TD-023: Layer 1 git auto-sync done; cross-surface adapters open | DEMOTED by WIP cap (3/agent) – OSCAR had 5 running at once. Not cancelled; requeued.

AOS licensable product: schema->Claude/ChatGPT/Grok compiler + governance

Parse Connections.csv (2,335 contacts) -> company->[name,role] for outreach routing

Parse Gmail job-alert/recruiter mail; target 2-3 jobs/day

Deep research + network lookup -> draft intro-ask + Gmail draft, queued for Greg

Let Greg log 'I did X' into the tracker; HUNTER to propose scraper scope

Every skill callable from any Claude surface; needs a design pass

Title+priority -> auto-creates task, fans out to LYRA + WENDY; no gates

Portable least-privilege Drive read/write/search for ChatGPT/Gemini/others

chrome://policy on Greg's connected Chrome profile blocks navigation to drive.google.com, pivotnreboot.com, everything tested tonight (2026-07-15). Not a login/access gap – browser-level policy enforcement, cannot be bypassed via 'already logged in' or incognito. Fix needs Greg to check chrome://policy and remove/adjust the policy, OR use a different unmanaged browser profile for claude-in-chrome. Blocking: any browser-automation task on this Mac.

Same media-26 data source as /dashboard – one canonical feed, all boards realtime | DEMOTED by WIP cap (3/agent) – WENDY had 4 running at once. Not cancelled; requeued.

Greg idea 2026-07-17: a daily board where every agent reports what it learned + what knowledge it's accumulating. Source = qa/qa_lessons.jsonl (already accumulates every correction) + per-agent knowledge. WENDY designs the display; LYRA turns each agent's entry into a per-agent build-in-public post. Ties the memory/learning work into something visible.

Greg idea 2026-07-18: a standard 'prime' that brings any fresh channel to baseline awareness (connectivity via /connectivity, roster, live sprint/backlog + agent activity) and a blank-channel TEST that probes 4 awareness questions (how do we deploy .cloud / what sprint+blockers / what are agents doing / can you reach the brain) scoring answered-from-loaded vs re-derived. RED-TEAM (OSCAR): storage!=retrieval – the real fix is a consult-trigger (load /connectivity before any external action) + QA-QUINN blocking uncited connectivity actions + supersede-not-append discipline. Static awareness primeable now (paste block / repo); live cross-LLM awareness gated on Supabase brain. Align to Greg's ChatGPT agent-block template (Greg to share).

Unblock kit built; tier-3 (Supabase/cross-LLM bootstrap) pending. Same blocker as #9. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had. | DEMOTED by WIP cap (3/agent) – OSCAR had 5 running at once. Not cancelled; requeued.

The single build that 3 of the 4 open TEN THINGS items are all waiting on. Supabase brain as shared peer store so ChatGPT/Gemini stop needing Greg to copy-paste. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

QUINN AUDIT 07-16: demoted from completed. Secrets quarantined locally (CREDENTIAL_MAP/RECOVERY docs confirm); Google-side token revocation UNVERIFIED – needs read-back on the provider console. Pairs with SS2-B04.

Metrix follow-up drafted ($150k/yr confirmed); Beacon Hill/SitusAMC apply; fresh roles cast

P0 funnel conversion work

P0 content supporting the funnel + job search

2,473 raw inbox rows to curate; only curated exposed to keyless surfaces

De-fragment brain homes + folder/dup cleanup; delete steps Greg-gated

TD-022: 6 squash-merged; 6 conflicted (OSCAR/agent cluster) need read+resolve

Shared-store decision made; wiring pending

RECLASSIFIED (QA-QUINN 07-18): NOT Greg-only – it's an investigation. Pull wp-admin revision history via cookie+nonce (same access that deployed .cloud) to see who changed footer getstarted@->info@. TD-029: getstarted@->info@ on prod, no run-record author; check wp-admin revision history

RECLASSIFIED (QA-QUINN 07-18): 'MCP cannot rename' is obsolete – KAT rename script + Mac Drive reference can rename (non-destructive). Read-back-prove before done. TD-001/002/003: MCP cannot rename folders; human rename in Drive UI

