v2.1.0

A new advisory rule for the moment a push is about to run a slow gate
inside its own connection.

Added

• push-preflight, advising. A git push in a repository where amont
  runs a test gate at push time, on a tree that has not been rehearsed, is
  told to run amont run pre-push first. git opens its connection to the
  remote before pre-push and holds it idle for as long as the gate takes;
  a remote that closes idle sessions — Forgejo's own git timeout is six
  minutes — kills the push after the gate has already passed, and the
  failure reads as network. Measured 2026-09-04: three pushes in a row died
  that way, each paying the full suite, before a --no-verify retry of the
  already-attested tree went through — the bypass this rule exists to make
  unnecessary. amont ≥ 1.27 stamps the tree a passed gate ran against and
  amont run pre-push stamps HEAD with no connection open, so the push
  that follows skips the suite. examine fires on the shape of a push (not
  --dry-run, not --no-verify, not amont's own notes push); confirm
  stays silent unless amont guards the repository (an amont shim in
  hooks/pre-push), amont list --json --stage pre-push says a test gate
  runs, and refs/notes/amont-gate carries no pre-push-* token for
  HEAD's tree.

Removed

• packaging/amont-agent.rb, the seed used once to create the tap's formula.
  It has said version "0.0.0" with zero checksums ever since, while the
  real formula moved to 2.0.2 — a file that looks authoritative, is not, and
  drifts further with every release. Nothing referenced it.

  The tap is the single source for the formula, and scripts/bump-tap.py
  rewrites it on each release. amont keeps no seed either.

  What is still missing, and is the real gap: nothing verifies that the
  tap's formula can install what a release actually shipped. publish-tap
  runs ruby -c, which proves the file parses and nothing more. amont's own
  formula kept a bin.install "amont-agent" line for three releases after
  that binary left the archive, so brew install failed outright the whole
  time while every release went green — the checksums matched, the syntax
  was valid, and nobody ran brew. A stale copy in this repository would not
  have caught that; a post-publish brew install from the tap would.

---

What's Changed

• remove: the formula seed, which was drifting by @fredericrous in #3
• build: verify brew can install what was published by @fredericrous in #4
• build: verify attestations with attest by @fredericrous in #5
• fix(install): drop a loop that ran once by @fredericrous in #6
• docs: attest joins the Related section by @fredericrous in #7
• feat: push-preflight advises rehearsing the gate by @fredericrous in #8
• chore(release): 2.1.0 by @fredericrous in #9

Full Changelog: v2.0.2...v2.1.0