AIOS — flow and process for daily use

Context

AIOS (AI Operating System) is Aspire's 6-layer business operating system powered by Claude Code. New team members and agents consistently ask "ok how do I actually use this thing?" — this page is the canonical answer.

The 6 layers (mental model)

Mental model: layers 1-3 are input (what AIOS knows), layer 4 is synthesis (daily output), layers 5-6 are execution (acting on it).

Daily flow (one human, one Claude session)

07:00  Run /daily-brief
       → reads Context OS + last Data OS snapshot + recent Intel OS
       → writes aios-workspace/briefs/YYYY-MM-DD.md
       → highlights: yesterday's revenue, today's bookings, blockers,
                     hot threads from meetings, weather/ops alerts
       → ~2-3 min to read, sets direction for the day

09:00  Triage inbox / Slack / email
       → if a task is repetitive → tag for /task-audit later
       → if a task needs strategy → use marketing/engineering skills inline
       → if a task is novel → /explore to walk through automation potential

13:00  Mid-day sync (optional)
       → /data-os if you need fresh numbers
       → /intel-os "what did <person> mention about <topic>?"

17:00  Wrap
       → if you discovered a new playbook today, add to aios-workspace/marketing/sop/
       → if a new automation crystallized, /n8n-automation to scaffold the workflow JSON
       → AIOS doesn't journal for you, but Recall MCP indexes everything you touched

Engineering lifecycle (sequential — each phase gates the next)

For new products / features:

/spec   → discovery questions → SPEC.md
          (user MUST approve before /plan)
            ↓
/plan   → vertical task slicing → tasks/plan.md + todo.md
          (user MUST approve before /build)
            ↓
/build  → test-driven implementation per task
          (each task: RED → GREEN → full suite → commit)
            ↓
/review → 5-axis review (correctness, readability, architecture, security, perf)
          (no Critical issues open before /ship)
            ↓
/ship   → 6-domain pre-deploy checklist → deploy → verify
          (health check passes, clean logs)

Each command has its own skill file under aios-workspace/engineering/skills/ with evidence gates (concrete proof required before "done"), anti-rationalization tables (pre-loaded rebuttals for common AI shortcuts), and red flags (indicators the skill was misused).

Cross-cutting engineering skills (aios-workspace/engineering/skills/external/)

Adopted from mattpocock/skills (Sept 2026 selection), adapted to Aspire's vault-as-doc-source model:

These 3 are adjuncts to the 5-stage lifecycle, not replacements. Sequence:

• grill-with-docs runs BEFORE /spec to stress-test the proposed scope
• diagnose runs DURING any stage when bugs surface, then returns to the original stage
• improve-codebase-architecture runs OUTSIDE the lifecycle as a periodic background task (quarterly per project)

Marketing flow (campaigns / content / data)

Single tasks → use the skill directly:

Multi-deliverable campaigns → use the lead agent:

@campaign-strategist (Opus — orchestrator)
  ↓ delegates to:
  @market-researcher   (Ahrefs + web search + briefs)
  @content-creator     (blog, email, lead magnets, copy)
  @creative-designer   (Canva, Figma, HTML, PPTX)
  @data-analyst        (Ahrefs, SendGrid, dashboards)

All marketing agents MUST read these 3 context files before producing anything:

1. aios-workspace/marketing/context/product-marketing-context.md
2. aios-workspace/marketing/context/brand-voice.md
3. aios-workspace/marketing/context/marketing-strategy.md

Decision tree: agent vs skill vs raw slash command

Is the task strategic OR needs synthesis across multiple data sources?
├── YES → use an @agent (campaign-strategist, market-researcher, etc.)
└── NO  → Is the task executional with a clear template/formula?
         ├── YES → use a skill directly (blog-writer, email-sequence, ad-creative...)
         └── NO  → Is this a business-operations question (briefs, automations)?
                  ├── YES → use a slash command (/daily-brief, /brainstorm, /explore)
                  └── NO  → Just ask Claude; the right skill auto-loads if applicable

Routing examples:

Connected MCP tools — what you can actually call

Setup flow (onboarding AIOS for a new business)

1. Set Context OS — /context-os walks through who/what/team/values/products → writes aios-workspace/context/{business,strategy,integrations}.md
2. Connect MCP tools — add the relevant servers to claude_desktop_config.json or ~/.claude.json. For credentials, use the Aspire Hub secrets vault, not 1Password.
3. First Data OS snapshot — /data-os to pull live numbers; sanity-check totals match what the business owner remembers.
4. First daily brief — /daily-brief and confirm the highlights feel right; if not, the Context OS needs more detail.
5. Run /task-audit — surfaces what's repetitive enough to automate.
6. Pick 1-2 automations from /brainstorm — small wins first; scale to bigger flows once trust is built.
7. Install /modules — only after the above 6 steps; modules layer on top of working context.

