Repository navigation
Document Pulumi Agents and reorganize Infrastructure AI docs - #22287
Conversation
Sentinel — merge gate statusAll gates green — nothing is blocking this merge.
Evaluated at head Need help, or think a gate got it wrong? Ask in |
Lighthouse Performance ReportCommit: 6de586e | Metric definitions
|
There was a problem hiding this comment.
🔧 One-click fixes for findings that block merge — each matches a Do this block on the author card. Commit one and the card picks up the fix; you can still disagree on the card instead.
Generated by Claude Code
|
/new-review the last review run failed Generated by Claude Code |
Reviewer's guide v3 — not for the authorTip Check the PR author: nothing here is yours. Your to-do list is the "Author action guide" comment. Review requested from: @pulumi/docs-guild, @pulumi/docs-marketing-review, @pulumi/docs-tools — an approval from anyone with write access satisfies the merge gate. Note What this PR changes:
What would block a reader is setup steps or permission statements that don't match the shipped product, because readers grant real credentials based on them. The review ran claim verification, the Hugo build and link check, a frontmatter sweep, readthrough structure detectors, and Vale. Review confidence:
|
| ID | Where | Finding |
|---|---|---|
| F4 | content/docs/ai/agents/integrations/datadog.md L17-22 |
"Prepare Datadog access" grants permissions to the service account before "Create the credentials" step 1 creates that account. A reader can still finish, since the account can be created first and permissioned afterward. Confirm the order is acceptable, or ask the author to move account creation ahead of the permissions paragraph. |
| F24 | content/docs/ai/agents/custom.md L47 |
Only the PR's own permissions page backs the claim that roles and credentials, not instructions, enforce an agent's access limits. If it's wrong, readers will over-trust prompt restrictions. Confirm with someone who knows the product, or accept the author's answer to F7. |
| F26 | content/docs/ai/agents/integrations.md L78 |
Only sibling pages in this PR support the claim that Neo and configured agents share the same integration connections. That affects how admins reason about credential blast radius. Confirm it with someone who knows the product. |
| F28 | content/docs/ai/agents/tasks.md L33 |
The rule that Slack channel membership alone grants no access to an agent's tasks rests only on the Slack integration page. It's a security-relevant statement, so confirm it with someone who knows the product. |
Editorial stances introduced by this PR
Superlatives and comparisons the diff adds. Framing isn't fact-checkable, so confirm each is a stance the docs should take.
- L44
content/docs/ai/agents/permissions.md— "To start a task, a teammate needsagent_task:createand access to the agent. For an agent with an assigned role, the teammate must hold that role directly…" — positioning (found by regex)
✅ What you can rubber-stamp
- Facts: 314 factual claims checked — 175 verified clean, 3 flagged in the
⚠️ list, 136 settled — see the evidence page. - Mechanics: frontmatter sweep ran; Hugo build green; 122 added internal link(s) resolve; 5 detector finding(s) filed above.
- Style: 15 advisory suggestions left with the author; never blocking.
Pre-existing issues in touched files: 2 — details on the evidence page.
Full evidence: verification trail, investigation log, review history.
Review v3 · updated 2026-10-23T18:18:33Z · head commit 6de586e
For the reviewer: something look wrong? Comment /update-review <your point> and the review re-checks with your input.
Author action guide v3 — nothing blocks mergeImportant You = the PR author. Answer every item below before this PR can merge. To answer: push a fix, or reply This PR adds a Pulumi Agents docs section (configuration, permissions, triggers, tasks, and cloud and MCP integrations) and reorganizes the Infrastructure AI docs. The review checked its facts, links, the Hugo build, readthrough structure, and style. Style suggestionsOptional polish from pattern-based linting and the review's own read; never blocking. content/docs/ai/agents/_index.md — 1 (1 heading capitalization)
content/docs/ai/agents/integrations/azure.md — 1 (1 wordiness)
content/docs/ai/agents/integrations/datadog.md — 1 (1 wordiness)
content/docs/ai/agents/integrations/google-cloud.md — 1 (1 wordiness)
content/docs/ai/agents/manage.md — 1 (1 wordiness)
content/docs/ai/agents/permissions.md — 1 (1 difficulty qualifier)
content/docs/ai/agents/triggers.md — 2 (2 wordiness)
content/docs/ai/coding-agents/_index.md — 1 (1 heading capitalization)
content/docs/ai/neo/integrations/cli/_index.md — 2 (2 misspelling)
content/docs/ai/neo/integrations/mcp/_index.md — 3 (2 wordiness, 1 heading capitalization)
content/docs/ai/neo/integrations/slack/_index.md — 1 (1 article before product name)
✅ Resolved since last review15 resolved items — click to expand
Full evidence: verification trail, investigation log, review history. Review v3 · updated 2026-10-23T18:18:33Z · head commit 6de586e Please don't edit or delete this comment; it's the review's record. |
|
/update-review with latest changes |
|
/update-review Confirm all unresolved findings -- @kramhuber is the owner of the functionality, he's the SME. |
Adds documentation for Pulumi Agents covering pre-built and custom agents, configuration, integrations, permissions, tasks, events, schedules, and troubleshooting. Incorporates Neo as the pre-built personal agent while preserving its dedicated guides.
Reorganizes Infrastructure AI navigation into Get started, Concepts, Guides, Integrations, and Reference. Updates the landing pages to distinguish Pulumi Agents from using Pulumi with external coding agents, and preserves existing URLs.
Marketing site updated covered in https://github.com/pulumi/marketing-web/pull/252