adk-bot-review

name	adk-bot-review
description	Reviews adk-bot pull requests by validating technical correctness against source code, checking PR states and formatting, and following Google's developer documentation style and ADK docs conventions. Triggers on "review bot PR", "check bot changes", or "verify bot issue".

Review changes proposed by adk-bot (adk-bot-review)

This skill guides a rigorous review of adk-bot changes. It ensures documentation updates are technically accurate, appropriately sized, and follow ADK docs conventions, based on Google's developer documentation style.

Step 1: Verify and research the proposed adk-bot change

• Check that the PR title briefly describes the topic being updated, not only the issue.
• Confirm changes are made directly in the .md files and no "do not merge" status tags were prematurely removed.
• Check the PR's mergeable state and CI status. Note git conflicts and failing checks as findings to resolve or to fix in the consolidating PR.
• The input is one adk-bot-authored PR, or occasionally a few. From the PR, trace back to the issue and the specific item it addresses. Identify the item from the PR title ("Update ADK doc according to issue # - ") and body. adk-bot often opens more than one PR for the same item. If multiple PRs are provided, confirm they all address the same item and review them together as one consolidation set. If they address different items, ask the user which one to focus on before continuing.
• Search closed issues and merged/rejected PRs across the repository for similar topics. Ensure the bot is not re-introducing changes, APIs, or stylistic choices that human maintainers have previously debated and rejected.
• Summarize the item's intended change and stated reasoning, and carry that intent into the technical validation in the next steps.

Step 2: Assess update criteria and architecture

• Prioritize smaller, surgical updates over comprehensive overhauls unless the existing docs are entirely unclear.
• Verify that the update provides a useful learning experience and is actionable for a developer.
• Ensure the content is placed appropriately within the ADK docs information architecture.
• Confirm the update length and prioritization are appropriate for the specific code change (check if it is already covered by existing docs or API references).

Step 3: Validate the change against source and intent

• Never assume an AI agent's code snippet is correct.
• Cross-reference all code examples against the https://github.com/google/adk-python repository.
• Verify the information against relevant pull requests and actual ADK code in the https://github.com/google/adk-python repo.
• Confirm the change actually implements what the linked issue and adk-bot PR set out to do, and resolves the issue item it references.
• Flag any part of the change that drifts from that stated intent, or any part of the intent the change leaves unaddressed.

Step 4: Check quality and style

Headings

• Maintain a logical h1, h2, h3 hierarchy without empty headings (stacking headers).
• Do not number headings (Use ## This heading, not ## 1. This heading).
• Use imperative titles (e.g., "Build agents", not "Building agents").
• Do not use command names in headings; call out the task to be done.

Language

• Follow Google's developer documentation style (https://developers.google.com/style) as the basis for conventions; consult it for anything not covered here.
• Use active voice phrasing (e.g., "The system does X", not "X is done").
• Never use the word "this" without a noun object (e.g., "This function is used...", not "This is used...").

Code snippets and formatting

• Enforce an 80-character line length limit for general Markdown text (excluding long URLs, unbreakable code blocks, or tables).
• Check for rendering errors caused by spaces before backticks (```).
• Use the !!! warning/tip/note mkdocs-material admonition format for notes and warnings instead of quote formats (>).
• Do not use ALL CAPS except for code syntax; use bold or italics for emphasis.

Step 5: Report findings and decide the recommendation

First, summarize the findings from the verification steps above, ordered blockers first, then required edits, then nits. Then choose one outcome below, justified by those findings:

• Close the bot PR(s). (sometimes) The change is wrong, outdated, targets a deprecated or unreleased API, or is already covered by live docs or a merged PR. Record the reason and link the superseding doc or PR.
• Approve as-is (rare). The bot PR(s) are correct, well-placed, and need no edits.
• Edit in place, then approve (sometimes). The existing branch is usable but needs conflict resolution, corrections, consolidation, copyedits, or relocation.
• Open a new consolidating PR (most common). Rework the change across the 1-3 bot PRs into one PR that fixes code errors and hallucinations, edits for quality, and places the content in the correct file. Then close the superseded bot PR(s).

Also confirm the change targets the correct file (files may have moved or been restructured, or the edit may belong on a different page).

Step 6: Produce an action plan

Output the concrete next actions for the chosen recommendation:

• The specific local edits needed: file paths, what to change, code fixes, consolidation of the 1-3 bot PRs, and correct placement.
• A short list of the verification findings that justify them, ordered blockers first (unverified/hallucinated code, wrong file, failing CI checks, broken rendering), then required edits, then nits.
• Which bot PR(s) to close.
• A PR description following the repo convention:
  • a short bullet summary of what changed
  • Rendered page:
  • Agent PR: <link(s)>
  • Original issue: # -

Style rules to bake in

• Model strings should be gemini-flash-latest, gemini-pro-latest, or other valid *-latest aliases in sample code rather than specific versioned model strings, which increase the maintenance burden when new model versions are released.
• Imports: from google.adk.agents import Agent; MCP uses from google.adk.tools.mcp_tool import McpToolset. Agent variable is root_agent (Python).
• No em dashes or verbose AI-generated content: Use a colon in **term**: definition bullets; split prose into sentences otherwise.
• When referring to ADK, use "ADK", never "Google ADK" or "The ADK".
• Do not start a sentence with an inline-code word.
• Internal ADK-docs links are site-relative (e.g. /sessions/memory/); external links must resolve. Do not invent links.
• Verify all code samples against actual APIs and library code. The canonical source repositories are listed in docs/community/contributing-guide.md.