Plan, don't mutate
`release` is dry-run by default — real mutation requires `dry_run: false`, the same `--yes` a human would give.
Changelog → semver bump → tag → publish — one ops wire for humans, agents, and campaigns.
Tarsi is the org's release plane: changelog → semver bump → tag → publish, exposed as tarsi-serve ndjson ops. Repo config is tarsi.toml; org conventions ride type: tarsi tamias profile packs; there is deliberately no tarsi CLI.
Every repo was reinventing release plumbing — changelog rendering, commit→bump inference, tag schemes, publish ordering. Tarsi makes release semantics a shared, deterministic ops surface that humans, agents, and peregrin campaigns all drive identically, with dry-run as the default so probing never mutates.
| You want to… | Go to |
|---|---|
| Cut a real release end to end | Tutorial |
| Do a task (write tarsi.toml, author a profile, drive ops) | How-to |
| Understand the design (ops wire, semantics, boundaries) | Explanation |
| Look up an op, config field, or pack layout | Reference |