Reads Supabase brain curated bucket via public anon key (RLS-scoped, Greg ruling 2026-07-13); inbox lock re-tested on every refresh

Tiers 1-2 done (Claude + GitHub); tier-3 ChatGPT/Gemini needs the Supabase load. Same blocker as #5. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

task_status.json unfrozen + board regenerated; awaiting QA-QUINN two-gate verification before Done

TD-031, Greg's #1 ownership priority; research prepped, only Greg can file (USPTO + copyright.gov)

EVIDENCE CLOSED 07-16 (OSCAR JWT decode + KAT full-vault sweep + TD-027 + 07-12/07-13 standups, all agree): key NEVER rotated at provider – vaulted key is the ORIGINAL mint (iat 2026-07-10T19:06:29Z, same second as anon). The 'rotation' Greg was told about = WP password rotation 07-11 + vault-passphrase rotation (different things). sbp_ management PAT exists in NO vault (index doc corrected). NEXT: Greg mints sbp_ PAT (Supabase dashboard -> Account -> Access Tokens) -> straight into vault flow -> agent runs supervised rotation.

Prior run FABRICATED an NVIDIA entry. Needs re-run on a real Gmail surface before any output is trusted. hunter.jobsearch launchd job is DISABLED pending Greg's call. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

Dashboards render (verified live 07-16). QUINN note-fix: real page inventory = dashboards(22)/mission-control(24)/board(14)/dashboard(27)/memory(28) – old IDs 44/45/47 were wrong.

PENDING[9]: decrypt 219, 3/3 queries; brain ~6,539 total / ~2,410 curated

COMPLETE for real 07-16 8:10 PM PDT (Greg 'go'): SessionStart->session_recall.sh + Stop->autosync.sh registered in ~/.claude/settings.json alongside SessionEnd capture. Both pipe-tested (recall returns brain context incl. tonight's ID legend; autosync commits+soft-fails push honestly) + jq schema-validated. Live from next session start.

Curated bucket anon-readable; raw transcripts stay service-role locked (TD-019)

Two-column Founder|Community, read-back proven live

Cookie-auth workaround (TD-030); header-strip logged

Internal, Greg curates. QUINN path-fix: list lives at ~/Projects/pivotnreboot-revenue-engine/outreach/2026-07-10_enterprise-targets.md (exactly 10 targets, verified).

QA-QUINN_CONTRACT.md crystallized + docs/qa/QA-QUINN-INTERNAL-DASHBOARD-DEPLOY.md

ARIA retired (confirmed _retired/aria.md). QUINN note-fix: executive-assistant duties (mail triage/calendar/PA) absorbed by KAT per oscar.md GAP-CLOSED 2026-07-12; Parker->OSCAR, Mara->LYRA.

QUINN note-fix to observed reality: 2,301 transcript + 148 auto-memory staged, 22 secrets redacted (was misreported 1,993/318/18).

Interview COMPLETED & PASSED (Greg, 2026-07-18). Deck submitted 2026-07-15 (ABC_Health_Toptal_Presentation.pptx, 15 slides). Screener Alvaro requested 2hr push; Greg confirmed via sent email 11:04 AM PDT. Interview now 1:00-2:00 PM PDT. Next: rehearsal + Zoom/notetaker check.

CLOSED by Greg sign-off 2026-07-16 8:2x PM PDT. BOB end-to-end proof same evening: gateway 200 with token / 401 without, semantic search live, 2,416 brain rows behind it, all memory-plane paths resolve. Token = AGENTMEMORY_GATEWAY_TOKEN (env-only, never in repo), consumed by n8n gateway + ChatGPT + Gemini.