Maintenance cadence

Common scenarios — what command to use

Planned trials (under evaluation)

safishamsi/graphify — code-graph audit tool

Status: ✅ ADOPTED as quarterly audit step (2026-05-20) Trial result: Gate met — 3 new cross-coupling facts surfaced (auth function blast radius quantified, Telegram comms dominance, 8-app shared-deps cluster) on first run against aspire-digital-group monorepo. See graphify-trial-2026-05-20 for full output. Scope: codebase → knowledge graph (interactive HTML + MCP server, 31 language extractors)

Where it fits:

• 🟢 Quarterly audit of aspire-digital-group monorepo (15+ apps + shared packages)
• 🟢 Spotting copy-paste duplication across the 7 zinga-* apps
• 🟢 God-node detection in heavy legacy code (alby-studio, larger zinga apps)
• 🟢 Visual artifact for onboarding new engineers

Where it doesn't fit:

• 🔴 Cross-repo flows (single-repo scope per run)
• 🔴 Knowledge OS vault (markdown not code; KO Stage 2 LightRAG is the right tool when query volume justifies)
• 🔴 Tiny repos (zac, edm) — too small for graph analysis to add value
• 🟡 Overlaps existing tools for trivial questions (IDE Go-to-Def + git grep + pnpm why already win on speed)

Tiered adoption plan:

1. ✅ Phase 1 — Trial DONE (2026-05-20): uv tool install graphifyy + graphify update . on aspire-digital-group. Result: 2,209 nodes / 2,935 edges / 220 communities / 435 files / 12 sec extraction. Outputs graphify-out/{graph.json, graph.html, GRAPH_REPORT.md}.
2. ✅ Phase 2 — Gate decision: YES. 3 cross-coupling facts surfaced that were genuinely new (auth blast radius quantified, Telegram comms dominance, 8-app shared-deps cluster). Adopt as quarterly audit step.
3. 🟡 Phase 3 — expansion in progress:
4. [x] alby-studio (149 files / 783 nodes / 1427 edges — DONE 2026-05-21; see graphify-audit-2026-05-21). 3/5 new findings — cn() 88-edge god-utility, getVendorBySlug() storefront spine (21), authOptions 30-edge auth surface.
5. [ ] zinga-pms (31 source files — small but the largest Zinga product)
6. [ ] MCP servers batch 1 (group of 5-8 servers for cross-server pattern detection)
7. [ ] Re-run aspire-digital-group quarterly

Install / use:

uv tool install graphifyy          # one-time
cd <repo>
graphify update .                  # build/refresh graph (no API cost, AST-only)
graphify explain "<symbol>"        # explore a specific node
graphify query "<question>"        # BFS over the graph
graphify path "<from>" "<to>"      # shortest path between two symbols

Operating rules:

• Add graphify-out/ to each audited repo's .gitignore (build artifact, ~3 MB per repo)
• Run quarterly via external/improve-codebase-architecture skill — NOT in daily workflow
• Do NOT wire the graphify MCP into agents by default — too narrow a use case
• Re-run after major refactors to confirm complexity moved (not just relocated)

Cost / risk: $0 (open source), ~12 sec per run on medium monorepos, no lock-in (open HTML/MD/JSON outputs).

Hand-off across Claude accounts (multi-account coordination)

Multiple Claude accounts share this Mac's filesystem. To prevent duplicate work:

1. Read ~/Desktop/Claude/PROJECTS.md FIRST — master project index (slug, aliases, canonical path, status, owner)
2. Check the project's _STATUS.md — know what previous sessions did and what's queued NEXT
3. Check .claude-lock — if present + fresh (<30min), coordinate before editing same files
4. On session end — append to _STATUS.md (date, account, model, files touched, NEXT)

Provenance

Source: CLAUDE.md (per-project instructions), ~/.claude/projects/.../MEMORY.md (user auto-memory), aios-workspace/engineering/sop/dev-lifecycle.md, aios-workspace/marketing/sop/{campaign-launch,content-production,agent-routing}.md.

Related

• aspire-brand-architecture
• aspire-tech-stack
• aspire-llm-gateway
• knowledge-os-stage-1

🔗 Relationships