From research-vault
Two-tier memory for the research vault. The vault `CLAUDE.md` is the hot cache (~30 most-used people, terms, projects). The `memory/` directory holds full storage — `glossary.md`, `people/<name>.md`, `projects/<name>.md`, `context/`. Use this skill whenever decoding shorthand, recording a new person/project/term, or deciding whether something belongs in the hot cache vs. deep storage.
How this skill is triggered — by the user, by Claude, or both
Slash command
/research-vault:vault-memoryThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Two-tier memory makes Claude understand the user's shorthand without a re-brief
Two-tier memory makes Claude understand the user's shorthand without a re-brief
every session. The hot cache lives in the vault's CLAUDE.md. Deep storage
lives in memory/.
<vault>/
├── CLAUDE.md ← Hot cache (~50-80 lines): top people, common
│ terms, active projects, preferences
└── memory/
├── glossary.md ← Full decoder ring (everything)
├── people/<name>.md ← Per-person profiles
├── projects/<name>.md ← Per-project context (separate from the task
│ file; this is *background*, not tasks)
└── context/ ← Org-wide context, tools, processes
CLAUDE.md goal: cover 90% of daily decoding needs without a directory walk. Keep it under ~100 lines.
memory/ goal: cover 100% of decoding needs, no size limit, searched when the hot cache misses.
User: "ask <handle> about the cal status"
1. Check vault CLAUDE.md hot cache:
- "<handle>" → ✓ resolved to a person in the roster
- "cal" → ? not in hot cache
2. Check memory/glossary.md:
- "cal" → calibration (a data-prep step)
3. If still not found → ask:
- "What's X? I'll remember it."
| Type | Hot cache (vault CLAUDE.md) | Deep storage (memory/) |
|---|---|---|
| Person | Top ~30 frequent contacts | All people: people/<name>.md + glossary nickname row |
| Term/acronym | ~30 most common | All terms: glossary.md |
| Project | Active projects from projects.md | All: projects/<name>.md + glossary codename row |
| Nickname | Listed under top-30 person | All nicknames: glossary.md |
| Preferences | All preferences | — |
| Org-wide context | Quick reference only | context/<org>.md, context/tools.md |
| Historical / stale | Remove | Keep in memory/ |
When the user says "remember this" / "X means Y" / introduces a new collaborator:
Glossary item (acronym, term, codename):
memory/glossary.md under the right table.CLAUDE.md.Person:
memory/people/<handle>.md (use the #handle form
from vault-people skill).CLAUDE.md People table if they're in the top ~30.Project background:
memory/projects/<name>.md for project context
(history, why it matters, key tradeoffs). The task spine stays in
<name>.md at the vault root.Preference:
CLAUDE.md.Promote to hot cache when:
Demote to deep storage only when:
Run a promotion/demotion pass during /update --comprehensive.
memory/people/<handle>.md)# <Full Name>
**Also known as:** <handle>, <alt-names>
**Role:** <role>
**Affiliation:** <institution>
**Email / contact:** <if known>
## Projects with the user
- <project-file.md> — what they do on it
## Communication
- <preferred channel, style notes>
## Context
- <anything useful for execution: who they report to, working hours, etc.>
## Notes
- <personal touches: shared interests, history>
memory/projects/<name>.md)# <Project name>
**Codename / aliases:** <if any>
**Vault file:** <name>.md
**Status:** <one line>
## Background
<Why this project exists, the funding source, the high-level goal>
## History
<Decisions made, paths not taken, prior versions>
## Key constraints
<Deadlines, dependencies, sensitivities>
Run /start on a fresh vault to scaffold the directory and walk a recent
task list (or external sources, if MCP connectors are configured) to seed
the glossary, people, and projects deep storage.
npx claudepluginhub gillyspace27/research-vault-plugin --plugin research-vaultCreates bite-sized, testable implementation plans from specs or requirements, with file structure and task decomposition. Activates before coding multi-step tasks.