Describe¶

Auto-generates PR descriptions: title, type, summary, code walkthrough, and an optional Mermaid diagram.

Run it from the CLI:

mergemate-review --pr_url https://github.com/owner/repo/pull/42 describe

Or comment on the PR:

/describe

Within ~30 seconds the tool fills in — or updates — the PR description.

Pass config overrides inline:

/describe --pr_description.generate_ai_title=true --pr_description.publish_labels=true

Automating description¶

[github_app]
pr_commands = [
    "/describe",
]

[pr_description]
publish_labels = true

Preserving your own description¶

MergeMate places anything above the generated "PR Type" header into a preserved zone. Write your own notes there and they'll survive subsequent /describe runs.

If your original description gets clobbered, two things happened:

You edited the description while the tool was running — a race condition.
Your content sat below the auto-generated markers, so MergeMate treated it as replaceable.

Always write your initial description before opening the PR so the tool can see it.

Mermaid sequence diagrams¶

The tool can render a horizontal Mermaid flowchart of component interactions right in the PR body. Controlled by enable_pr_diagram (default: true).

Configuration¶

Option	Default	Notes
`publish_labels`	`false`	Applies type labels (bug, enhancement, docs, etc.) to the PR.
`publish_description_as_comment`	`false`	Posts as a comment instead of overwriting the PR body.
`publish_description_as_comment_persistent`	`true`	Edits the same comment on re-runs.
`add_original_user_description`	`true`	Preserves your description above auto-generated content.
`generate_ai_title`	`false`	Rewrites the PR title using AI.
`extra_instructions`	`""`	Additional guidance for the model.
`enable_pr_type`	`true`	Shows the "PR Type" field in the output.
`final_update_message`	`true`	Posts a note when the description updates to the latest commit.
`enable_semantic_files_types`	`true`	Generates the "Changes walkthrough" section.
`file_table_collapsible_open_by_default`	`false`	Expands the file table by default.
`collapsible_file_list`	`"adaptive"`	Collapses the file list when there are >8 files. Set to `true` or `false` to override.
`enable_large_pr_handling`	`true`	Makes multiple AI calls for large PRs to cover more files.
`enable_help_text`	`false`	Appends help text to the output.
`enable_pr_diagram`	`true`	Renders a Mermaid flowchart of the changes.
`auto_create_ticket`	`false`	Creates a ticket in the linked ticketing system on PR open.

Markers — template mode¶

Enable markers to control exactly where each section lands in the PR body:

[pr_description]
use_description_markers = true

Then place tokens in your PR description:

## PR Type:
mergemate:type

## PR Description:
mergemate:summary

## PR Walkthrough:
mergemate:walkthrough

## PR Diagram:
mergemate:diagram

MergeMate replaces each token with the corresponding generated content. If no markers exist, the description is left untouched.

Param	Default	Notes
`use_description_markers`	`false`	Switches to marker-based rendering.
`include_generated_by_header`	`true`	Adds a "Generated by MergeMate" header.

Custom labels¶

The built-in label set is generic: Bug fix, Tests, Enhancement, Documentation, Other. Define your own:

[config]
enable_custom_labels = true

[custom_labels."performance"]
description = "Use when the PR changes affect runtime speed or memory usage"

[custom_labels."api_change"]
description = "Use when a public API signature is added, removed, or modified"

Each description should be a conditional statement that tells the model when to apply the label. Labels that no longer apply are automatically removed on the next /describe or /generate_labels run.

You can also manage labels from your repo's label page — prefix the description with mergemate: (e.g. mergemate: A new SQL query was added).

Tips¶

Set /describe as a default pr_command so every PR gets a consistent description with zero effort.
Use markers if you have a strict PR template and want generated content in specific spots.
Custom labels pay off fast. Spend five minutes defining labels that match your team's vocabulary and the model will use them accurately.