The role-template registry moves out of rig's tree (heavy-duty/rig-templates,
ruled 2026-07-24: pinned to the rig version by default, overridable per mint).
This is the schema-and-resolution half rig keeps: RIG_TEMPLATES_DIR >
RIG_TEMPLATES_REF > the in-tree RIG_TEMPLATES_PIN (the BOX_RELEASE
discipline), template.env parsed against an allowlist and never sourced,
every refusal naming the failing key, and the lint the registry repo's CI
will run on every definition.
The pin currently names the registry's pre-seed head; it bumps to the seeded
tree in this PR's course (the seed PR is the other half of the build task).
install from this checkout via RIG_INSTALL_SOURCE, assert what landed,
a converging re-run proven by an empty diff, uninstall --all ending in
the absence assert (! -e AND ! -L). test/install-lifecycle.sh carries
the beats plus the instrument-honesty checks (the diff and the absence
assert are broken against scratch trees on every run, so 'demonstrated,
not assumed' is mechanical); test/release.sh gains the local channel's
contract — dir, tarball, loud refusal, never a silent download fallback.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Both fixture digests are computed outside the implementation, so a refactor
that changes the prefix, the hash or the slicing renames the whole fleet and
fails here. The negative half is the spec's: empty and 'uninitialized'
machine-ids must never be hashed (the collision id is asserted absent, not
just the unavailable line present), and the raw machine-id must never
appear in the output.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Option A per triage's normalization on #95: derive, never mint. The
derivation is pinned — sha256("rig-machine-id:<machine-id>"), first 32 hex
as 8-4-4-4-12 — and the prefix is the contract that keeps the id
uncorrelatable with other tools' derivations, per machine-id(5)'s own
guidance not to expose the raw value.
Missing, empty and 'uninitialized' machine-id files degrade loudly to an
'unavailable (reason)' line: hashing nothing would hand every such machine
the same identity, the worst possible failure for an identity field.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The round's one blocking point, from all three reviewers: drill.sh:387
still said the exported pin beats "box's default (main)", the exact
expired claim #133 exists to remove — bootstrap's unset path has been
BOX_REF="${BOX_REF:-$BOX_RELEASE}" since rig#103 landed. Rewritten like
the other two spots: the why survives, the fallback is named as the
BOX_RELEASE pin, main is gone. The pre-flight hazard-class block
(302-306) stays untouched per the spec; the wide grep
(default.*main|main.*default over drill/, drills/, README.md) now finds
only that protected block.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
drill/README.md and drill.sh's assert_installed_from comment claimed both
installers default to main 'until #103 lands'. #103 landed (box installs
the BOX_RELEASE pin), and rig's install.sh has resolved the latest release
since #32 — six days before the README was written. The behaviour was
always right; the reason given for it expired, so both places now carry
the durable reason: a sane default is what makes an unstated ref
invisible, and the clean-looking record it produces proves nothing about
the candidate.
The --box-ref example (three copies: README, file header, --help — the
header IS the --help via sed 2,33p, so the window's line count is
preserved) moves from release/0.10.0 to the 0.9.0 tag: after #103 the box
that ships is the BOX_RELEASE tag, so a release/… branch is the wrong pin
for box while staying right for rig. The hazard-class citations at the
pre-flight refusal are untouched — they cite #103 as the hazard, not as a
pending change.
Closes#133
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The runner leg reads the newest run ID before dispatching and only
judges a run with a different ID — workflow_dispatch takes seconds to
materialize a run, and the previous run's 'completed' was one poll away
from being read as ours. --help's sed range stops where the header does.
(ceremony flow: issue #105)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
drill/README.md is the repeatable procedure #107's second checkbox asks
for: prerequisites (the throwaway machine, the tagged key, the fork's
drill workflow, the pins), the invocation, what each leg asserts, where
the record lands. drills/README.md's harness disclaimer flips to point
at the instrument, its legs list and example record match what
drill.sh actually runs and emits, and CONTRIBUTING's drill sentence
names the script. Changelog entry under Unreleased.
(ceremony flow: issue #105)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
test/drill.sh awk-extracts the harness's decision functions (the
release.sh pattern) and drives them against fixtures: the ref refusal
names both refs, a loud skip never classifies as a pass, the idempotence
verdict is a real diff that goes non-empty when convergence is broken —
demonstrated mechanically on every CI run — and the record emitter
cannot produce a clean-sweep reading over a skipped leg. CI runs it in
the check job.
The tests caught three real harness bugs before any reviewer could:
printf eating a '- '-leading format as options (a silently empty Failed
section — the exact lie the record exists to prevent), tree_of trusting
GNU readlink -f's exit 0 on a dangling final component, and the arg
refusals sitting behind the root check in violation of the repo's own
validated-before-root doctrine.
(ceremony flow: issue #105)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Box's shape, rig's legs: the reporting verbs and set -u-only discipline
(a failing check is data, not a crash), a fatal INSTALLED_FROM assertion
on BOTH pinned refs before anything is believed, convergence asserted on
effective state, idempotence decided by a mechanical capture-and-diff,
db driven through test/db-integration.sh with its loud-skip contract
kept, the runner lifecycle against a fork, a pinned coolify install, and
a record emitter that writes drills/<version>.md in the schema the
drill-recorded gate reads — skips counted and named, never folded into
passes.
The --host yes leg stops at 'the pinned box installed and its host stack
stands', in as many words in the output: the isolation boundary is box's
drill's assertion, joined to this record by the shared run ID.
(ceremony flow: issue #105)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The ceremony conversion (#112) extracted the panel= line verbatim from the
pre-ceremony labels-reconcile.sh BOTS array, which predates kimi-bot joining
the family bench. A name missing from that line is a reviewer the machinery
never waits for: the handoff would report a full panel while one verdict
short — the exact defect cast@2612967 fixed after cast#143 shipped it.
labels.conf is the source of truth the reconciler reads; CONTRIBUTING
mirrors it for humans requesting reviewers by hand. Both move together.
Closes#120
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Root LABELS.md was rig's own copy of the family taxonomy doc; the
vendored .ceremony/LABELS.md is the driftless replacement, so the copy
goes. drills/README.md's gate pointer follows the guard upstream.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
changelog-armed returns (rig#44's revert, now version-keyed upstream);
docs-sync guards the doctrine mirror the next commit vendors.
test/release.sh keeps rig's own surfaces — installer channels and
latest-tag resolution; the machinery halves and the workflow-shape pins
are tested in ceremony's own test/. test/labels-reconcile.sh goes whole:
it drove the deleted reconciler.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The conf format takes no comment lines: every non-blank, non-panel line
must parse as name|color|description, so the file is data only.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The fourth agent tenant, through the one parameterized mechanism
(lib/tenant-config.sh + bootstrap-tenant.sh) — never a fifth
hand-maintained copy. The official installer (code.kimi.com/install.sh)
is uv-managed and lands kimi in ~/.local/bin, run as the tenant user
(grok's lesson: a root-owned install under a 0700 home is a CLI that
exists and cannot run); no node, uv brings its own Python. The context
file lands at ~/.kimi/AGENTS.md — the <dotdir>/AGENTS.md convention —
with an honest note that kimi documents only project-level AGENTS.md
today.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
