UPGRADES — Deferred Items

Cross-business deferred items. Add here when making an SSOT decision to defer something.

AI Department

JOB_DESCRIPTION.md updates (all depts) — AI, Processes, IT, AI/Tools dept scope defined ✅ (AI/Tools confirmed 2026-07-09). Remaining stub (Scope: To be defined): Cancer50 (dormant mission anchor — activate on first real need). 2026-06-15.
Vault-wide /rocks-refresh skill — generalize /filter-focus-refresh to refresh the Core/DASHBOARD Biggest Rocks roll-up. Deferred 2026-07-09: with ~5 curated rows sourced from already-curated business lists, manual editing wins until volume justifies a skill.

Utilities

pdf2md: add config.yaml to package data — pyproject.toml only packages src/pdf2md; config.yaml excluded from wheel. After uv tool install, __file__-relative path resolves to site-packages and misses config. Fix: add [tool.hatch.build.targets.wheel] include = ["config.yaml"] or use TOOL_HOME env pattern. Workaround active: manual copy to installed tool lib dir (fragile, breaks on reinstall).

Vault Structure

~~Address MBR\Clippings\~~ — resolved: all 90 Business\Clippings files → Core\Misc\Clippings\ (2026-06-09)
FSS\ strategy content — TalbotStevens.com specific content to be carved out (Phase 2)

Post-Phase-7 Remaining Tasks (Business-OS-Architecture follow-ons)

SMTM_System.md update — Core\Processes\Simple Markdown Task Management\SMTM_System.md still references old Business vault structure and Dynalist. Needs rewrite for unified vault. (~~Ideas Workflow.md supersession~~ — resolved 2026-07-09: unique content migrated into Core\Processes\Capture-Routing.md, old file is now a pointer.)
SDC Core Value Propositions — SDC\Strategy\Core Value Propositions.md is an empty placeholder (> [!todo] complete this). Needs content when SDC activates.
SSoT review: MBR vs SDC Threats — MBR\Strategy\Identity\Internal View\Threats.md and SDC\Strategy\Identity\Internal View\Threats.md have parallel structure. Review for shared content that belongs in Core\Strategy\ vs legitimate per-biz differences.
~~Graphify high-quality re-run~~ ✅ done 2026-07-13 — whole-vault, --mode deep, 2,103 nodes / 2,671 edges / 329 communities. See KB-OS/logs/2026-07-13_Graphify-Fullvault-Rerun.md.
FSS tasks in Core — Core\_WorkingOn\Tasks\ still holds FSS-specific files (ecommerce.md, ts.com-migration.md, ts-site-upgrade.md). Move to FSS\_WorkingOn\Tasks\ when FSS work resumes.

Autonomy Levels (A2/A3/A4)

A2 — Autonomous within budget (activate when budget tracking in place)
A3 — Autonomous within scope
A4 — Autonomous within mission

Dept Templates (activate on first need)

Specs\ per dept — activate when first spec is written
UPGRADES.md per dept — activate when first deferred item logged
Questions.md per dept — activate when first open question logged
Sub-departments beyond pre-active set

Frontmatter Fields (TaskNote)

decay-after: — decay curve timer; pair with “decaying soon” Bases view (tasks nearing expiry surfaced in dashboard before they vanish)
priority: — explicit priority field
scope: — scope tagging
agents: — named agent assignment
parent_goal: — OKR linkage / goal ancestry chain
next_wake: — heartbeat scheduling (scheduled resume timestamp)
claimed_at: — Paperclip timestamp (pair with claimed_by:)
budget_cap: — per-agent budget cap; companion Budget-YYYY-MM.md tracks actuals against cap
delegates_to:, cadence:, memory_root:, budget_root: — Stage 3+ fields

Skills (deferred)

/triage-inbox — for unrouted residue when Inbox grows unmanageable; scans all active dept Inboxes and routes items to correct Tasks/, Projects/, or archive
/program-review — weekly cross-project review ritual (extracted 2026-07-09 from archived myMBR-OS task, where it was designed pre-KB-OS): read all active projects’ STATUS.md, check alignment against Filter+Focus / Biggest Rocks, update roadmap views, surface blocked/ready-to-start work, suggest the single next highest-leverage rock. ~10–15 min; replaces “I wonder what’s next” with a structured scan. Largely overlaps today’s dashboards — evaluate whether a skill still adds value or the Bases views suffice.

