Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Context Compass: Repo Prep & Proof-Driven Changes

    1

    Find where a requested change meets the code—and where its tests are missing—before editing an unfamiliar repository.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Context Compass: Repo Prep & Proof-Driven Changes

    Context Compass: Repo Prep & Proof-Driven Changes

    Example session with this skill installed

    Synthetic local repository: contracts/shipping.md says shipping is free from 5000 cents; below that it costs 500 cents. The source and tests are supplied.

    Use Context Compass to prepare a baseline for a later shipping change. Run only the bundled local indexer, then inspect the contract, implementation and test assertions. Show the missing evidence and the next useful step. Do not edit files or run project quality commands.

    • Read your context and instructions
    • Compiled the context compass

    Shipping baseline

    The supplied test file checks the free-shipping threshold. It does not assert the 500-cent charge below that threshold.

    This is an actual local inspection of a synthetic repository. The bundled scanner ran; no project test or quality command ran.

    LinkWhat was inspectedResult
    Rulecontracts/shipping.md:3 specifies free shipping at 5000 cents or more, otherwise 500 cents.Documented
    Codesrc/shipping.mjs:1–3 contains both branches.Implementation located
    Assertiontests/shipping.test.mjs:5–7 checks that 5000 cents returns zero shipping.Threshold asserted; test not executed
    Missing evidenceNo assertion in the supplied test file checks a below-threshold order.Partial coverage

    Next step: if this rule remains the intended contract, add a case immediately below the threshold before changing it. The expected shipping charge is 500 cents under the supplied contract. No behaviour has been inferred from file names.

    Available: npm run check in package.json. Run for this baseline: none. All supplied fixture files remained unchanged. This baseline does not establish test execution or whole-repository correctness.

    in package.json

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

    About this skill

    You have a change to make in an unfamiliar repository. A file list helps you navigate, but it does not tell you which rule governs the change or whether a test checks it.

    Context Compass helps your coding agent follow that trail: documented behaviour → implementation → callers → test assertions. The report shows the evidence for each link and leaves gaps visible, so the next person knows what to inspect or test.

    Two ways to use it

    • Prepare the repository. Build a read-only baseline with file references, unresolved decisions and available checks. The bundled Node.js scanner provides leads; the agent inspects the actual source and assertions.
    • Make one requested change. Agree the behaviour, inspect its contract, demonstrate the failing case, then update code, tests and documentation and record the checks actually run.

    A typical result is practical: “The free-shipping threshold is implemented, but the below-threshold charge has no assertion.” That is a specific next test to consider, rather than a blanket claim that the repository is covered.

    The ZIP contains both workflows and the local scanner. It requires repository access through your agent and Node.js for the scanner; no API key is required. Analysis stays read-only until a change is requested. If a required behaviour decision has no authoritative answer, the workflow asks for it.

    File matches remain leads, and this cannot find every contract or guarantee correctness. The separate public Context Compass plugin remains free and MIT licensed.

    Start with one behaviour you need to understand or change.

    The scanner finds incomplete file/path leads. Actual contracts, callers and assertions still need inspection; no match does not prove absence.
    A required decision without user direction or authoritative repository evidence remains unresolved. Conflicting sources can pause the workflow.
    No guarantee of correctness, complete coverage or measured time, token or defect savings.
    The scanner is local and read-only by default, makes no network calls, skips symlinks and known private-memory directories and does not intentionally search for secrets.
    The assistant host may process repository content. Individual host compatibility has not been certified.
    The output is text. It does not create a PDF; marketplace-generated display cards are separate.

    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 25 days ago

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    **/*

    Uses Terminal/Shell and Read Files for the local Node.js inventory and source inspection. Write Files is used only for requested implementation, inside the selected repository. The bundled scanner skips symlinks and known private-memory directories, makes no network calls, and does not intentionally search for secret values. Project checks follow repository rules and user authorization.

    Listed1 month ago
    Updated16 days ago

    What's inside

    Frequently Asked Questions

    No. It helps trace the requested behavior through inspected contracts, implementation, callers, tests, and docs, then reports what remains unverified. It does not infer missing required behavior: if user direction or authoritative repository evidence is missing or conflicting, it stops and asks. The host agent still needs to review the real behavior and final diff.

    The workflow is language agnostic when the host can inspect the repository. The bundled inventory uses local Node.js built-ins. The ZIP was tested locally; individual agent hosts and every repository layout have not been certified.

    The ZIP contains a SKILL.md with repository-baseline and Proof-Driven Change workflows, a bundled local scanner, and license notices. The scanner is separately available in the free MIT-licensed public plugin. The paid workflow adds a traceable repo baseline and task-specific path from requested behavior through contracts, tests, implementation, docs, and verification.

    It turns a specific request into criteria grounded in the user's words or direct repository evidence, then traces those criteria through actual contracts, callers, code, tests, and docs. Inferences only guide inspection; unsupported required decisions pause for clarification. If implementation is requested, it uses test-first changes and reports exact verification results. It cannot guarantee correctness.

    Only when the user requests implementation. Analysis and planning stay read-only. For requested implementation, the workflow follows repository instructions, makes a narrow test-first change, updates relevant docs, and runs only authorized, safe checks. It pauses before behavior choices that the user request or repository evidence does not settle.

    The bundled scanner uses Node.js built-ins and makes no network calls. Its default scan excludes symlinks and known private-memory directories such as mem0, .remember, and .private; it does not intentionally search for secret values. The assistant host may process repository content according to its own product behavior.

    No. The package produces text briefs and reports only. Agensi's preview may display an automatically generated PDF-style card; that is a platform artifact, not a PDF produced by this skill.

    New in Code Quality & Review