Realtime 60s poll off canonical task_status; WENDY deploy via TD-030 cookie path; data vehicle media 26 alt_text (host blocks non-image uploads); QA-QUINN PASS w/ flags 2026-07-16

launchd com.pivotnreboot.dashboard-feed-sync every 5 min on the Mac: Drive canonical to WP media 26 (hash-deduped, fail-soft, logged, write-verified). Any surface updates Drive canonical and every board refreshes itself within 6 min – no human, no Mac session needed. 2026-07-17. INCIDENT same night: a '‹=' in this very note tripped WP kses entity-escaping and corrupted the feed; sanitizer + write-verification added to the sync (QUINN-gated). | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

Scrum lanes To Do/In Progress/QA/Blocked/Done, live 60s feed; scrum.org terminology; QA press-tested. 2026-07-18. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

Epics only (12 epics / 66 items); Done never Shipped; QA read-back PASS. 2026-07-18. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

SOLVED_MECHANISMS (SM-001/002) auto-loaded via CLAUDE.md; BOARD_REGISTRY; central logs/. Roster exactly 7. 2026-07-18. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

Root cause: no feedback + no entity-decode. Patch deployed to page 27; new binding QA rule in contract. TD-049 cache filed. 2026-07-18. | RECONCILED 2026-07-18: recovered from a diverged mirror that the repo copy never had.

Human folder-rename (Greg)

Human folder-rename (Greg)

Human delete/merge (Greg)

KAT relocate under control plane

BOB – wire journal-append into agent loop

Generalize `qa/qa_quinn.py` runner

Add promotion cadence

Wire WordPress/Hostinger connector

KAT consolidate to canonical, mark legacy `_STAGING`

Check `OPENAI_API_KEY` / `EMBEDDING_PROVIDER=openai` in the AgentMemory container on the VPS

Remaining value is only what the native meter lacks: prediction/extrapolation, a history trend, and cross-device view (phone/PC). Don't build the 2-hr dashboard unless Greg still wants those

Spike: Supabase-vs-Drive decision doc (cost, connectivity, migration effort)

HUNTER – HTML dashboard surfacing resume/draft status + links

WENDY – web-URL/change dashboard + connectivity check (see TD-009) + Wendy backlog

Greg trash the old mirror file in Drive UI

Greg deletes the 4 stale board HTMLs (deletes are human-gated); index already routes agents away from them

Auto-reconcile: on any status change, commit + regenerate board without waiting for a verbal "push"; or a periodic pull so the terminal never reads stale git. Interim: agents say "push" after edits; treat git as possibly-behind before reporting "not done"

Durable fix: n8n flow on the existing instance (Drive file → Gmail draft attachment, bytes never touch LLM context, verifiable server-side) OR a local token with gmail.compose scope. Also still owed: Greg's target-comp figure

Reconcile: either the agents/ contracts adopt the v1.0 schema, or scope/rename the schema so it doesn't claim to govern them

Confirm intent or map "candidates found" to exit 0 with a distinct signal (e.g. stdout marker)

OSCAR+BOB: define the completion-record schema, write on completion via scripts/supabase_connect.py, make the done-check query it

QA-QUINN: add the done-check as a mandatory pre-escalation step; a blocker with no done-check evidence is itself a BLOCK

LYRA+OSCAR: verify API rules, build publish adapters; migrate to revenue-engine/BACKLOG.md when that repo is in-session

OSCAR: check n8n logs on Mac/VPS. Scope: trace the 4 failing jobs, identify root cause (timeout, script error, missing credential, permissions), restore nominal 13/13. Investigate first after Hunter P0 work (SS2-P01 Toptal presentation) completes.

WENDY+OSCAR: design telemetry integration. Scope: uptime / HTTP status, page load metrics, form conversion funnel (AI Parachute sales funnel – SS2-P03), content freshness (last publish date per page). Queue after Hunter + SKED-036 investigation.

rotate-api authored once in the repo (canonical). Propagate the identical version to Cowork + web/chat so `/rotate-api` resolves everywhere at one version

Still OPEN (documented in merge contract env notes). KAT sweeps the root orphans; durable fix TBD

Mac-session action via `rotate-api`: encrypt → `SUPABASE_API_KEY.encrypted` (600) → wipe env/history → VERIFY real Supabase call → then revoke → register → launchd. Cannot be executed from the cloud/web channel

Design captured in docs/LOGGING_SYSTEM.md. Build `logs/activity.jsonl` (append-only, monthly-rotated) + `scripts/log_event.py` + session-start/end append; existing logs become views/feeders. Owner OSCAR/BOB

docs/SINGLE_SOURCE_OF_TRUTH.md captures the map + propagation checklist. Durable fix: off-repo homes shrink to POINTERS to repo CLAUDE.md, not copies. Off-repo ARIA cleanup staged in docs/ARIA_REMOVAL_OFFREPO_STAGED.md

