More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Walkthrough Buddy

    1

    Point it at a PR, folder or whole repo and get one page that takes you from "never seen this code" to making your first change: diagrams, ELI5 analogies, real code linked to exact lines, and tested commands.

    Free

    Security scanned
    Walkthrough Buddy

    Walkthrough Buddy

    Example session with this skill installed

    walk me through pallets/click PR #3877 so I can make my first change in this area

    • Read your context and instructions
    • Compiled the walkthrough buddy

    Wrote .walkthroughs/click-pr-3877.html: one page, ~12 min read, 4 diagrams, 12 code excerpts and 62 file:line links, all checked against commit 2247b35. Below is a cut-down copy of its text. The real page also has the diagrams, the code excerpts and the quiz.

    Parameter.spec: naming a parameter the way the user types it

    Click now names options in errors, warnings and help the way users type them: --my-option, not my_option.

    Not covered here: tox -e typing/style/docs were not run. Tests ran on one local Python, not the CI matrix.

    The big picture

    Before this PR, some error messages named an option by its Python name: The option 'my_option' is deprecated. The user never typed my_option. They typed --my-option. The PR adds one property, Parameter.spec, that returns the user-facing spelling, and deprecates human_readable_name.

    Like you're five: a coffee shop. The customer writes an order slip (the decls: '-g', '--gluten-free'). The barista turns it into a kitchen ticket (the name: gluten_free). The menu board shows the drink as the customer orders it (the spec: --gluten-free). Before this PR, the barista shouted the ticket number at the customer. Now they read from the menu board.

    Where it breaks: an argument has no flag, so its menu-board entry is its name in capitals (RECIPE).

    Worked example (real output)

    1. @click.option("--my-option", deprecated=True) builds an Option. _parse_decls derives name = 'my_option' (src/click/core.py:3486).
    2. User runs cli --my-option hello. The deprecation branch runs (src/click/core.py:2960).
    3. self.spec returns '--my-option'.
    4. stderr: DeprecationWarning: The option '--my-option' is deprecated.

    Test it

    • ✅ ran: python -m pytest tests/test_options.py -k test_spec: 7 passed
    • ✅ ran: python -m pytest: 2269 passed, 25 skipped, 1 xfailed in 3.58s
    • ⚠️ not run here: tox r -e typing (mypy + pyright)

    Recipe: change what users see as a parameter's name

    1. Pick spec (messages) or get_help_spec(ctx) (the --help column).
    2. Edit Parameter.spec at src/click/core.py:2570, or Argument.spec at :3932 for arguments only.
    3. Add a case to test_spec (tests/test_options.py:102).
    4. Grep the tests for the old wording. test_deprecated_warning asserts on it.
    5. Add a CHANGES.md bullet and a versionchanged in the docstring.

    Watch out: don't call human_readable_name from Click code. It warns, and Click's tests turn every warning into an error (pyproject.toml:86).

    Next steps

    Answer the 4 check-yourself questions on the page, then try the starter task. It gives the exact test command and what passing output looks like.

    5. Add a CHANGES.md

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

    About this skill

    New to a codebase, or stuck reviewing a big PR? Ask your agent "walk me through this" and Walkthrough Buddy writes one page you read once and then make your first change from.

    What's on the page

    • Big picture: one sentence on what the code (or PR) does, and a before/after diagram for changes
    • Diagrams of the real flow: what calls what, following one real request, not a folder tree
    • ELI5 analogies for every hard idea, plus where the analogy breaks
    • Real code, not made-up snippets: every excerpt is copied from git at a pinned commit and linked to the exact lines
    • Read these yourself: a short list of the files and line ranges worth reading, with what to look for
    • How to contribute: setup, the exact test command (already run), conventions, a step-by-step recipe for the most common change, and the traps
    • Check yourself: a few questions, then a starter task with what "done" looks like

    It checks its own work

    Before handing the page over, a bundled script compares every snippet against the code, follows every file:line link and renders every diagram. A box at the top of the page says what was checked and what the page did not cover.

    ADHD mode

    Say "I have ADHD" or "make it easier to read" and the same content comes in short timed parts, with a "do this first" step, a progress bar, Done ticks and one next step at the end.

    Works on

    • a pull request (#482 or a GitHub/GitLab URL)
    • a branch, commit or range
    • a folder or file (src/billing)
    • a whole repo
    • an idea ("how payments get retried")

    Output is a single HTML page (light and dark), or Markdown if you ask. It's saved to .walkthroughs/ and kept out of git status. Commands run in a throwaway worktree, so your checkout is never touched.

    Needs: git, bash, python3. Optional: gh CLI for PR threads, and Chrome/Edge/Brave to check diagrams render.

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

    • Needs the code checked out locally as a git repo. It reads files from git, so it can't work from a pasted diff alone.
    • Commands only run if the project can be set up on your machine. Anything it couldn't run is marked "not run here" instead of being claimed.
    • Very large repos get a focused page (one main flow and the most-changed areas), not a tour of every module. The page lists what it did not cover.
    • Diagram render checks need Chrome, Edge or Brave. Without one, the page says diagrams were not checked.

    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

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 10 days ago

    • Free with an Agensi account

    Software
    GitBashPython 3
    Optional
    GitHub CLI
    Chrome, Edge
    orBrave

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Environment Variables
    Network Access

    Shell: runs git, the bundled scripts and the project's own tests, inside a temporary git worktree. Read files: reads the code it explains. Write files: saves the page to .walkthroughs/ and adds that folder to .git/info/exclude. Network: optional, for gh to read the PR description and review comments; the HTML page loads Mermaid and highlight.js from a CDN when opened.

    Listed10 days ago

    What's inside

    Frequently Asked Questions

    No. Tests and commands run in a temporary git worktree, and the page is saved to .walkthroughs/, which is excluded locally so it never shows up in git status.

    Every excerpt is copied out of git at a pinned commit and linked to its exact lines. Before the page is handed over, a bundled checker compares each snippet with the code and follows each file:line link. The result is shown in a box at the top of the page.

    No. Point it at a PR, branch, commit range, folder, single file, a whole repo, or an idea like "how payments get retried". Ask for Markdown if you want to paste it into a PR or commit it.

    Yes, the 'Execution Validation' feature runs your build and test scripts in a detached worktree to confirm that the environment requirements and setup steps provided to the new developer actually work.

    You receive the full suite of automation scripts (gather, excerpt, and check), the AI agent configuration for generating the walkthroughs, and templates for Markdown and HTML outputs.

    Walkthrough Buddy is designed to handle complex PRs and large refactors by tracing the main execution path from entry points to side effects, rather than just listing file changes.

    New in Agents & Orchestration

    Free