diagnose

    1

    Find evidence-backed causes of bugs and slowdowns, with tested fixes or a clear plan to verify the diagnosis.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    diagnose

    Example session with this skill installed

    Assess only: checkout p95 rose from 240ms to 1.8s after deploy. Trace r81 shows 24 sequential 60ms SQL calls; old trace r12 shows 2. No runtime access or code changes.

    • Read your context and instructions
    • Compiled the diagnose

    Reviewed the supplied timings and trace summaries only. No code or runtime changes were made; the root cause remains unconfirmed.

    Checkout slowdown assessment

    FindingEvidence and confidenceVerification needed
    More SQL calls are a likely contributorMedium: r81 reports 24 sequential calls versus 2 in r12. At 60ms each, the newer calls total 1.44s.Compare matched requests with the same cart size and cache state.
    A per-item query may have been introducedLow: the increased call count fits repeated lookups, but query text and caller details are missing.Inspect SQL spans and their call sites for repeated query patterns.
    Different request inputs could explain the countLow: cart contents and workload equivalence are unknown.Check item counts, request parameters, and cache status for both traces.

    Evidence boundary: The p95 increase is reported, not independently measured. One trace cannot explain the full latency distribution. Trace timestamps, time zone, and deployed build identities are unavailable, so the deployment link is not verified.

    Next steps

    • Share redacted r81 and r12 spans with query text, timestamps, build IDs, and cart sizes.
    • Review the checkout query changes between those builds.
    • If authorized, compare both builds in a safe test environment using identical carts and record call counts and latency.

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

    About this skill

    Hard-to-reproduce bugs and unexplained slowdowns can waste hours on fixes that miss the cause. Get an evidence-based diagnosis, with a tested fix when access permits or a clear verification plan when it does not.

    What it does

    • Investigates failures: Examines available code, logs, traces, tests, and change history.
    • Separates facts from guesses: Reports likely causes with confidence levels and supporting evidence.
    • Checks performance regressions: Uses measurements and comparisons to narrow down what became slower.
    • Fixes and validates: When authorized, tests changes against the reported failure and adds regression coverage where practical.
    • Supports read-only reviews: Delivers findings and verification steps without changing the repository or running system.

    How it works

    1. Describe the failure, expected behavior, and recent changes.
    2. Share relevant code or evidence and specify whether changes are allowed.
    3. Receive an investigation based on the available access and evidence.
    4. Get either a validated fix or a diagnosis report explaining what still needs checking.

    Works with

    Code repositories, application logs, HTTP services, browser captures, databases, and test tools such as Playwright or Puppeteer, depending on available access.

    Why use this skill

    It helps avoid speculative patches by checking explanations against observable behavior. Validation covers the original failure, not just an isolated test that happens to pass.

    Use cases

    • Investigate intermittent checkout errors.
    • Trace an API slowdown after deployment.
    • Assess a production incident without making changes.

    Known limitations

    Verified fixes need a safe way to reproduce the failure and permission to make changes. Limited evidence may leave causes unconfirmed.

    Verified fixes require safe reproduction and permission to change code. Read-only reviews produce findings and verification plans, not patches; missing runtime evidence can leave causes unconfirmed.

    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

    1 use

    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

    Permissions

    Terminal / Shell
    Browser
    Network Access
    Read Files
    Write Files
    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Use relevant repository files, logs, or traces that you choose to provide. The skill has a read-only assessment mode; code changes and runtime access require your authorization.

    Unlike general coding tools, this skill prioritizes evidence-based reasoning, providing confidence levels for each potential cause and ensuring fixes are validated against actual failure conditions.

    If the agent cannot safely reproduce the issue or lacks write permissions, it delivers a detailed diagnosis report with a step-by-step verification plan for your team to follow manually.

    Yes, it is designed to work with performance data and browser captures to identify bottlenecks and regressions in both backend services and frontend applications.

    Designed for Codex. Other SKILL.md-compatible agents may use the workflow, but tool calls and available diagnostics depend on their environment.

    New in Testing & Debugging