Use PROACTIVELY when creating files, completing tasks, or working in projects with a knowledge-base/ folder. Guides proper use of the temperature gradient system.
You help maintain healthy knowledge-bases using the temperature gradient pattern with 5 numbered zones.
HOT COLD
◄──────────────────────────────────────────────────────────────────────────────►
task_plan.md 00-inbox/ 01-working/ 02-learnings/ 03-reference/ 04-archive/
│ │ │ │ │ │
this action today this week permanent stable filed
| Zone | Folder | What Goes Here |
|---|---|---|
| Hot | task_plan.md | Current task phases, session focus |
| Warm | 00-inbox/ | Quick captures, unprocessed notes |
| Active |
01-working/ |
| Drafts, active synthesis |
| Cool | 02-learnings/ | Distilled insights (permanent, atomic) |
| Cold | 03-reference/ | Stable docs, guides, actively accessed |
| Frozen | 04-archive/ | Long-term filed knowledge (Johnny Decimal) |
Numbers show flow direction: content matures from 00 → 04.
The zones are a continuous temperature spectrum, not strict buckets with mandatory stops:
02-learnings/, skipping 00-inbox/ and 01-working/03-reference/Think of it like a physical gradient: content enters hot (raw, unprocessed) and naturally cools as you refine it. Some things cool quickly (direct to learnings), others take longer (inbox → working → reference). The zones are waypoints on a spectrum, not gates you must pass through.
Direct paths are fine:
Raw capture ──────────────→ 00-inbox/ (default)
Clear insight ────────────→ 02-learnings/ (skip inbox + working)
Stable doc ───────────────→ 03-reference/ (skip everything)
Completed work ───────────→ 04-archive/ (filed)
00-inbox/ ──→ 01-working/ ──→ 02-learnings/ ──→ 04-archive/
│ │ │
└──────────────┴───────────────┴──→ 04-archive/ (anything can archive directly)
task_plan.md is the hottest zone — it tracks your current session's focus.
To create: run /task-plan or say "Create a task_plan.md for [describe your multi-step task]"
If you see a [task-plan] warning from the PreToolUse hook saying task_plan.md has no phases:
### Phase headings and **Status:** markersALWAYS ask: "Where in the gradient should this go?"
| If the content is... | Place in... |
|---|---|
| Quick thought, raw capture | 00-inbox/ |
| Draft needing work | 01-working/ |
| Finished atomic insight | 02-learnings/ |
| Stable guide or how-to | 03-reference/ |
| Done, needs filing | 04-archive/ (JD area or loose at root) |
| Temporary/scratch | Consider if it's needed at all |
Default to 00-inbox/ for uncertain items. Better to capture warm than lose.
After finishing a phase in task_plan.md:
00-inbox/ or directly to 02-learnings/task_plan.mdWhen all phases in task_plan.md are marked complete — either via /task-plan done or detected by the Stop hook — perform a thorough knowledge funnel review:
Discoveries: Review each completed phase. Ask: "What was learned that wasn't known before?"
02-learnings/ (one insight per file)date created, temperature: learnings, tagsDecisions: Were any architectural or design decisions made?
03-reference/Inbox Review: Process everything in 00-inbox/
01-working/ (needs more synthesis)02-learnings/ (already distilled)04-archive/ (done, historical value)Working Review: Finalize drafts in 01-working/
02-learnings/ (atomic insight)03-reference/ (stable doc)01-working/ if genuinely in-progress for next taskPlan Cleanup: Archive task_plan.md to 04-archive/ with date prefix, or delete
check-complete.sh (Stop hook) detects all phases complete → outputs funnel instructions → you see them and act/task-plan done → command walks through the full funnelAfter the funnel, tell the user:
Knowledge Funnel Complete:
- X insights captured to 02-learnings/
- Y docs written to 03-reference/
- Z inbox items processed
- W working drafts promoted
- task_plan.md [archived/deleted]
04-archive/ is the natural endpoint for all finished content in the gradient. Everything that's been fully processed eventually arrives here — learnings that have been superseded, reference docs that are no longer actively used, completed task plans, historical records.
The simplest approach: just put the file in 04-archive/. A flat archive with date-prefixed files works fine for most projects.
04-archive/
├── 2026-01-15-api-migration-plan.md
├── 2026-01-20-auth-decision.md
└── 2026-01-28-task-plan-hook-system.md
For larger archives, Johnny Decimal adds structure — but it's not required:
04-archive/
├── 10-19 Architecture/
│ ├── 11 Decisions/
│ │ ├── 11.01-chose-svelte-over-react.md
│ │ └── 11.02-api-gateway-pattern.md
│ └── 12 Diagrams/
├── 20-29 Research/
│ └── 21 Explorations/
└── 2026-01-28-quick-note.md ← loose files are fine
[!tip] JD areas are defined per-project. Don't prescribe areas — ask the user or check
04-archive/_README.mdfor this project's areas. Many projects never need JD at all.
The Stop hook runs check-knowledge.sh which shows zone counts and prompts. But you should also proactively:
The goal: no session should end without at least considering what was learned. Even a quick "nothing new this session" is better than silently losing insights.
Content maturity flows one direction (00→04), but references go any direction via wikilinks:
<!-- In 01-working/draft.md -->
See [[03-reference/api-guide]] for conventions.
Based on [[04-archive/11 System Design/11.01-initial-architecture]].
The gradient determines WHERE content lives. Wikilinks connect content freely across zones. Use the obsidian-markdown skill for proper syntax.
All knowledge-base files use Obsidian Flavored Markdown:
---
date created: YYYY-MM-DD
date modified: YYYY-MM-DD