Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Code & Dependency Research for Agents

    1

    Stops your agent inventing APIs. It reads the real source at the exact version you have, then answers.

    $15

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Code & Dependency Research for Agents

    Code & Dependency Research for Agents

    Example session with this skill installed

    Does httpx retry failed requests automatically? Check the real source for our
    pinned version before answering.

    • Read your context and instructions
    • Compiled the code & dependency

    No. httpx does not retry by default

    Verified against the pinned version in requirements.txt, not from memory.

    QuestionAnswerEvidence
    Automatic retries?NoNo retry transport mounted in httpx/_client.py by default
    Retry-After honoured?Only if you implement itNo handling in the base client
    Timeouts applied by default?Yes, since 0.20An explicit timeout object is required

    What the source shows

    httpx.Client mounts transports; retry lives in the transport layer, not the
    client. The default mount is a plain HTTPTransport, which sends one request
    and returns. Nothing loops.

    That means a transient 429 or a dropped connection surfaces to your code -
    which is usually what you want, because silent retries on a rate-limited API
    hide the fact that you are being throttled.

    The wrong answer you were about to get

    "httpx retries three times with exponential backoff by default." That is a
    conflation with requests + urllib3.Retry, which is a different library and
    still not a default.

    Next steps

    1. Add a retry transport with a hard attempt cap and jitter.
    2. Honour Retry-After when present.
    3. Log every retry with the status that caused it.

    Sources: local source of the pinned httpx version, plus the transport
    documentation at encode/httpx

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

    About this skill

    An agent with no source access will confidently give you a function signature that does not exist. It is not lying - it is recalling the most probable shape of an API, and the probability is wrong often enough to cost you a debugging session.

    This pack is the workflow layer for the codesearch MCP server: read the actual source at the version you actually have, then answer.

    What you get

    • 20 worked scenarios. Signature checks, behaviour changes between versions, CVE triage, finding real usage examples, comparing libraries, checking whether a package is maintained, and understanding an unfamiliar repo's entry point.
    • Version truth. The rule that you never answer an API question before establishing the version, why "latest" is the wrong default, how to detect drift between docs and source, and what to say when the version is unknowable.
    • Advisory triage. How to handle a vulnerability list without panicking the user: reachability, direct versus transitive, affected ranges, and when a finding does not apply to them.
    • A failure playbook. Ten modes including the one that catches everyone: long grep.app queries return zero results because the engine uses AND semantics.
    • Five client configs. Claude Desktop, Cursor, Cline, Continue, Zed.

    Why this and not the repo

    The server is free, MIT, and on PyPI - pip install skdev-codesearch-mcp gives you all seventeen tools. This pack is the discipline that makes the answers trustworthy: the call order, the version pin, and the refusal to guess when the source will not load.

    Requires

    • The codesearch MCP server (pip install skdev-codesearch-mcp)
    • No API keys required. GITHUB_TOKEN is optional and raises rate limits.

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

    Requires skdev-codesearch-mcp server connection. Effectiveness depends on package availability within the codesearch index and existence of local manifest files for version pinning.

    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 23 days ago

    Software
    codesearch MCPPython 3.11+Skdev Codesearch MCP

    Install the MCP server with pip install skdev-codesearch-mcp.

    Permissions

    Read Files
    Network Access

    Network access is needed because the server queries live package registries, code indexes and advisory databases. File read access is used to resolve the exact dependency version from your manifest or lockfile before answering. Environment variables are optional keys that raise rate limits.

    Listed23 days ago

    What's inside

    Frequently Asked Questions

    This skill solves 'version drift' and LLM hallucinations by enforcing a strict 'pin-then-read' workflow, ensuring the AI uses the actual source code of your specific dependency version instead of relying on outdated training data.

    It requires the `codesearch` MCP server (skdev-codesearch-mcp) and an MCP-compliant client like Claude Desktop or Cursor; it supports major languages including JavaScript, Python, Go, and Rust.

    Unlike generic prompting, this pack provides deterministic scenario maps and failure recovery playbooks that force the agent to verify versions via manifests (like package.json) before searching, eliminating guesswork.

    The pack includes production-grade playbooks, specific tool sequences for signature and type retrieval, and a failure recovery manual for handling missing packages or ambiguous search results.

    Yes, it includes specialized logic for transitive analysis, allowing the agent to distinguish between direct project dependencies and nested vulnerabilities during security audits.

    New in LLM Ops