github-releases

name	github-releases
when_to_use	when the user asks for the latest release(s) of a GitHub repo — tag, publish date, release name
description	Fetches the N most recent releases of a GitHub repository via the public releases API and formats tag, published_at, and name as a compact JSON array. Uses `jq` projection so the exec preview never overflows 2KB.
entry	run.sh
args	[{"name":"repo","type":"string","required":true,"description":"owner/repo — e.g. microsoft/TypeScript"},{"name":"count","type":"number","required":false,"description":"How many releases to return. Defaults to 3. Max 20."}]
requires	["curl","jq"]

Behavior

1. Validates count is between 1 and 20 (defaults 3).
2. GETs https://api.github.com/repos/<repo>/releases?per_page=<count>.
3. Projects each release with jq to {tag_name, published_at, name} — drops body, assets, author, and all other large fields so the subprocess stdout stays under a few KB even for a full 20-release list.
4. Prints the resulting JSON array on stdout.

The agent should bind the result with bind_as if it intends to reference specific fields across multiple turns (e.g. formatting per release). Since the canonical output shape will be json<Array<Object>>[N], the preview is already the parsed array — no need to re-run with a different command.

Exit codes

• 0 — success
• 2 — missing repo argument
• 3 — GitHub API returned a non-200 status (e.g. rate limit, not found)

Why this skill exists

Without guidance, models often call curl ... | head -n 3 which cuts the JSON stream mid-object, producing malformed text that makes the model hallucinate the rest. This skill enforces the jq-project-then-emit path so the preview is always valid structured data.