Codebase Orientation: Map a Repo in 10 Tool Calls

    1

    A cheap, deterministic procedure that gets your agent oriented in an unfamiliar repository before it spends tokens reading files or risks editing production.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Codebase Orientation: Map a Repo in 10 Tool Calls

    Example session with this skill installed

    Where do we handle payment webhooks in this repo? I want to add idempotency checks.

    • Read your context and instructions
    • Compiled the codebase orientation

    Orientation: shopfront-api

    Mapped in 9 tool calls with 2 full file reads.

    Type and stack: Python 3.11, Django 5, Stripe SDK, Celery
    Run and test:   make dev, make test (pytest)
    Git:            branch main, clean, last commit "Bump stripe to 9.x"
    
    Layout:
      api/        HTTP views and serializers
      services/   billing, inventory, email logic
      jobs/       Celery tasks
      tests/      pytest suites mirroring api/ and services/
    
    Entry points:  manage.py, config/urls.py
    Config:        config/settings/base.py, config/settings/prod.py, .env.example
    Tests:         64 files under tests/
    Largest files: services/billing.py (812 lines), api/views.py (540 lines), jobs/sync.py (410 lines)
    
    For this task:
      api/webhooks.py       found via concepts "webhook" and "api" plus "stripe", 14 hits
                            imports services.billing; imported by config/urls.py
      services/billing.py   9 hits; already has a record_event() helper
      tests/api/test_webhooks.py  3 existing tests
    
    Environment:   no production signals
    Open questions: should idempotency keys live in Postgres or Redis?
    

    Next steps

    • Add an idempotency check at the top of the handler in api/webhooks.py, keyed on the Stripe event ID.
    • Reuse record_event() in services/billing.py rather than writing a new table helper.
    • Extend tests/api/test_webhooks.py with a duplicate-delivery test.

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

    About this skill

    Dropping an agent into a repository it has never seen usually goes one of two ways. It reads dozens of files and burns context before doing any work, or it guesses at a file and edits the wrong thing. This skill replaces both with a short, repeatable procedure built on file listing, search, and a handful of reads.

    The agent identifies the project type and stack from manifest files, captures git state, maps the tree by directory with entry points, config, tests, and the largest files, then traces imports and importers for the files the task actually touches. When the task names a concept rather than a file, a concept-to-regex reference maps "auth", "webhooks", "retries", and two dozen other ideas to search patterns that work across languages. Files are ranked by hit count and only the top few get read.

    Before any change, the agent checks for production signals: framework environment variables, hosted database hosts, a stray .env file, a kube context with prod in the name. If one fires, the agent stops and asks you to confirm the target instead of assuming. It never prints the secret values it is checking.

    The whole thing targets 8 to 15 tool calls and at most three full file reads. Example commands are given for both Unix shells and PowerShell. The output is a 40-line summary your agent can store in project memory so the next session skips the scan.

    Built by Korey Dillon (LatentAxis). Source: https://github.com/koreydillon/cipher-harness

    Above roughly 2,000 code files the agent scopes the scan to the subtree the task concerns and says so. Import tracing is pattern based, so dynamic imports and dependency injection can be missed. Production detection is heuristic: it catches common signals, not every possible setup. Best with an agent that can run shell commands; read-only file tools work but take more calls.

    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

    Requires

    Python

    Permissions

    Terminal / Shell
    Read Files
    Environment Variables

    The agent lists files, runs read-only searches (find, grep, git status, git log), and reads a few files. For the production check it looks at environment variable names and hosts, and never prints secret values. It makes no network calls and does not write files.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Instead of reading files one by one, the agent follows a fixed procedure: manifests for type and stack, git state, a directory map with entry points and tests, then search hits ranked by count. It targets 8 to 15 tool calls and at most three full file reads before it reports back.

    Yes. It detects the stack from manifest files such as package.json, pyproject.toml, go.mod, Cargo.toml, Gemfile, and .csproj, and the concept-to-regex search map is written to match patterns across languages.

    Any agent that reads SKILL.md files and can list, search, and read files. Shell access is best, with example commands for both Unix shells and PowerShell. There are no scripts or dependencies to install.

    Before the first change, the agent looks for production signals such as framework environment variables set to production, hosted database hosts, a stray .env file, or a kube context containing prod. If one appears, it stops and asks you to confirm the target before editing. It never prints the values it checks.

    Yes. It runs the manifest step at the root and in each package directory, and above roughly 2,000 code files it scopes the scan to the subtree your task concerns and tells you it did.

    SKILL.md with the seven-step procedure, a tool-call budget, and an output template, plus references/search-map.md mapping 28 concepts (auth, webhooks, retries, and more) to cross-language search patterns. It is free and MIT licensed.

    New in Productivity