diff --git a/.claude/skills/release/SKILL.md b/.claude/skills/release/SKILL.md
index 29ce7b6f..382a1986 100644
--- a/.claude/skills/release/SKILL.md
+++ b/.claude/skills/release/SKILL.md
@@ -16,5 +16,21 @@ tagging, run `bash tools/changelog_fragments.sh cut <version> <YYYY-MM-DD>`, whi
 section, bumps `VERSION` and deletes the fragments, and commit that as the cut PR (`changes/README.md`).
 Check `git tag` for published versions. The front-door docs do not carry a copied latest-version line.
 
-Homebrew tap: github.com/InauguralSystems/homebrew-eigenscript (tracks the
-latest release).
+The Homebrew tap at github.com/InauguralSystems/homebrew-eigenscript must
+track the latest release automatically.  Its `bump-formula.yml` workflow runs
+on a schedule and can also be started with `workflow_dispatch`.  It reads the latest
+EigenScript release from GitHub's public API, downloads that tag's source
+tarball, computes the formula's `sha256` from the downloaded bytes, and opens a
+formula-bump PR when the formula is behind.  Do not copy a checksum into the
+formula by hand and do not add a cross-repository credential to this release
+workflow.
+
+After a release, look in the tap's pull requests for the automated formula
+bump and its `brew test-bot` result. GitHub may require a maintainer to approve
+the test-bot runs on a PR created with `GITHUB_TOKEN`; approve those runs on
+the generated PR before waiting for its formula checks. A manual dispatch of
+the test-bot runs only setup/syntax checks and does not replace PR formula checks.
+A red PR is repaired in the tap; do not
+paper over it by declaring the release complete here.  If no PR appears after
+the scheduled run, dispatch `bump-formula.yml` in the tap and inspect that
+workflow's log.
diff --git a/changes/documentation/1297-homebrew-tap-automation.md b/changes/documentation/1297-homebrew-tap-automation.md
new file mode 100644
index 00000000..8a488237
--- /dev/null
+++ b/changes/documentation/1297-homebrew-tap-automation.md
@@ -0,0 +1,2 @@
+- Document the Homebrew tap's automatic formula-bump workflow, including its
+  computed checksum, test-bot gate, and manual recovery path.
diff --git a/tools/release_skill_check.sh b/tools/release_skill_check.sh
new file mode 100755
index 00000000..4cc0f853
--- /dev/null
+++ b/tools/release_skill_check.sh
@@ -0,0 +1,45 @@
+#!/usr/bin/env bash
+# Keep the release runbook explicit about the independently automated tap.
+set -eu
+cd "$(dirname "$0")/.."
+
+check_file() {
+    file=$1
+    failed=0
+    for phrase in \
+        'track the latest release automatically' \
+        'bump-formula.yml' \
+        'workflow_dispatch' \
+        "GitHub's public API" \
+        'computes the formula' \
+        'sha256' \
+        'brew test-bot' \
+        'dispatch `bump-formula.yml`'
+    do
+        if ! grep -Fq "$phrase" "$file"; then
+            echo "release_skill_check: RED: missing required guidance: $phrase"
+            failed=1
+        fi
+    done
+    return "$failed"
+}
+
+if [ "${1:-}" = --selftest ]; then
+    # Check the actual runbook before requiring an edited copy to fail.
+    # Otherwise a pre-existing omission makes the negative control vacuous.
+    check_file .claude/skills/release/SKILL.md
+    scratch=$(mktemp "${TMPDIR:-/tmp}/release-skill.XXXXXX")
+    trap 'rm -f "$scratch"' EXIT HUP INT TERM
+    cp .claude/skills/release/SKILL.md "$scratch"
+    sed 's/brew test-bot/brew verification/' "$scratch" > "$scratch.mutant"
+    mv "$scratch.mutant" "$scratch"
+    if check_file "$scratch" > /dev/null 2>&1; then
+        echo 'release_skill_check selftest: RED: missing test-bot guidance was accepted'
+        exit 1
+    fi
+    echo 'release_skill_check selftest: PASS: missing test-bot guidance rejected'
+    exit 0
+fi
+
+check_file "${1:-.claude/skills/release/SKILL.md}"
+echo 'release_skill_check: PASS'
diff --git a/tools/selftests.txt b/tools/selftests.txt
index 099c9f0f..eb602c2f 100644
--- a/tools/selftests.txt
+++ b/tools/selftests.txt
@@ -35,6 +35,7 @@ tools/obs_marker_check.sh | timeout 30 bash tools/obs_marker_check.sh --selftest
 tools/obs_reader_sync_check.sh tools/obs_marker_check.sh | timeout 30 bash tools/obs_reader_sync_check.sh --selftest | [{"pattern":"^SELFTEST: ([0-9]+) passed.*$","count":11,"floor":true,"value":true}]
 tools/performance_observer_docs.py docs/PERFORMANCE.md bench/observer_callgrind.txt bench/baseline.txt bench/observed_loop.eigs bench/unobserved_loop.eigs bench/conservative_observed_loop.eigs bench/conservative_unobserved_loop.eigs | timeout 30 python3 tools/performance_observer_docs.py --selftest
 tools/replay_diff.sh | timeout 30 bash tools/replay_diff.sh --selftest | [{"pattern":"  selftest ok:","count":11},{"pattern":"^SELFTEST: all planted faults caught","count":1}]
+tools/release_skill_check.sh .claude/skills/release/SKILL.md | timeout 30 bash tools/release_skill_check.sh --selftest | [{"pattern":"^release_skill_check selftest: PASS:","count":1}]
 tools/section_plan.sh tests/suite_plan.sh .github/workflows/ci.yml | timeout 1860 bash tools/section_plan.sh --selftest
 tools/section_fragments_check.sh tools/runner_text.sh tools/section_plan.sh tools/suite_label_check.sh tools/enrolment_check.sh tools/child_exit_check.sh tests/run_all_tests.sh tests/sections/* | timeout 60 bash tools/section_fragments_check.sh --selftest
 tools/stop_gate.sh | timeout 30 bash tools/stop_gate.sh --selftest | [{"pattern":"^stop_gate selftest: committed-fragment PASS; no-fragment BLOCK$","count":1}]