Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    mind-palace: Persistent Project Memory

    2

    Persistent project memory for Claude Code: carry key decisions, conventions, commands, and open tasks across sessions, /clear, and compaction—kept in a small, editable Markdown file.

    $10

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    mind-palace: Persistent Project Memory

    mind-palace: Persistent Project Memory

    Example session with this skill installed

    Continue the unfinished NO_COLOR work in this repository. Before making changes, consult the mind-palace skill and verify existing memory against the code. Preserve the decision that an explicit color=True takes precedence, and that the NO_COLOR check belongs in should_strip_ansi(). The helper exists but is not wired in yet. Keep the remaining work clear: wire it in, add tests in tests/test_utils/, and update the changelog. Save only durable facts and remaining tasks in .claude/memory/context.md; merge with existing notes, keep it under about 150 lines, and never store secrets or append a session diary.

    • Read your context and instructions
    • Compiled the mind-palace

    Memory updated: .claude/memory/context.md

    I kept the existing project memory concise and recorded the decisions and unfinished work needed to resume the task. I did not add a session-by-session transcript or store secrets.

    .claude/memory/context.md

    Project memory

    Commands

    • Tests: uv run pytest (warnings are errors)

    Decisions

    • NO_COLOR handling is centralized in should_strip_ansi().
    • An explicit color=True takes precedence over NO_COLOR.

    Active tasks

    • Wire the existing NO_COLOR helper into the execution path.
    • Add tests in tests/test_utils/.
    • Add a changelog entry.

    Working rules

    • Verify memory entries against the current code before acting on them.
    • Keep only durable, high-value project facts; merge or prune old entries instead of appending forever.
    • Never save secrets, tokens, passwords, or personal data.

    Next step

    Inspect the relevant implementation and tests, then finish the wiring, add coverage, update the changelog, and run the test suite. No source code changes are implied by this memory update alone.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Pick up where you left off.

    Every new Claude Code session starts without yesterday's conversation. mind-palace gives your project a small, reviewable memory file so important decisions and unfinished work survive across sessions.

    It stores durable project facts in .claude/memory/context.md and loads that file into Claude's context at session start, after resume, after /clear, and after compaction.

    What it does

    • Remember — /mind-palace:remember <fact> saves a durable fact under the right heading and merges it with existing notes instead of endlessly appending to a log. With no argument, it saves the useful facts from the current session.
    • Recall — /mind-palace:recall shows the current memory file and its line count.
    • Compact memory — /mind-palace:compact-memory checks entries against the code, then merges, compresses, and prunes stale notes.
    • Load context automatically — the SessionStart hook injects the project memory into new and resumed sessions, including after /clear and compaction.
    • Guard manual compaction — when memory has not changed, the PreCompact hook can block one manual /compact and remind Claude to save first. It does not block automatic compaction.

    Designed to stay lean

    The memory budget is about 150 lines (with an early warning around 9 KB). It is plain Markdown that you can inspect, edit, commit, or keep local. It focuses on architecture and project facts—not conversation transcripts or copied source files.

    The skill's working rules prioritize correctness: inspect the actual code before changing it, treat memory as a lead rather than proof, locate files before reading them, and verify stale notes against the repository. It instructs Claude not to store secrets, tokens, passwords, or personal data.

    Lightweight and local

    • No Python or Node.js runtime dependency for the hooks; they use POSIX sh.
    • No plugin-operated server, telemetry, analytics, tracking, or network requests.
    • On Windows, Git for Windows is needed to provide sh.
    • Claude Code may ask you to approve writes inside its protected .claude/ directory.

    The memory file is still part of Claude's context and is sent to Anthropic with the rest of your Claude Code conversation. Review it before sharing or committing it to a repository.

    What the benchmark says

    In a small blind benchmark on pallets/click (30 headless runs across five tasks), the cross-session continuation task scored 10.0/10 with mind-palace versus 8.3/10 without it. Across all tasks, average token cost was 10.6% higher. This is a small experiment, not a universal guarantee: mind-palace is intended to improve continuity on work that spans sessions, not to reduce token usage on every task.

    Built for Claude Code; the hooks and namespaced slash commands are not a drop-in integration for other agents. Memory is injected into every session, so it can increase token cost rather than save tokens; the repository's small benchmark measured +10.6% average cost. The memory budget is about 150 lines / 9 KB and is intended for durable facts, not full transcripts or file contents. The PreCompact guard applies only to manual /compact and does not block automatic compaction. Hooks use POSIX sh; Windows requires Git for Windows. Writes under .claude/ may require user approval. Review memory before committing it because it is loaded into future sessions.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 22 days ago

    Requires

    Claude Code
    Git for Windows
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    .claude/memory/context.md
    ~/.claude/plugins/data/**

    Reads relevant project files to verify stored facts. Maintains .claude/memory/context.md and creates short-lived empty session markers under ~/.claude/plugins/data/. The plugin itself makes no network requests and has no telemetry. Memory text is included in normal Claude Code context and is therefore sent to Anthropic with the rest of the conversation. Claude Code may ask for approval before writing inside .claude/.

    Listed22 days ago
    Updated18 days ago

    What's inside

    Frequently Asked Questions

    No. The memory is included in every session's context. In the project's small blind benchmark, average token cost was 10.6% higher. The intended benefit is better continuity when work spans multiple sessions, not lower token usage on every task.

    Durable project facts such as the tech stack, repository structure, build and test commands, architectural decisions and their rationale, coding conventions, preferences, and unfinished tasks. It is designed not to store secrets, tokens, passwords, personal data, full file contents, or transient session details.

    This package is built for Claude Code. Its SessionStart / PreCompact hooks and /mind-palace:* commands depend on the Claude Code plugin runtime. Do not advertise it as fully compatible with Cursor, Codex, Gemini, or Copilot unless you create and test separate integrations for them.

    The plugin operates locally and makes no network requests or telemetry calls. However, the memory file is added to Claude's normal context and is sent to Anthropic with the conversation. Review the file before sharing or committing it.

    Claude Code protects files under .claude/. In its default permission modes, it asks before writing there. You can approve edits to the project's .claude folder for the current session when prompted.

    Yes, with Git for Windows installed so the hooks can run through sh.

    New in Prompt & Skill Engineering