← Portal
Snapshot · 3 engagement(s) tracked · auto-tagged

State · Active Engagements

Updated 2026-05-18

VG Home Buyers

Latest activity: 2026-05-18 · 02_active/vg/
Filter:
next-notes.md · updated 2026-05-18

Scratchpad for potential ideas and uncommitted work items. Not decided work — see next.md for committed next steps.


2026-05-18 — Portal section reorganization complete

Portal structural reorganization verified and locked down. 11 docs relocated to tool-specific sections: actualsboard-guide + budget-source-guide → Reports; 5 intake docs (property-intake-checklist, property-data-lifecycle, intake-variable-registry, estimating-variables-by-wo, plan-template-variables-approval) → Forms; 3 WO cheat sheets → Work Orders; interview-guide + who-approves-what → Workflow. diagrams/ folder renamed to workflow/ with canonical hand-crafted SVG diagrams. Reconciliation guide depersonalized (removed person name). Dropbox aspirational guide retired; new quick-reference guide authored from actual folder structure. All 14 affected docs verified visible on target hubs (100% disk presence + index linkage). 87 outputs tracked; 85 OK, 2 STALE (improvement from baseline 84 OK/3 STALE). Portal sidebar structure unchanged (6 sections). See docs/superpowers/plans/2026-05-18-portal-section-reorg.md.


2026-05-18 — Extract source docs in client-files/ into canonical reference/

Companion follow-up to today's reference-folder cleanup (moved source/outdated docs out of reference/ into client-files/archived/, client-files/people/, client-files/meetings/, etc.).

Several folders inside client-files/ contain raw source material that has useful contextual signal but is not yet canonical reference. The deferred work is to extract that signal into proper reference/ files with an approve/deny workflow so Randall can review each extracted item before it joins canonical context.

Sources to process (in priority order):

  1. client-files/people/detailed/*.md (17 files: andrew, brian, edgardo, edmar, elmie, hector, herson, jao, jenn, jj, jose, kim, mark, michelle, ritchie, shlomo + jenn) — these have richer role/responsibilities/workflow content than reference/people/roster.md. Extract:
    • Role descriptions → expand roster columns or add reference/people/role-detail.md
    • Workflow involvement (which streams, which pipeline stages) → cross-link into reference/operations/workflows/
    • Standard contact info → roster (where appropriate; PII per RWP/CLAUDE.md rules)
    • Strip personal/private details that don't belong in canonical AI context
  2. client-files/people/ raw files — JDs, offer letter, OnlineJobs.ph listing, Ritchie's Roles and Tasks.docx, Daily Role Functions & How-To Steps.gdoc, remote-admin-member-support-specialist-longform.md. Useful for filling out roster role descriptions but should not become reference verbatim.
  3. client-files/meetings/ — selectively pull recurring patterns (process changes, decisions, vocabulary clarifications) into reference/ updates. Do not import transcripts wholesale.
  4. client-files/archived/work-order-system/WorkOrderSystem_MASTER.md, Work Order System Intro.docx, Work Order Numbering Final.csv may have content that should be reconciled against current reference/cost-codes/ files (or confirmed redundant and left archived).
  5. client-files/archived/finance-team-docs/ — 5 Finance_Team_*.docx files. If any contain still-applicable policy, lift into reference/operations/ or reference/accounting/.
  6. client-files/incoming/ — random staging area; sweep for any items that should move to temp/, be processed into reference, or be archived.

Workflow:

  • Read source → propose extracted reference additions as diffs/inserts
  • Present each proposal individually for approve/deny
  • On approve: apply to canonical reference/ file
  • On deny: skip (source stays in client-files/)
  • Track progress so we can pause/resume across sessions

Why deferred: today's cleanup was structural (moves only). Extraction is content work that benefits from focused sessions with Randall's review attention, not a rushed batch pass.


2026-05-18 — Team-driven budget upgrade flow

After the budget source-tagging system landed (see docs/superpowers/plans/2026-05-18-budget-source-tagging.md), team members still need to edit YAML files to upgrade a placeholder to a real estimate. That works for Randall but won't scale to delegated team members.

