Release Tweet Skill

Compose a release announcement tweet from GitHub release notes. This skill fetches the release, summarises changes, identifies contributors, cross-references X/Twitter handles, and outputs composed text for review. It does NOT post the tweet.

Workflow

Step 1 — Fetch release notes

Determine the repository from the current working directory or user input.

If the user specifies a tag:

gh release view <tag> --repo <owner/repo>

If no tag is given, find the latest release:

gh release list --repo <owner/repo> --limit 1

Then fetch its body:

gh release view <tag> --repo <owner/repo>

Extract the release body markdown for processing.

Step 2 — Summarise changes as tweet bullets

Parse the release markdown sections (Added, Changed, Fixed, Removed) and rewrite each item as a concise, outcome-focused bullet:

• One line per bullet
• Merge related items where possible
• Drop purely internal items that don't matter to end users
• Keep the list to a reasonable length for a tweet (aim for 5-10 bullets max)
• Use backticks for CLI commands, tool names, and code references
• Focus on what the user gains, not implementation details

Step 3 — Identify external contributors

Scan release notes for attribution patterns like by [@username](https://github.com/username).

Extract GitHub usernames, then determine the repo owner:

gh api repos/<owner>/<repo> --jq '.owner.login'

Filter out:

• The repo owner
• Bot accounts (e.g. dependabot, github-actions)

The remaining usernames are external contributors.

Step 4 — Cross-reference X/Twitter handles

For each external contributor, run the multi-signal verification procedure documented in references/handle-verification.md.

Classify each result as high confidence or low confidence.

Step 5 — Compose the tweet

Assemble the tweet using the template and rules in references/tweet-format.md.

Output

Present two things to the user:

1. The composed tweet text — ready to copy and post
2. A confidence table — showing each contributor's GitHub handle, resolved X handle (if any), confidence level, and reasoning

Example confidence table:

GitHub	X Handle	Confidence	Reasoning
@alice	@alice_dev	High	GitHub twitter_username field set
@bob	bob_codes	Low	Same handle exists on X but no corroborating signals
@charlie	—	—	No X presence found

The user reviews everything and posts manually.