reference data classification

Classification levels

Level	Description	Storage rules	Backup rules
L1 — Public	Already public; no restriction	any	any
L2 — Internal	Operational data, no PII	mode 600/700, owner-only	quarterly
L3 — Sensitive PII	Names + emails + phones	mode 600, Keychain for tokens, FileVault recommended	quarterly + iCloud mirror
L4 — Restricted	Financial / SSN / health / OAuth tokens	Keychain ONLY, never in plaintext on disk	NEVER in unencrypted backups

Current GC tenant target: L3 baseline, with L4 secrets in Keychain (not on disk).

Inventory

L4 — Restricted (Keychain only)

Account	Service	Last rotated	Notes
`cloudflare_api_token`	SkyRun	2026-04-24 (.env mtime)	Account-level, full edit. ROTATE EVERY 90D.
`ntfy_topic`	SkyRun	2026-04-22	Public-by-design (push channel name) — Keychain'd for consistency, not security

Future L4 to migrate (currently using cookie-CSRF via chrome_bridge — should move to scoped API tokens):

HubSpot API token (currently relying on browser CSRF)
Gmail OAuth refresh token (currently held by Gmail MCP)
SmartLead API key (not used; CSV browser injection only)
BeenVerified session cookies (currently chrome_bridge'd)
Track / KeyData credentials (not yet automated)

L3 — Sensitive PII (mode 600 files, mode 700 dirs, FileVault recommended)

Path	Content	PII fields	Records
`~/Desktop/SkyRun/DNC_active_homeowners.json`	Current customer roster — DO-NOT-CONTACT canonical source	email, phone (last10), name tokens	~141 emails, 97 phones, 263 name tokens
`~/Desktop/SkyRun/knowledge_graph.json`	Operator KG — people, deals, meetings	name, email, phone, deal stage, transcripts referenced	39 people, 11 deals (post-2026-05-02), 26 meetings
`~/Desktop/SkyRun/Prospecting & Leads/master_lead_inventory_807.json`	Lead inventory — full prospect data	name, email, phone, address, AV, purchase price, OOS flag	865 leads
`~/Desktop/SkyRun/Prospecting & Leads/STR Engine Archive/Grand_County_STR_Engine_v3.5_filtered.xlsx`	Master workbook (4 tabs)	same as inventory + Score Reason + tier classification	865 rows × 41 cols (Lead Details)
`~/Desktop/SkyRun/Prospecting & Leads/Postcard_Targets_GrandCounty_v3.5.xlsx`	Postcard targets	name, mailing address	~789 rows
`~/Desktop/SkyRun/Call Transcripts/transcripts/*.txt`	Verbatim call transcripts	speaker names, content	~36 files (~558K chars)
`~/Desktop/SkyRun/Call Transcripts/notes/*.txt`	Meeting notes / summaries	same	many
`~/Desktop/SkyRun/Email Scans/*.json` + `insights_today.jsonl`	Gmail-scan output	email subjects, sender, snippets	accumulates daily
`~/Library/Application Support/SkyRun/pwa/data/pending_*.jsonl`	Approval queues	recipient names + emails + draft bodies	varies
`~/Library/Application Support/SkyRun/known_hs_duplicates.json`	Lead-id ledger of acked dups	lead_ids only (no PII directly)	30 entries
`~/Library/Application Support/SkyRun/health/*.json`	Heartbeats	task summaries, occasionally include names in summary text	30-day rolling
`~/Desktop/SkyRun/audit/<date>/*.json`	Skill audit summaries	varies	accumulates
`~/.claude/projects/-Users-josephbowens-Desktop-SkyRun/memory/project_active_deal_*.md`	Per-deal memory	full prospect context	~7 active deal files
`~/.claude/projects/-Users-josephbowens-Desktop-SkyRun/memory/feedback_*.md`	Voice + drafting feedback	includes draft examples with recipient names	many
`~/Desktop/SkyRun/Property Deliverables/<address>/*.{docx,pptx,pdf,xlsx}`	Per-property deliverables	owner name, address, financials, Track comp data	~per-prospect

L2 — Internal (mode 600/700, no PII directly)

Path	Content
`~/Library/Application Support/SkyRun/{chrome_bridge,dnc_check,system_hygiene,build_pwa,deploy_pwa,quarterly_backup,notify,pwa_auto_rebuild,fleet_status_push,recompute_health_summary,secrets,schema_guards}.{py,sh,md}`	Helper scripts
`~/Library/Application Support/SkyRun/.env`	Keychain-backed secret shim (no values in plaintext)
`~/.claude/scheduled-tasks/*/SKILL.md`	Skill prompts (operator-agnostic logic; no PII baked in directly)
`~/.claude/projects/-Users-josephbowens-Desktop-SkyRun/memory/reference_*.md`	Reference docs
`~/.claude/projects/-Users-josephbowens-Desktop-SkyRun/memory/MEMORY.md`	Index — paths + summaries (no raw PII)

L1 — Public

Path	Content
`~/Library/Logs/skyrun-*.log` (CARE — operational logs may contain snippets that approach L3; rotate aggressively)
`~/Desktop/SkyRun/Brand & Marketing/*`	Brand assets, public marketing

Handling rules

NEVER

❌ Email PII outside the operator's Workspace account
❌ Paste PII into a public LLM playground (use the local SkyRun Claude project only)
❌ Upload PII to a public diagram/paste service
❌ Commit PII to a public Git repo
❌ Include PII in a screenshot shared on social/Slack/anywhere external
❌ Store L4 secrets in plaintext on disk (use Keychain via secrets.py)

ALWAYS

✅ chmod 600 on L3 files; 700 on L3 dirs (gate-proof + system_hygiene auto-restore drift)
✅ Use Keychain (secrets.py get/set) for all L4 tokens
✅ Quarterly backup runs include all L2 + L3 (3-source: user + runtime + memory + secrets manifest)
✅ Token rotation reminder fires at 90d; rotate at 90d
✅ Verify FileVault is ON (currently OFF — top operator action)
✅ Audit trail: every write to L3 files logged in ~/Desktop/SkyRun/audit/<date>/

Subject Access / Right-to-Be-Forgotten

If a lead requests removal:
1. Find their lead_id (search master_lead_inventory by name/email/address)
2. Add to DNC_active_homeowners.json with reason "do-not-contact-by-request" + remove lead_id from outbound queues (SmartLead, postcards, pending_drafts)
3. Mark KG entity do_not_contact: true
4. Optionally: blank PII fields in master inventory (keep lead_id + DNC status; remove email/phone/name)
5. HS contact: set lifecyclestage=other and add a note recording the request
6. Ledger the action in ~/Desktop/SkyRun/audit/<date>/dsar_<lead_id>.json with timestamp + scope

Cross-references

reference_dnc_system.md — DNC enforcement
reference_security_posture.md — overall security architecture
reference_vendor_security_posture.md — vendor risk + DPA tracking
reference_incident_runbook.md — IR scenarios
reference_disaster_recovery.md — DR procedures