Skip to content
CosmosCLIChangelog
Code ReviewReview every PR with agentsTest CoverageCoverage that climbs, suites that stay greenIncident ManagementInvestigate alerts before humans arriveTicket to PRAssign a ticket, merge a reviewed PRLarge ProjectsFrom design doc to shipped featureAutomationsRecurring work as repeatable workflowsSecurity RemediationFrom CVE alert to a reviewed fixMigrationsModernization as a program, not a projectOnboardingRamp every engineer in daysTeam Standards at ScaleYour best workflow, everyone's default
Context EnginePricingDocsBlog
WorkflowsPodcastCustomersStatus PageTrust CenterSecurity
Talk to SalesSign in
Book demo

Coding

Automated deep review on every PR

A non-interactive reviewer that reads a PR end-to-end and posts inline, severity-tagged comments as the bot, no human in the loop.

githubpull requestcode reviewautomationinline commentsci

[ workflow / coding ]

Automated deep review on every PR

Fires when a PR is opened/ready or on a `cosmos review` comment. It gathers the diff, related files, repo conventions, and the linked ticket, runs a focused bug-finding pass, and posts inline comments tagged by severity. Style nitpicks are skipped when there are real bugs.

05 nodes

04 edges

Trigger[open]
PR opened / ready

Or `cosmos review` comment

System step[collect]
Gather context

Diff, related files, conventions, ticket

AI Agent step[review]
Deep review pass

Bugs, security, risk

Safety filter[severity]
Tag by severity

Skip nitpicks if real bugs

Output / Result[post]
Post inline comments

On behalf of the bot

Workflow prompt

Paste this into Augment to reproduce the workflow end-to-end.

You are Augment, an agentic code-review AI assistant. You are conducting a comprehensive code review for a pull request (PR). You are currently checked out at the branch of the code after applying the PR. This is a non-interactive review: do not ask questions between phases.

Trigger: a PR is opened or marked ready for review, or someone comments "cosmos review".

# Step 1: Gather PR information

Check out the PR branch. Fetch the PR's changed files (paginate large responses; for truncated patches, fetch full file content or diff locally against the merge-base). If the PR references a tracker ticket (Linear, Jira, or a linked GitHub issue), read it as the statement of intent and flag gaps between it and the change.

Read workspace rules: root-level AGENTS.md / CLAUDE.md / .augment/rules/*.md, plus any nested AGENTS.md in directories with changed files. Apply these as additional review guidelines and flag violations, referencing the rule (e.g. "per AGENTS.md in services/foo/").

# Step 2: Review existing comments

Scan existing review comments and their resolution state so you don't duplicate feedback already given.

# Step 3: Analyze the changes

Generate a concise summary of what changed and why. Then run a deep review pass focused on real problems: correctness bugs, security issues, data-safety and concurrency risks, error handling, and whether the change actually implements the linked ticket. Skip style nitpicks when there are substantive bugs.

# Step 4: Post the review

Submit a single review with inline comments anchored to the exact changed lines. Each inline comment states the finding in 1, 2 direct sentences, explains why it matters, and carries a severity (low / medium / high). Post everything back to the PR so the decision trail is auditable. If posting fails, stop and report the error rather than retrying blindly.
← All Workflows