Goal: make it easy for a team member to upgrade a low-confidence budget line directly from the dashboard, without touching files.

Possible approaches (not yet decided):

  • Click-to-edit on the actualsboard — clicking a or * marked budget number opens a small form that lets the user enter a new amount + source. Submits to a backend that writes the YAML. Requires: hosted endpoint (Netlify function? n8n webhook? local FastAPI?), some form of auth, write-back to budgets.yaml + automatic regen.
  • Google Sheet front end — a per-property sheet where each WO line shows current budget + source. Team edits in Sheets; a sync job writes YAML. Reuses the Google Sheets infra already in place for pm-workbook.
  • Simple intake form per property — a one-page form (like the existing capture forms) for "I have updated estimates for property X" — team fills it out, submits, agent processes and writes YAML.
  • Slack command/vg-budget 32-clinton P1WO5 41200 detailed_estimate "ABC Roofing bid" — captures the line directly into YAML.

Tradeoffs to think about:

  • Write authority — who can upgrade what? (e.g. only buyer should set buyer_initial; only Randall sets committed)
  • Auditability — should every upgrade be logged with who/when?
  • Conflict handling — what if two people upgrade the same line concurrently? (low risk given team size, but worth noting)
  • Regen cycle — does the dashboard auto-refresh after an upgrade, or does Randall manually rerun?

Worth a brainstorming session before designing. Save for after the current active-property placeholder-fill walk (which is the next manual task that will reveal which lines really need easy upgrade paths).


2026-05-08 — Actualsboard follow-up

  • Map remaining unmapped QB items — check the Unmapped Items exception queue in the actualsboard report; add missing QB Product/Service names → WO codes to workspaces/actualsboard/config/cost_code_mapping.yaml
  • Fill missing budgets — 32 Clinton and 519 Highland have no budget entries; review all active-construction properties in workspaces/actualsboard/config/budgets.yaml for any missing WO budget lines; pull numbers from approved WO packets
  • Redesign the report — ✅ Done 2026-05-13: WO-matrix layout shipped, split into construction + non-construction sibling boards. See docs/superpowers/specs/2026-05-13-actualsboard-redesign-design.md.

Randall Harness

Latest activity: 2026-05-11 · 02_active/randall/
Filter:
next-notes.md · updated 2026-05-11

Scratchpad for potential ideas and uncommitted work items. Not decided work — see next.md for committed next steps.


2026-05-11 — agentic-stack dream-cycle sandbox: installed and torn down same day

Installed the .agent/ dream-cycle layer here, then removed it the same session after realizing two things:

  1. Wrong location. I (Randall) don't actively work in 02_active/randall/ right now — current daily work splits ~60% VG, ~30% RWP root, ~10% BC/personal. A sandbox here would never see meaningful tool-call volume.

  2. Wrong tool for the workflow. The dream cycle's PostToolUse hook writes tool inputs and outputs to a local JSONL — fine in your-own-code repos, but a privacy issue in client folders (VG, BC) whose contents include buyer/seller details, financials, and meeting-transcript snippets. The pii_guard.py PreToolUse hook blocks PII writes, not telemetry logging. Putting the sandbox in any client folder would funnel that data into a Drive-synced JSONL.

  3. Mismatched signal shape. agentic-stack clusters by tool-call categories tuned for software-engineering work (deploy / migrate / schema). The actual high-value pattern in this workspace is markdown-evolution: repeated edits to reference docs (CLAUDE.md, task-vocabulary.md, client workflow files) that suggest rules ready to be promoted upward. The dream-cycle clustering doesn't surface that pattern.

Outcome: see 01_inbox/markdown-evolution-tracker-design.md for the purpose-built alternative (no tool-call logging, watches reference markdown files only, weekly review surfaced into a digest file). No build commitment yet — it's a design Randall can decide on later.

Net change to this repo: none. Sandbox files removed; .gitignore restored.

Personal OS

Latest activity: 2026-04-28 · 02_active/personalos/
Filter:
next-notes.md · updated 2026-04-28

Scratchpad for potential ideas and uncommitted work items. Not decided work — see next.md for committed next steps.