SMTM Enhancements

Spec → Test → Impl → Audit loop — formalize as an SMTM task sub-type for code/build work; prevents “done” declarations before evidence is captured
Knowledge freshness markers — per-file reviewed: date + “stale after N days” signal; surface in Bases or vault health check
CHANGELOG.md per dept — each active dept gets a CHANGELOG.md updated by /task-complete; pairs with LESSONS.md (already per-project)

Paperclip Patterns (Phase 2+)

A2 autonomous-within-budget checkout
Cross-dept task borrowing (beyond atomic checkout)
Paperclip pilot at 60-day mark — run one non-critical workload fully autonomous to stress-test agent handoff, budget caps, and heartbeat scheduling before broader rollout
Reference org-chart deep dive — explore paperclipai/companies for patterns applicable to KB dept structure and agent role definitions

Graphify (Phase 2+)

graphify --watch background daemon
Daily/on-commit graph refresh
Agent convention: query graph before grep (document in Core\AI\JOB_DESCRIPTION.md)
Bulk legacy ingestion — 30+ years of artifacts (PDFs, PPTs, older docs); batch through pdf2md → graphify pipeline when vault content stabilizes

CodeGraph — code-dev context (CLI installed & verified 2026-07-13 ✅)

CLI installed globally (v1.4.1, ~/.local/bin/codegraph, telemetry off) and verified end-to-end on packages/sd-math (34 Py files → 513 nodes/1,261 edges in 0.4 s; codegraph explore returned the exact payment helper + call paths + untested-symbol flags). Full write-up: KB-OS-Context §5.
CLI>MCP tension resolved: CodeGraph ships a real codegraph explore CLI with the same output as its MCP tool — agents use it via Bash, no MCP server needed. .codegraph/ per-project index is gitignored (per-device disposable cache, Principle #8).
Division of labour: graphify = default for everything (docs, strategy — “how does it connect”); CodeGraph = specialist for code-dev projects (monorepo, rate-scanner — “what is this code / what breaks if I change it”).
Remaining decision (Talbot): whether to also enable the MCP integration for auto-sync + auto-approved tool use in Claude Code’s main agent. If yes, wire it through ai-config (not a raw codegraph install, which edits deployed CLAUDE.md/settings directly — SSOT violation). Owner: IT/AI dept.
Per-project usage: cd <repo> && codegraph init once, then codegraph explore "<question>".

Memory & Context layer (KB-OS ROADMAP M3.6)

Active tasks promoted 2026-07-13: Memory Phase 3 SessionStart hook + graphify full-vault re-run (see KB-OS/Tasks/). Still deferred below. Detail/rationale: KB-OS-Context §5.

Hybrid-query speed — investigate qmd flags to skip query expansion + GPU offload in WSL, to make qmd query usable interactively (currently minutes on CPU). Low urgency: BM25 qmd search covers daily recall.
Transcript bootstrap import — one-time, sentinel-guarded pull of historical Claude Code session transcripts into the sessions archive. Deliberately deferred — the vault is already a curated history, so marginal value is low.
CodeGraph MCP integration — not done by design (CLI codegraph explore suffices, matches CLI>MCP). Only if a future need justifies seamless main-agent auto-use; if so, wire through ai-config (not a raw codegraph install, which edits deployed config directly).

Inter-Department Patterns (Phase 2+)

Inter-dept mailbox — cross-dept request queue pattern; dept A drops a request, dept B agent picks it up on next heartbeat; avoids direct coupling between agent loops

Briefings (Phase 2+)

Daily briefing cadence
Monthly briefing cadence
Quarterly briefing cadence
Yearly briefing cadence

MBR-specific deferred

Risk-register Bases dashboard
Strategic-OKR Bases dashboard
VP-Strategy staff file (when delegation volume justifies)
Compliance-Auditor as separate staff file
Quarterly external audit cadence
MBR\Customer\ dept — customer support escalations

SDC + FSS

SDC dept activation on first task arrival
FSS dept activation on first task arrival
Personal\ Life-OS slot (reserved)

Per our KB-OS evolution to department segmentation …
- /task-complete should put log into appropriate dept log folder, and update CHANGELOG.md and LESSONS.md
- migrate \Core\Logs into ..<dept>\Logs

Confirm these are done … (from bottom of D:\FSS\KB\Core_WorkingOn\Tasks\Business-OS-Architecture-Content-Drafts.md, purged)

Drafts NOT in this file (for later Sonnet cycle from template)

The following are uniform/templatic — Sonnet handles efficiently from dept-template:

MBR\Offerings\JOB_DESCRIPTION.md
MBR\Mktg\JOB_DESCRIPTION.md (+ sub-depts WealthCare50, Content, Viral-Growth)
MBR\IT\JOB_DESCRIPTION.md (+ sub-depts UX, Data, Processes, Apps, Sites)
Core\Cancer50\JOB_DESCRIPTION.md
Core\AI\JOB_DESCRIPTION.md (+ Tools sub-dept)
Core\Processes\JOB_DESCRIPTION.md
MBR\Strategy\Staff\CEO.md (uses Staff template; per-position content brief)
MBR\Strategy\Briefings\BRIEFING-Weekly.md (Phase-1 launch state + Q2 OKRs — Talbot may want to author personally)
MBR\IT\UX\JOB_DESCRIPTION.md — links F.A.S.T. Glossary; can use template
Core\README.md — vault navigation guide; mechanical from layout diagram
Core\UPGRADES.md — inventory already in plan v6; mechanical copy