mise-versions

stores version numbers of common mise plugins

Static version data

mise reads these static files from https://mise-versions.jdx.dev/data/:

• <tool>.toml: every known version of one tool, with created_at and flags.
• index.v1.json: one file covering every tool, with enough entries for latest and common prefixes such as 22 or 3.12. See docs/versions-index.md for the client contract.
• github/{owner}/{repo}/releases.json and tags.json: a GitHub repo's whole release or tag list, baked at deploy time for the repos clients have asked for (others are answered by the Worker at the same URL). See docs/github-versions.md for the client contract.

Tool Update Analysis

This repository tracks tool updates through git history analysis. The chart below shows the daily number of tool updates over the last 30 days.

Daily Tool Updates (Last 30 Days)

Timeline showing the number of tools updated each day over the last 30 days

Token pool observability

The scheduled Worker records GitHub core-rate-limit snapshots every 15 minutes, and the authenticated /admin dashboard displays them. It shows aggregate quota headroom, quota burn per hour, token checkout rate, estimated time until the 1,000-request-per-token reserve, and the latest headroom for each pool token.

Each run checks at most 45 tokens to stay within the Workers Free-plan external subrequest limit while leaving room for alerts. Larger pools rotate through batches and are shown as incomplete rather than being classified as critical.

The monitor warns only when the pool is at risk: only one token is available, 10% or more of the checked tokens fail their check (and at least 10 tokens, or 10% of a smaller pool, so a few dead tokens in one early batch don't count), less than 35% of lendable quota remains, or the pool is within six hours of reserve. It becomes critical at zero available tokens, 15% lendable quota, or two hours to reserve. A few rate-limited, below-reserve or unreachable tokens, or a rotation that has not yet covered the whole pool, do not warn on their own; the dashboard still shows them.

Alerts use Resend. TOKEN_ALERT_TO and TOKEN_ALERT_FROM are non-secret Worker variables in wrangler.jsonc; configure RESEND_API_KEY as a Worker secret:

aube exec wrangler secret put RESEND_API_KEY

Alerts are sent on warning/critical transitions, repeated after 12 hours while unhealthy, and followed by a recovery notification. The deploy workflow syncs the existing GitHub Actions RESEND_API_KEY secret into the Worker.