More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Déjà Vu Search: Stop Re-Debugging Solved Problems

    1

    Search past AI coding sessions with the deja CLI before your first edit — reuse solved fixes, honor settled decisions, and never recommend a tool the user already rejected.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Déjà Vu Search: Stop Re-Debugging Solved Problems

    Déjà Vu Search: Stop Re-Debugging Solved Problems

    Example session with this skill installed

    I am about to add Redis caching to our API service. Before I design anything, check whether we already decided something about caching for this service.

    • Read your context and instructions
    • Compiled the d j vu

    History check — Redis caching for the API service

    Search run before any design work: "redis caching decisions" for this service.

    Envelope read first: real match (not a relevance near-miss), total 1, capped false, nothing withheld. The hit is a session from February where you rejected Redis in favor of in-memory caching for this service, citing operational overhead. No superseded marker — the decision stands.

    Action: Redis is not proposed. The design proceeds with in-memory caching, per your February decision. One line to you: "Déjà vu: in February you decided against Redis here (operational overhead) — sticking with in-memory unless you want to revisit that."

    What was avoided: a full design round for a tool you already rejected, a debate you already had, and a recommendation that would have contradicted your own earlier judgment. The February reasoning is in the session record if you want to revisit it — the decision is honored, not buried, and revisiting it is one question away. The check took seconds; the design round it replaced would have taken an afternoon, and the recommendation it prevented would have been wrong on arrival. The check took seconds; the design round it replaced would have taken an afternoon, and the recommendation it prevented would have been wrong on arrival.

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

    About this skill

    Past sessions hold decisions, dead ends, and fixes that never made it into the repo. This skill searches that history so you do not re-debug a solved error, re-litigate a settled decision, or recommend a tool the user already rejected — and it insists the search happens before the first edit, not after the mistake is repeated. The cost of a search is seconds; the cost of re-debugging a solved problem is an afternoon.

    It teaches you to read the result envelope first: tier ("relevance" means nothing matched — do not count near-misses as hits), total and capped (judge coverage from those, never from printed hits), policy_withheld (an empty result and a withheld result are different answers), and superseded markers (never repeat a rejected approach). Two worked examples show it in action: a Redis proposal stopped by a February rejection, and a JWT clock-skew error fixed in minutes from a March session. A rubric gates the discipline: below 6 out of 8, the search happened but the judgment around it did not.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Productivity.

    • Result windows are bounded: it never reports corpus-wide counts or claims a complete audit from the hits shown.
    • A hit without a superseded marker is information, not approval — and a stale hit needs confirmation before you act on it.
    • If deja is not installed or the index is empty, history search is unavailable; the skill says so rather than inventing what it might have found.
    • History informs, it does not override: your current instructions always win, with the conflict surfaced, not silently resolved.

    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 buying a skill.

    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 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Tier "relevance" means nothing actually matched — those are the nearest sessions the search could find. A real match (like tier "error" for an error query) means the query matched by signature. Do not count near-misses as hits.

    Because the search is cheap and the mistake is expensive. Searching after you have already repeated the error confirms what you knew — it does not prevent the wasted work. The discipline is timing, not just technique.

    Absence is a valid answer — proceed. But if the task is high-stakes (production auth, a migration, an irreversible operation), say the history is silent and proceed with extra verification, not with extra confidence.

    New in Productivity