ovh/shai/memory
Claude Code and agent skill by ovh/shai
- community
Checking that the SKILL.md file is still in the repository.
Save, organize, and curate persistent memories (MEMORY.md index plus topic files) across sessions. Load before writing detailed notes, topic files, or when asked to remember context that should outlive this session.
Install
npx vskill@latest install ovh/shai/memorySKILL.md
Memory
Shai keeps persistent memory as plain markdown you manage with the standard file tools plus two index tools. Memories are loaded into every future session, so they must stay small, durable, and curated.
Layout
Two scopes exist, each with the same shape:
- The index (
MEMORY.md) is the only thing auto-injected into the prompt.
One bullet per memory: - [YYYY-MM-DD HH:MM:SS] short fact, optionally ending with a pointer like (see testing.md).
- Topic files hold detail. They are never auto-loaded; read them with the
read tool when the index line is relevant.
- Project scope = facts about this repo (conventions, decisions, architecture
notes that are not derivable from the code). Global scope = facts about the user (preferences, working style) that apply everywhere.
What to save
Save when the information would be useful in a future session and cannot be derived from the codebase, git history, or AGENTS.md:
- Corrections the user gives you ("no, we use pnpm here", "never force-push")
- Confirmed preferences (answer style, workflow choices)
- Decisions and their rationale ("chose X over Y because Z")
- External references (issue tracker, dashboards, deploy targets)
- Ongoing work context that is not in git yet
Do not save: anything readable from the code or AGENTS.md, file paths or architecture the repo already shows, one-off debugging fixes, or session task state (that belongs in the todo list).
How to save
- One-line fact →
memory_write(addscopeonly for global facts).
Keep it under ~120 characters. Deduplicate first: if the index already says it, do not write it again.
- Detailed note → create or update a topic file with the
write/edit
tools. Name it after the topic (testing.md, deploy.md, user_preferences.md). Then add or update a single index line pointing at it: memory_write {"content": "Testing conventions and commands (see testing.md)"}.
- Never write multi-paragraph content through
memory_write— it rejects
newlines by design.
Curation
The index is capped (200 lines / 25KB per scope). If memory_write reports the cap, or the injected index says it is truncated, curate before adding more:
- Remove stale or superseded facts with
memory_remove. - Merge near-duplicates:
memory_removethe variants,memory_writeone
consolidated line.
- Move growing detail out of index lines into topic files, leaving a pointer.
- Topic files can be rewritten freely with
edit— keep each focused on one
topic and short enough to read on demand.
Prefer updating an existing memory over stacking a new one that contradicts it.
Examples
User says "we deploy with the cds pipeline, not kubectl":
then write the pipeline names/links into .shai/memory/deploy.md.
User says "stop adding emojis to commit messages" (personal preference):
An index line is now wrong:
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingshai-core/skills/memory/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/ovh/shai/memory)