Anthropic=Enabled, customerCount 50-60, etc.) — scripts/refresh-qa.mjs automates them and exits non-zero on red flags. READ BEFORE any atlas-portfolio scrape/refresh work; run refresh-qa.mjs after every refresh. Set 2026-05-25 in v1.0.1. |
| atlas-portfolio OneDrive Backup | Backup playbook for the MICROSOFT CONFIDENTIAL data layer. Local working dir at C:\ssClawy\atlas-portfolio\ (incl. data/ and dist/ — gitignored, MS Confidential) mirrored weekly to corp OneDrive at C:\Users\ssutheesh\OneDrive - Microsoft\_AtlasPortfolio-backup\. Scheduled task \Atlas\AtlasPortfolioOneDriveBackup runs Sundays 06:30 NZST (30 min before AtlasPortfolioFullRefresh at 07:00, so the backup captures the previous week's state BEFORE the refresh overwrites). Robocopy /MIR mode; excludes node_modules, .git, .copilot, *.log. Script at scripts/backup-to-onedrive.ps1 (manual run: pwsh ...\backup-to-onedrive.ps1, dry: add -DryRun). Includes restore procedure (clone from GitHub + robocopy from OneDrive + npm install + QA verify) AND cleanup TODO for a pre-existing PERSONAL OneDrive copy that violates Standing Rule #4. Set 2026-05-25 with Sush at work. |
| Atlas HELM Design Philosophy | 🎨 THE LIVING CONSTITUTION for HELM visual design. Read first before ANY phase that touches a HELM visual surface (today-tab, overview-tab, customer cards, scout cards, etc). Set 2026-05-29 (session 386fd1f6) from Sush's "make each card tell its own story" directive — bento mosaic principle (cards have shape-fit-to-content, not stacked uniform rectangles), 5-second cold-user test, story-shape mapping table for Today cards (hero=2×2 anchor · timeline=4×1 strip · top-3=1×3 tall · hot pipeline=3 squares · atlas brief=2×3 serif narrative · scout=3×1 strip), IBM Plex Sans + JetBrains Mono + IBM Plex Serif typography plan (replaces HELM's current Inter, matches what PAC actually uses), amber/cyan discipline (amber reserved for STAKES only — money/customer/time at risk), 8 better-than-asked queued moves (sticky hero, freshness ink, growable cards, sparklines, pinning, hover halos, two-state cards, quiet-day creative space), Path A grid-template-areas skeleton with mobile collapse, risks + guardrails + change log. Intentionally FLUID — expect v2/v3/v10. Each session may EDIT this doc. Phase 200 series in ~/.copilot/atlas-portfolio-phases.md carries the multi-session execution plan. |
What does NOT live here¶
- Rules that must be followed every session — those stay in
copilot-instructions.md - Sensitive operational details — internal tenant info, customer-specific paths, anything Microsoft-confidential
- Secrets, tokens, passwords — never in either file. They live in
~/.copilot/secrets/ - Calendar items (self-reminders with dates) — those stay in
copilot-instructions.mdso they fire on session start
Editing rules¶
When future-Copilot adds a new entry here:
- Use a focused, scannable format — tables, bullet lists,
!!! noteadmonitions - Lead with the answer, then why, then examples
- Cross-link liberally between pages
- If a rule emerges that must apply every session, copy the rule line into
copilot-instructions.mdand leave the depth here - Run
mkdocs build --strictbefore committing to catch broken links - Add a row to the table above when creating a new file. The table is the directory; the playbooks are the books. (Set 10 May 2026 alongside the Learn-Doc-First Rule.)
- Naming:
<topic>-playbook.md(ongoing) ·<topic>-system.md(design system) ·<area>-quickref.md(cheat sheet) ·<event>-incident.md(post-mortem) ·<topic>-followups-prompt.md(kickoff prompt for next session)
Why this section exists¶
Sush asked for it. Direct quote from May 2026:
"make sure my study sections are intact and create a new memory section for you and move all things related to you there. and reorganise everything as it makes sense, and rebuild things as it makes sense. You have full creative freedom and control and I trust you."
The design intent: shrink copilot-instructions.md to the minimum set of cross-cutting rules, and offload everything else here so future-Copilot can grow this annex over time without fattening the always-loaded instructions file.