More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    recall

    1

    Search every past Claude Code and Codex session on your machine, across every checkout, in milliseconds.

    Free

    2 usesSecurity scanned
    recall

    recall

    Example session with this skill installed

    What did we decide about scoping idempotency keys in the payments service, and did we have to migrate anything? I think it came up during the duplicate-charge incident.

    • Read your context and instructions
    • Compiled the recall

    Summary: Found it in a session from 3 June on the hotfix/dedupe branch, in a different checkout of payments from the one you're in now. You decided to scope keys per merchant, and yes, it needed a migration with a specific ordering.

    What was decided

    QuestionDecision (2026-06-03, session 6e2b8d15)
    Key scopePer merchant, never global. The storage key is the pair (merchant_id, idempotency_key).
    WhyClients choose the key string, so a global namespace lets two merchants both pick order-1 and collide. The collision doesn't fail: it returns the first caller's stored response to the second. That was the cross-merchant receipt leak.
    MigrationYes, and it ships before the code change.
    Key lifetime24 hours: long enough for a client retrying after a network partition, short enough that the table doesn't grow without bound.

    The migration, in the order the session gave

    1. Add a merchant_id column to the keys table.
    2. Backfill it from the linked charge.
    3. Only once the backfill reports zero nulls, swap the unique constraint from the bare idempotency_key to (merchant_id, idempotency_key).

    The session was explicit about the ordering: "Keep the old constraint until the backfill reports zero nulls — dropping it first turns a partial backfill into a silent duplicate."

    Where this came from

    $ recall turns "idempotency key" merchant
    3 of 3 matched turns for "idempotency key merchant"
    6e2b8d15…  2026-06-03  github.com/acme/payments  hotfix/dedupe  assistant  ×9
        Scope the idempotency key per merchant, never globally. The key is a string the
        client chooses, so a global namespace means any two clients picking `order-1`
        collide — and the collision does not fail loudly, it returns the first caller's
        stored response to the second …
    ── 3 sessions · 3 searched · conversation only — tool output NOT searched (--results)
    ── live to 2026-05-12 · archived before that · refreshed just now
    ── scanned 3.0 KB · 14 turns · 37 ms
    ── 1.7 KB · ~441 tokens
    

    The footer shows that only conversation was searched, not tool output, and only this repo. The decision was stated outright in conversation, so a wider search wouldn't change the answer.

    Next steps

    • Check that the composite unique constraint is live: the session agreed on the plan but didn't record running it.
    • Read the whole thread with recall show 6e2b8d15 if you want the TTL discussion in context.
    • Ask "when did we first talk about rate limiting?" to place a related topic in time (recall when).
    
    > The demo is from recall's public demo corpus (a fictional `acme/payments` repo shipped with the source), so it contains nothing from anyone's real sessions. The `recall turns` block is verbatim output, trimmed only where marked `…`.
    

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

    About this skill

    Stop wasting time re-solving technical debt or searching through thousands of lines of past chat logs. This skill enables your agent to search the history of every coding session on your machine to find specific decisions, previous attempts, and context you have already established. It prevents re-deriving solutions that were already settled in past checkouts or different repositories.

    What it does

    • Find sessions by identifying which past conversations discussed a specific topic and how frequently.
    • Rank passages to extract specific turns and quotes across all sessions based on query relevance.
    • Context recovery by showing the turns immediately surrounding a match in a specific session to understand the "why" behind a decision.
    • Timeline tracking by placing a topic in time to see when it was first raised or where a task left off.
    • Cross-repo search by reaching into every local repository history, not just the current working directory.

    How it works

    1. Tool Check: The agent verifies if recall_* MCP tools or the CLI binary are installed and functional.
    2. Contextual Search: You query for a topic like "what did we decide about the auth flow" or "previous attempts at the dockerfile."
    3. Query Expansion: The tool applies curated synonyms (e.g., db for database) and near-neighbor corrections to find relevant hits.
    4. Analysis & Citation: The agent presents the past conclusion, the reasoning, and the session metadata (date, branch, ID).

    Frameworks & tools

    This skill utilizes the recall open-source binary and its associated MCP tools. It is compatible with major AI coding clients including Claude Code, Codex CLI, Cursor, Windsurf, and Copilot.

    Why this beats prompting it yourself

    Standard agents only know the current conversation context. This skill gives the agent a persistent "long-term memory" of every interaction you have ever had with a coding agent on your local machine, eliminating the need to manually copy-paste old logs.

    Use cases

    • Resume work by asking "where did we leave off with the migration script" after a long break.
    • Avoid regression by checking "what did we already try for the memory leak fix" to avoid repeating failures.
    • Audit decisions by finding the specific date and reasoning for a past architectural change.
    • Discover patterns by searching across multiple repos for how you handled a similar integration before.

    Requires local installation of the recall binary. Only searches local agent transcripts from supported clients. Default search excludes tool output unless specified.

    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

    2 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    GitHub
    recall (github.com/mayberuk/recall)
    macOS, Linux or Windows
    Claude Code or Codex CLI

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files

    Allowed Hosts

    raw.githubusercontent.com

    File Scopes

    ~/.claude/projects/**
    ~/.codex/sessions/**
    ~/.local/share/recall/**
    ~/.local/bin/recall

    recall reads your coding agent's session transcripts (~/.claude/projects, ~/.codex/sessions) and never writes to them. It keeps its own archive of the conversation text in ~/.local/share/recall so sessions stay searchable after your agent's cleanup deletes the originals. Searching makes no network requests. Network is used only by the one-time install (which downloads the binary from GitHub and verifies its sha256), by `recall update`, and by a once-a-day version check in `recall doctor` (turn it off with RECALL_NO_UPDATE_CHECK=1). The skill asks before installing anything.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Standard AI agents only see the current chat, but 'recall' gives your agent long-term memory by indexing every past coding session and repository on your machine, preventing you from having to re-solve the same problems twice.

    This skill is fully compatible with major AI coding environments including Claude Code, Codex CLI, Cursor, Windsurf, and GitHub Copilot, provided the 'recall' open-source binary is installed on your host system.

    Yes, unlike standard searches that are limited to your current project, 'recall' allows your agent to search and retrieve insights across every local repository you have ever worked on.

    The agent uses query expansion and curated synonyms (like 'db' for 'database') to find relevant past discussions even if you don't use the exact keywords from the original session.

    The purchase includes the core 'recall' skill logic and MCP tool configurations; however, you must have the 'recall' open-source binary installed on your local machine for the skill to function.

    To maintain clarity and focus, the search excludes raw tool outputs and the current active session by default, ensuring the agent only retrieves meaningful past human-AI interactions.

    Yes, and it was built for exactly that. recall identifies a repo by its git remote, not its folder, so every checkout is one history. Add `--all` to search every repo on the machine.

    No. Agent answers are capped at about 4,000 tokens by default, and every answer states its own size.

    New in Agents & Orchestration