2026-07-18 17:34:34 +00:00
# Labels
How this repo uses GitHub labels. The taxonomy is shared across the
heavy-duty repos (box, rig, cast) — only the `scope:` set differs per repo,
because it names this repo's actual surfaces.
## State — who is the ball with? (PRs, exactly one)
Every open PR carries exactly one `state:` label, and it answers the only
question a board scan actually asks: *who is this PR waiting on?* The states
mirror the review loop this repo runs — PRs open as drafts, three reviewer
bots pick up ready PRs with reviews requested, each round is answered in a
single reply, and a human takes the final review.
| Label | Color | Waiting on | Enters when | Leaves when |
|---|---|---|---|---|
| `state:building` | `#FBCA04` | the coding agent, still building | PR opened as draft | marked ready + bot reviews requested |
| `state:bots-reviewing` | `#1D76DB` | the reviewer bots to finish the round | ready with reviews requested, or fixes pushed and reviews re-requested | all three bots have reviewed the round |
2026-07-20 17:50:00 +00:00
| `state:addressing` | `#D93F0B` | the coding agent to reply, fix, or ask | all bots reviewed and not all approved; or nobody was asked; or a blocker is up | the round-reply is posted and fixes pushed — and any blocker named alongside is cleared |
2026-07-20 17:27:06 +00:00
| `state:needs-human` | `#8250DF` | the human reviewer | the PR **could be merged right now** : no blockers, three formal head-current approvals — and the human review is requested | merged — or changes requested, which cycles back to `state:addressing` |
2026-07-18 17:34:34 +00:00
`bots-reviewing` and `addressing` are deliberately distinct: staleness in the
first means *poke the bots* , staleness in the second means *the agent dropped
the ball*. Collapsing them loses exactly the information a sweep needs.
2026-07-20 17:27:06 +00:00
`bots-reviewing` therefore means strictly *a request is live and an answer is
coming* — a PR nobody was asked to review is the agent's ball, not the bots'.
## The second axis: `blocker:*`
State answers *whose ball is it* . Blockers answer *what is in the way* , and
unlike states they are *facts about the branch* — mutually independent, so a
PR carries as many as apply.
| Label | Color | Means | Clears when |
|---|---|---|---|
| `blocker:conflict` | `#B60205` | GitHub says `CONFLICTING` — the agent owes a **rebase** | it merges cleanly |
| `blocker:ci-red` | `#B60205` | a check failed — the agent owes a **fix** , which a rebase will not provide | checks are green |
2026-07-20 17:50:00 +00:00
| `blocker:unrequested` | `#E99695` | this head has no verdict from somebody — never reviewed, or staled by a push — and **nobody was asked** for one | reviews are requested |
refactor: one drill record per version, in drills/
Drill records move from sections inside drill/RUNS.md to one file per
version: drills/<version>.md. drill/RUNS.md is untouched — it stays the
harness's own run log, traps table and lore. drills/ is release evidence
only, and the docs now say which is which.
The old guard parsed headings: em-dash field matching, an optional
' — DATE' tail, whole-version comparison so 0.9.0-rc1 could not satisfy
0.9.0, avoiding \x escapes because CI runs mawk not gawk, and a non-blank
body rule. Every one of those existed only because records shared one
file, and two separate defects were found in review because of that
complexity — the sed '/./,$!d' whitespace bypass, and heading-grammar
drift from the sibling repos.
One file per version makes almost all of it unrepresentable: 0.9.0.md and
0.9.0-rc1.md are simply different files, so whole-version matching is
free rather than a trap. The guard keeps only the rule that was never
about headings — a record of pure whitespace is not a record — and keeps
the failure message that names the version, the path, the unblock, and
the recorded-waiver escape hatch.
Plain drills/, not .drills/ — a dot-directory is invisible to globs
without dotglob, which is what caused #116 and #118.
Also corrects the release-drill framing: the three repos' drills are
INDEPENDENT and run in any order. What dissolves the box<->rig recursion
is that every drill pins the same fixed candidate refs, not sequencing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-21 16:26:47 +00:00
| `blocker:drill-pending` | `#B60205` | a `release` PR whose version has no drill record at [`drills/X.Y.Z.md` ](drills/README.md ) — the ceremony is **correct but unevidenced** | the drill is run and recorded (or a waiver is recorded) at `drills/X.Y.Z.md` |
feat: CI refuses a release PR with no drill record
CONTRIBUTING.md has said since #96 that the release PR is where the full
real-hardware drill hangs, recorded in drill/RUNS.md. No release has ever
done it: #95, #114 and #148 all shipped as a VERSION bump plus a CHANGELOG.md
stamp, and RUNS.md carries no '## Release drill' section at all. A reviewer
bot finally blocked on it — which is the point. The one time it was caught is
the one time somebody happened to look, and that is not a gate.
So the rule moves out of the document and into CI.
