Guide · Agent skill
How to use the Agent Carnet agent skill for persistent notes
Updated · written with AI from the project's own documentation and checked against our catalogue data
Agent Carnet is an agent skill that lets a coding agent keep a shared markdown notebook inside your project. It is for developers who work with coding agents across many sessions and want findings, gotchas and decisions to carry over instead of being rediscovered. After this guide you will know what an agent skill is, how to install this one, how to trigger it with plain requests, and which safety and expiry rules to check before relying on it.
Agent Carnet Notebookyamadashy/repomix · MIT Download the skill Quick start
- 1Download the skill ZIP. Get the agent-carnet skill as a ZIP from the page where you found it.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Confirm the CLI is available. Make sure the agent-carnet command is installed, since the skill calls it for every note operation.
- 4Ask the agent to save a note. Say 'remember this' or 'save this' after a finding, and the agent will write a markdown note under .carnet/ with a summary.
- 5Recall notes in a later session. Ask 'what did we discuss about...' or 'find in carnet' to search summaries and reopen the relevant note.
What an agent skill is and what Agent Carnet makes the agent do
An agent skill is a folder of instructions and supporting files that a coding agent loads when a task matches the skill's description. Agent Carnet's SKILL.md tells the agent to save, recall, find and organize notes using a small command-line tool called agent-carnet.
Notes are stored as markdown files under .carnet/<category>/<slug>.md in your project. The agent saves findings, gotchas and decisions with a short summary, then searches and reopens them in later sessions instead of working them out again.
- Save research findings that took effort to derive
- Record non-obvious patterns or gotchas in the codebase
- Keep solutions to tricky problems and architectural decisions with their reasoning
- Note in-progress work that may be resumed later
How to install the Agent Carnet skill
Download the skill as a ZIP from the page you found it on, then unzip it into a skills folder. For a personal install use ~/.claude/skills/. To limit it to one repository, unzip it into that project's .claude/skills/ folder instead.
If you use the Claude apps rather than a local setup, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Single project install: unzip into .claude/skills/ inside the repository
- Claude apps: Settings → Capabilities → Skills, then upload
What you need before the skill works
The skill depends on the agent-carnet CLI, which the agent calls for every save, find, list, show, used, move and rm operation. The source files for this skill do not say where to install the CLI from, so check the CLI's own documentation before you rely on the skill.
The skill runs scripts, and it does not use the network.
- agent-carnet CLI available to the agent
- Runs scripts: yes
- Uses network: no
How to trigger Agent Carnet with natural requests
The skill activates when you ask to save, recall, find or organize notes. Phrases that trigger it include 'remember this', 'save this', 'note this', 'what did we discuss about...', 'check the notebook' and 'find in carnet'.
The skill also tells the agent to save proactively when it discovers something worth keeping across sessions, so you do not always need to ask. Before a related task, the agent can search for existing notes first.
- 'Remember this: the staging adapter is the thin proxy in front of POST /v1/stage'
- 'Save this fix for the iconv issue so we do not rediscover it'
- 'What did we discuss about the auth middleware?'
- 'Check the notebook before you retry the database migration'
- 'Find in carnet: esm'
How notes are saved, found and expired
Every save requires a decisive --summary and --agent claude-code. A reader should be able to tell from the summary alone whether to open the full note. Notes are grouped in folders by category, and subcategories such as deps/esm/iconv-issue are allowed.
Notes have a 30-day default lifespan. Reading a note with show or marking it applied with used resets that clock, while find does not. Stale notes are moved to .carnet/.trash/ automatically, and they are hard-deleted 7 days later. Setting keep: true pins a note as permanent.
- find searches summaries and does not reset the lifespan
- show reads the full note and resets last_used
- used marks a note as actually applied and increments use_count
- Auto-prune runs on every CLI invocation
- Deleted notes stay in .carnet/.trash/ for 7 days before hard delete
agent-carnet find iconv
agent-carnet list --sort use_count
agent-carnet show deps/iconv-issue
agent-carnet used deps/iconv-issueSafety notes before you rely on it
Notes are plain files on disk. Because the skill tells the agent to save proactively, a note could capture secrets or private data. Review what gets stored, especially on sensitive projects.
The skill's own documentation says the agent should run agent-carnet <command> -h for a subcommand's full flags. Use that help output rather than guessing flags.
- Check .carnet/ contents before committing or sharing the repository
- Do not treat notes as a long-term knowledge base, since unused notes expire
Tips for getting more from Agent Carnet
Keep summaries specific, because they are what future sessions scan first. Use used only when a note actually changed the outcome, so the use_count stays meaningful. Use tags such as vocab for shared project terminology or hypothesis for debugging dead-ends, following the cookbook patterns in the skill's references.
- Write summaries that name the problem and the answer
- Call used only after a note shaped your work
- Use list --sort use_count to surface the most useful notes
- Use move to reorganize categories instead of re-saving notes
What people use it for
- Keeping debugging findings, such as a dependency version that breaks types, so the next session does not repeat the search
- Recording architectural decisions and the reasoning behind them for a repository
- Building a shared record of project terminology so agents use the same names for the same module
- Tracking in-progress work that may be resumed in a later session
Frequently asked questions
How do I install the Agent Carnet skill?
Download the skill ZIP from the page where you found it and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
What does the agent-carnet skill require?
It requires the agent-carnet CLI, which the skill calls for save, find, list, show, used, move and rm. The source files do not say where to get the CLI, so check its own documentation. The skill runs scripts and does not use the network.
Is Agent Carnet free to use?
The source material for this skill is released under the MIT licence. It does not mention prices or paid tiers, so check the licence file in the repository for the terms.
Why did my Agent Carnet notes disappear?
Notes expire after a 30-day default lifespan unless they are read with show or applied with used. Expired notes move to .carnet/.trash/ and are hard-deleted 7 days later. Setting keep: true pins a note as permanent.
Does Agent Carnet send my notes anywhere?
The skill is documented as not using the network. Notes are plain markdown files stored on disk in the .carnet folder inside your project.
How is Agent Carnet different from a normal knowledge base?
It is a small plain-text notebook kept inside a repository and driven by the agent, not a long-term knowledge base. Notes auto-expire unless they are kept or used, so it suits working memory across sessions better than permanent documentation.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Agent Carnet suits developers and teams who work with coding agents across many sessions and want a plain-text record of findings inside the repository. Pick something else if you need a long-term knowledge base, since notes expire after 30 days unless they are used, or if your project could expose secrets in saved notes that you cannot review.
Download the skill