WENDY deploy as child of page 27 → pivotnreboot.cloud/dashboard/backlogs/ via SM-002 cookie+nonce (Mac `deploy_backlogs.command` already written). Read-back verify, then send Greg the URL

Execute docs/ARIA_REMOVAL_OFFREPO_STAGED.md: Greg = claude.ai instr paste-block; BOB = brain rows (service_role, Mac vault) + committed migration; KAT = Drive docs

Owner-only: Greg or Mac session removes the `anyone` permission in Drive UI/API; then audit Contracts folder for other `anyone:writer` files (KAT next-contract)

Needs a decision, not a unilateral fix: is Drive (`177MPZZ…`, the multi-surface/ChatGPT/phone-writable canonical per `dashboard-refresh-path-drive-canonical` memory) or the repo (`board/task_status.json`, the new git-heartbeat canonical per tonight's HANDOVER/PICKUP docs) the intended single source going forward? Until decided, do not keep patching one side – the other job will erase it within its next cycle. Left both jobs running; did not unilaterally stop either.

Write a `generate_standup_board.py` (mirror `generate_scrum_board.py`'s pattern) sourced from current `board/task_status.json`, then deploy via `scripts/wp/deploy_dashboard.py –slug standup`. Owner: WENDY.

Claude/OSCAR 2026-07-10: unified read-only board shipped (`board/index.html` via `scripts/generate_unified_board.py`, renders all 3 canonical sources). Drag-to-move across columns = v2

**DONE this session:** #31 opened (`salvage/td-022-agentic-core`, 23 net-new files: orchestrator engine + typed contracts from #17, key-rotation+promote_lessons from #20, spam_health+gmail Apps Scripts from #2); #2/#16/#17/#20 closed with linking comments. **RESIDUAL:** (a) Greg reviews + merges #31; (b) orphan branch `agent/lyra` (SHA 065dbef) delete BLOCKED – proxy rejects delete-ref pushes (send-pack sideband disconnect), needs a Mac/UI delete → TD-028; (c) two QA FLAGs on #31 → TD-026, TD-027

ARIA scrubbed from ALL editable source layers: canonical `.claude/` + core docs, live root docs (RECOVERY_AND_SECURITY, OWNER_INDEPENDENCE_SCORECARD), and design docs (`agents/oscar.md`, `ai-os-architecture/` README+diagram+handoffs.yaml+control-plane). Her responsibilities FOLDED INTO KAT (organizing/filing/notifying/light-technical) per Greg 2026-07-12. RESIDUAL (generated/Mac-side only, non-blocking): `brain/` REGENERATED on the Mac 2026-07-12 (`~/.claude/skills/aria/` deleted, `brain_extract.sh` re-run -> `[brain] agents: 7`, ARIA gone) – now just needs commit+push from the Mac (`save.command`) to land in git; regenerate `board/index.html` (needs Drive+revenue-engine sources, run on the Mac); let memory record `PIVOTNREBOOT_AGENTS_HUNTER_ARIA_V1` be superseded via the append-only memory flow (do not hand-edit). Live calendar/email-triage still unowned – fold into OSCAR if ever needed, do NOT revive ARIA

Greg confirmed solved UI-side 2026-07-10

Procedure is binding in hunter.md + QA-QUINN contract. Never report a binary upload done without the round-trip byte-compare

Reconciled onto main via PR #36 merge; both branches' items now coexist under distinct IDs

CLOSED by merge contract Step 4: `.claude/skills/rotate-api/SKILL.md` built (mint→encrypt→wipe→verify→revoke→register→launchd→QA-QUINN→Drive log; self-limits to plan+handoff on non-Mac surfaces). Install same-version copy in Cowork + web/chat to fully close TD-027

SM-001: Mac session runs `update_drive_file.py 177MPZZ… ` with `health:"ON TRACK"`; boards self-heal ≤6 min. Repo renderers already hardened to strip embedded days

Leave all `agent/lyra*` branches as-is. LYRA agent (`.claude/agents/lyra.md`) is active and untouched. Do NOT propose deleting any lyra ref again