Skip to content

Use cases

Every workflow below ships with auto-swe as a built-in template. Each page is generated from the template’s own spec, so its diagram and its list of human steps are what the run actually does. 30 templates, one engine: a team that needs something else authors its own on the same nodes.

A ticket goes in and a pull request comes out. These differ in how much checking happens between the two, and none of them merges.

Use case Stops for people Maturity
Ticket to reviewed pull request 1 place a person acts Flagship flow
Ticket to merge with a Qodo review 2 places a person acts Built from the flagship steps
Production hotfix no person inside the run Built from the flagship steps
Dependency update no person inside the run Built from the flagship steps
Model catalog refresh no person inside the run Built from the flagship steps
Code and CI only no person inside the run Built from the flagship steps
CI failure triage and fix no person inside the run Built from the flagship steps
Pull-request feedback no person inside the run Built from the flagship steps
Agent-reviewed pull request no person inside the run Built from the flagship steps
Two-reviewer consensus no person inside the run Built from the flagship steps
Parallel implementation no person inside the run Built from the flagship steps

The same engineering steps with people placed in the run: an approval, a decision between paths, a form to fill in, or a diff to review.

Use case Stops for people Maturity
Approve before the pull request opens 1 place a person acts Built from the flagship steps
Clarify scope first 1 place a person acts Built from the flagship steps
Security triage 1 place a person acts Built from the flagship steps
Human review of the diff 2 places a person acts Built from the flagship steps
Risk-tiered approval 3 places a person acts Built from the flagship steps
Four-eyes change 2 places a person acts Built from the flagship steps
Database migration with a dry run 1 place a person acts Built from the flagship steps
Fully supervised 4 places a person acts Built from the flagship steps

Runs that wait on systems outside auto-swe — a deploy pipeline, a monitoring check — before they continue.

Use case Stops for people Maturity
External approval gate no person inside the run Built from the flagship steps
Staged deploy confirmation no person inside the run Built from the flagship steps

From a brief to a PRD, from a PRD to tickets, from tickets to implementation.

Use case Stops for people Maturity
Draft a PRD 1 place a person acts Seeded, not yet proven
File a tracker issue 1 place a person acts Seeded, not yet proven
Brief to tracker issue 1 place a person acts Seeded, not yet proven
PRD to implementation 2 places a person acts Seeded, not yet proven

Customer replies drafted by an agent and published under your autonomy policy.

Use case Stops for people Maturity
Support ticket reply 1 place a person acts Seeded, not yet proven

Drafts and announcements written into the tools your team already reads.

Use case Stops for people Maturity
Draft an update in Notion 1 place a person acts Seeded, not yet proven
Draft with a brand review 1 place a person acts Seeded, not yet proven
Post a Slack update 1 place a person acts Seeded, not yet proven
Tracker issue to Confluence page 1 place a person acts Seeded, not yet proven

None of these ship. Each is an idea for a team to author on the same nodes the templates above use, with what it would be built from — including anything it needs that the platform does not provide. See architecture for the node types.

Once a week, compare the docs against the code they describe, and open a pull request fixing what no longer matches.

Built from: A scheduled work request on the repository, running code-and-ci with a custom skill on the implementer that says what drift to look for.

Draft release notes from the pull requests merged since the last tag, for a person to edit before they are published.

Built from: The template webhook, fired by CI on a release tag; an mcp node calling a GitHub MCP server (not shipped) for the merged pull requests; an agent node with contentWriter; a humanReview node; writeOutcome to a Notion page.

Turn an incident timeline into a first-draft postmortem the incident lead corrects, rather than writes from a blank page.

Built from: The template webhook, fired by the incident tool with the timeline as its payload; an agent node to draft; a humanReview node for the incident lead; writeOutcome to a Notion page.

Summarise the week in support — volume, recurring problems, tickets waiting longest — and post it to a channel once someone has approved it.

Built from: An external scheduler firing the template webhook — schedules inside auto-swe are tied to a repository; an mcp node calling a help-desk MCP server (not shipped); an agent node to summarise; a humanApproval node, then writeOutcome to Slack.

© 2026 Jorge Barnaby

Source-available under FSL-1.1-MIT

Source on GitHub

Built with Astro and Starlight