Guide · Agent skill
How to use the Agent Memory Discipline agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Agent Memory Discipline is a skill, a set of written instructions that an AI agent loads, which tells the agent when to read its long-term memory before acting and when to save decisions, corrections and failures afterward. It is for people whose agent has a memory tool connected but still forgets preferences or repeats old mistakes. After this guide you will know what the skill changes, how to install it, how to trigger it with ordinary requests, and what it needs to work.
Agent Memory Disciplinesickn33/agentic-awesome-skills · MIT Download the skill Quick start
- 1Pick a memory backend. Choose a local Markdown folder, a local MCP memory server or a hosted memory service. The skill needs a backend the agent can call.
- 2Download the ZIP. Download the skill ZIP from this page.
- 3Unzip into a skills folder. Unzip it into ~/.claude/skills/ for all projects, or into .claude/skills/ in one project.
- 4Or upload in the Claude apps. In the Claude apps, go to Settings, then Capabilities, then Skills, and upload the skill.
- 5Make a request. Ask the agent to recall before project work, or say remember this when you want a fact saved.
- 6Correct and review. When the agent is wrong, correct it directly and check that the saved note is one fact with its source.
What the skill makes the agent do
Connecting a memory tool does not make an agent use it well. The skill adds standing rules. Before starting project work, choosing a library or tool, writing tests or commits, or answering questions like how the team usually does something, the agent reads relevant memory first.
After a decision, a user correction or a failure, the agent saves a short note with one fact in it. When a memory changes, the old note is marked as superseded with an end date and a pointer to its replacement, not deleted. When two memories conflict, the agent shows both to you instead of quietly picking one.
- Recall before acting, using the words you used plus the project or repository name.
- Save one fact per entry, including what, why, when and where it came from.
- Skip recall for one-off factual questions, arithmetic or fully specified requests.
- Never save secrets, tokens, personal data, transient state or file contents that can be read again.
Choose a memory backend
The skill needs a memory tool the agent can call. It does not supply one. The source lists three kinds of backend that work with the same rules.
Authentication depends on the backend. A local folder needs none, a local MCP memory server uses its own configuration, and a hosted service uses an API key or OAuth sign-in. The skill says it never handles credentials itself and never writes them into memory.
- A memory/ folder of Markdown notes, one fact per file. No dependencies, searchable with grep and versionable in git.
- A local MCP memory server that keeps everything on your machine.
- A hosted memory service over MCP, which makes memory portable across tools and machines but stores the data elsewhere.
Install the skill
Download the ZIP from this page. You can install it in one of two ways, depending on how you use your agent.
For Claude Code, unzip the download into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to limit it to that project. In the Claude apps, upload the skill under Settings, then Capabilities, then Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings, then Capabilities, then Skills, then upload
Trigger the skill with natural requests
You do not need special syntax. The skill's description tells the agent when to use it, and the source lists phrases that should trigger recall or save. Saying remember this, what did we decide, recall, or save this for next time all point the agent toward memory.
Recall also triggers on phrases such as again, like last time or as we agreed. When you are correcting a habit, state the correction directly, because corrections are the strongest signal the skill saves.
- remember this
- what did we decide about the database?
- recall how we usually write tests here
- save this for next time
- stop using npm here, we are on pnpm
See what a saved memory looks like
Each memory is one fact in a short sentence that also says why, when and where it came from. The source example is a correction about a package manager, written as a single line.
When a fact stops being true, the skill keeps the old line, adds an end date and a pointer to the replacement, and writes the new fact next to it. The full entry format is in references/entry-format.md inside the skill.
- What: the decision or observation in one sentence.
- Why: the reason, briefly.
- When: the date it became true, and when it stopped if it has.
- Source: a file, commit, conversation or test run.
- Scope: where it applies.
Project uses pnpm, not npm. Stated by the user after a lockfile conflict. Applies to all packages in this repo.Safety notes: scripts and network
According to the listing, this skill runs no scripts and uses no network access of its own. It is a set of Markdown instructions. Your memory backend may still use the network, for example a hosted memory service, so check that backend's own setup.
The skill's metadata marks its risk as critical, which reflects that it writes persistent notes. Keep sensitive data out of memory, and review what the agent saves, especially early on.
- Runs scripts: no
- Uses network: no (the skill itself; a hosted backend will)
- Credentials are handled by your backend, not by the skill
- Review saved entries, and delete ones you do not want kept when you ask for deletion
What people use it for
- Stopping an assistant from reverting a coding convention you have already corrected several times.
- Keeping track of architecture decisions, such as a state management library, across weeks of work.
- Making sure the agent checks for a known environment quirk, such as a port or a service that must be running, before running tests.
- Flagging conflicting team rules to a person instead of letting the agent pick one silently.
Frequently asked questions
How do I install the Agent Memory Discipline skill?
Download the ZIP from this page 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, then Capabilities, then Skills.
Does the Agent Memory Discipline skill need a memory server?
Yes, it needs a memory backend the agent can call. The skill does not provide one. A local Markdown folder, a local MCP memory server or a hosted memory service will all work.
Is the Agent Memory Discipline skill free?
The listing says MIT licence, but the SKILL.md metadata names CC0-1.0 with a link to the LICENSE file. Check that file on the source repository to confirm the terms before you rely on either one.
Does the skill run scripts or use the internet?
According to the listing, it runs no scripts and uses no network itself. A hosted memory backend you choose would use the network, so review that service's setup separately.
How is this different from just connecting a memory tool?
A connected tool only makes memory available. This skill tells the agent when to read it before acting and when to save decisions, corrections and failures afterward, and how to handle conflicting entries.
Why isn't my agent using memory even though it is installed?
The skill has to be loaded and the memory tool must be connected and callable. Use phrases like remember this or recall, and check that the backend itself is running and authenticated. The skill does not supply storage or credentials.
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
Use this skill if you run an agent with a memory backend and keep repeating the same corrections or losing past decisions. Pick something else if you only ask one-off questions, want a ready-made memory product, or do not want the agent to store anything after decisions and corrections.
Download the skill