From clayworks-lite
Pick the right memory layer when saving a fact — Engram (cross-model structured/procedural), native Claude Code MEMORY.md (project-scoped), or Honcho (user modeling). Use when the user says "remember/save/store this", states a preference or identity fact, describes a project convention or build command, or when Claude is about to write to memory and the destination isn't obvious. This skill is a routing decision tree, not a memory implementation — the underlying plugins handle storage.
How this skill is triggered — by the user, by Claude, or both
Slash command
/clayworks-lite:clayworks-lite-memory-routingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You have up to three memory layers. Each is good at different things. Picking the right one isn't a hunch — it's a decision that follows from what the fact IS.
You have up to three memory layers. Each is good at different things. Picking the right one isn't a hunch — it's a decision that follows from what the fact IS.
The point is the decision, not the destination. Even if you only have one layer installed, deciding deliberately keeps that layer's contents coherent.
Install (free, runs locally):
/plugin marketplace add Gentleman-Programming/engram
/plugin install engram@engram
Use for:
Key trait: topic-keyed, searchable, cross-model accessible. Survives compaction and session boundaries. Save with mem_save; retrieve with mem_search.
MEMORY.md)Install: built into Claude Code — no plugin needed.
Path: ~/.claude/projects/<project-name>/memory/MEMORY.md
Use for:
Key trait: auto-loaded into context on session start when CC is invoked from inside that project directory. Free. No plugin needed. Plain markdown — edit directly with your favorite editor.
Install (cloud-hosted; $100 free credits then ~$2–3/mo at typical usage):
/plugin marketplace add plastic-labs/claude-honcho
/plugin install honcho@honcho
Use for:
Key trait: dialectical model that builds a persistent representation of the user across sessions. Auto-loaded as a profile at session start. Save with create_conclusion.
Alternative: Honcho is one user-modeling layer; not the only one. If you use a different system (a hand-maintained USER.md, a separate vector store, a custom service), the routing principles still apply — Honcho's slot in the decision tree is just "your user-modeling layer of choice."
Ask these questions in order. Stop at the first YES.
1. Is this a fact about WHO the user is — their role, working style, preferences, or behavioral patterns?
→ Honcho. (If Honcho isn't installed, fall through to Engram and tag with user-profile/....)
2. Is this fact specific to one project — a build command, file convention, or recent decision affecting only that codebase?
→ Native MEMORY.md in that project's memory directory.
3. Is this a cross-project decision, a tool config, a procedure, a bug-fix root cause, or something that should be searchable from multiple AI tools? → Engram.
4. None of the above clearly fit? → Engram. Broadest retrieval, lowest risk of misfiling. Add a clear topic key so it stays discoverable.
| Fact | Layer | Why |
|---|---|---|
| "User is a backend engineer at a fintech startup" | Honcho | WHO they are |
| "User prefers terse responses, no closing pleasantries" | Honcho | Working style |
| "User won't approve destructive operations without a written backup confirmation" | Honcho | Behavioral guardrail |
"Project X builds with ./gradlew :app:assembleDebug using JDK 21" | MEMORY.md (Project X) | Project-specific build command |
"Project X uses applicationId com.foo.app.debug for debug builds" | MEMORY.md (Project X) | Project-specific config |
| "Use the Bash tool (not PowerShell) for cross-platform scripts" | Engram | Tool convention across projects |
| "Bug: API X's bulk endpoint silently drops fields > 1KB" | Engram | Reusable discovery, cross-project |
| "Pattern: backup-then-install + SHA-256 hash-diff for idempotent installers" | Engram | Cross-project architecture pattern |
| "DocuSeal free tier doesn't support conditional logic" | Engram | Cross-project tool limitation |
| "User is allergic to peanuts" | Honcho | Identity fact, persistent across sessions |
| "Standup is at 09:30 daily" | Probably neither — this is a recurring event; use a calendar or the Nudge skill instead | Memory is for facts, not schedules |
The skill works degraded. Use what's available:
user-profile/role, user-profile/style, project-x/build, decisions/architecture/idempotent-installers.MEMORY.md only. Save everything in the active project's MEMORY.md. Add a ## User Profile section near the top for WHO-the-user-is facts so they load with the file.mem_save syntax, create_conclusion parameters, and native MEMORY.md formatting are documented in their respective sources. Read those when you're actually saving.Trigger conditions:
mem_save, write to MEMORY.md, or call create_conclusion and the destination isn't obviousWhen in doubt: invoke and run through the decision tree once. The 30 seconds it costs now beats the friction of finding a fact in the wrong layer later — or worse, saving it three times across all three layers and then maintaining the divergence.
The three layers drift apart over time if you don't actively maintain them:
Quarterly: skim each layer. Re-route entries that are in the wrong place. Delete entries that no longer reflect reality. The decision tree is most valuable on day one of a new fact — but periodically re-running the decision on existing facts catches the drift.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
Applies a firm's KYC/AML rules grid to parsed onboarding records: assigns risk rating, checks required documents, outputs rule outcomes with citations, and routes for escalation.
Generates daily or weekly digests of activity from connected sources (chat, email, docs, tasks, CRM), highlighting action items, decisions, mentions, and project updates.
npx claudepluginhub clayboicardi/clayworks-lite --plugin clayworks-lite