.github/scripts/drill-recorded.sh, keyed on VERSION the same way
changelog-armed.sh is: a -dev tree passes with nothing to assert (which is
what keeps the guard installable — a version that fired on every PR would be
switched off inside a day), and a bare VERSION must carry a section headed
'## Release drill — <version>', optional ' — <date>' tail, with prose under
it. The version is compared as a whole awk field, never as a substring, so
0.9.0 cannot be satisfied by a 0.9.0-rc1 drill or vice versa —
release-notes.sh's trap, solved the same way so the two cannot disagree
about what "the section for X" means.
What it asserts is a RECORD, deliberately, not a passing drill: CI cannot run
the drill (real hardware, the better part of an hour — ci.yml says as much
about the rehearsal job it runs instead). That also keeps the maintainer
waiver honest — a release that must ship undrilled writes that under the same
heading, so the skip is a reviewable line in the diff rather than silence.
Wired into ci.yml as its own step, NOT pull-request-only, for the reasoning
#143 applied to the monotonic guard: the merge that publishes a release is a
push to main carrying the same bare VERSION, so a PR-only check would leave
the tree that actually ships unasserted.
test/release.sh grows 27 cases (134 -> 161). Every fixture carries its own
VERSION and its own RUNS.md — reaching for $ROOT/VERSION is the coupling #146
had to fix, and it goes red on the ceremony tree, the one tree where the
release suite most needs to be trustworthy.
CONTRIBUTING.md now states the flow (draft -> ready -> bot round -> drill ->
state:needs-human -> merge), the heading format, that three releases shipped
through the gap, and the recorded-waiver escape.
It also describes the drill as ONE orchestrated run over the whole stack,
because box and rig are mutually recursive and cannot be linearly ordered:
rig sits below box as the host-builder ('rig bootstrap --host yes' installs
box and runs setup-host) and above it as the guest-converger (a box new seed
curls rig's installer and runs 'rig bootstrap <tenant>-box'), the inverted
edge bin/box already documents as rig#28. The run is host bootstrap -> box
new -> tenant converge -> cast. It drills CANDIDATE REFS, not released
artifacts: RIG_REPO/RIG_REF are mint-time environment variables defaulting
to heavy-duty/rig@main (bin/box:1116-1117), so a run pins the exact commits
under test and no repo must be released before another can be drilled.
Drilling the candidate is drilling the release — a release diff is VERSION +
CHANGELOG.md, so nothing executable differs. One run, one shared run ID; each
repo records its own legs citing that ID and the other two SHAs, and the
guard reads only this repo's file.
Recorded as a known gap, not fixed here: a released box still defaults
RIG_REF to main, so a box minted a week after a drill is not the drilled
combination. Pinning RIG_REF to a released rig tag in the templates is the
outstanding step from #81 (rig#32 step 5).
LABELS.md documents blocker:drill-pending — ceremony correct but unevidenced,
maintainer-created because the bot account gets a 403 on label creation, with
`blocked` standing in until it exists.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-21 15:20:37 +00:00
`blocker:drill-pending` is the one blocker that says the diff is *fine* . The
version is stamped, the changelog is right, CI's other guards are green — what
is missing is the evidence that the release was proven on real hardware, which
[.github/scripts/drill-recorded.sh ](.github/scripts/drill-recorded.sh ) refuses
to let a release ship without (CONTRIBUTING.md, "Releases"). Naming it
separately from `blocker:ci-red` matters because the two owe different work: a
red check is a fix in the branch, a pending drill is an hour on a real host.
It must be **created by a maintainer account** — the bot account gets a `403`
creating labels, so the labels workflow's bootstrap dispatch cannot mint it.
Until it exists in the repo, `blocked` stands in: it is the closest true thing
(the PR is waiting on something outside itself) and, usefully, the staleness
sweep already skips it, so a release parked overnight on a drill does not
collect a `stale` .
2026-07-20 17:27:06 +00:00
One rule joins the axes: ** `state:needs-human` requires zero blockers.** Any
blocker means the work is the agent's, whatever the review round says.
This split exists because the single-label version kept lying. Independent
facts were projected onto one totally-ordered label, so one always had to win
and the losers vanished off the board: a PR that was *both* conflicted and red
could only say one of them, and `needs-rebase` told an agent to rebase when
what it actually owed was a bug fix. Precedence between two blockers is not a
question a set has to answer, which is why every ordering bug this machine has
had — `needs-human` surviving a conflict, `MISSING` swallowing `STALE` — lived
on the axis that had to be totally ordered.
`state:needs-rebase` was the first attempt at this and is **retired** ; the
reconciler strips it on sight so no PR is left carrying a label nothing
recomputes.
2026-07-18 17:34:34 +00:00
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
**`state:needs-human` means one thing: a human could merge this right now.**
2026-07-20 17:27:06 +00:00
The label is the only signal a maintainer scanning the board (or a phone)
actually reads, and one that says "your turn" on an unmergeable PR is worse
than no label at all. So beyond the blockers, one review fact also outranks an
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
explicit human request:
- **nobody reviewed *this* head** — every approval staled by a push → `state:addressing` ,
because the agent owes a re-request
2026-07-20 17:27:06 +00:00
That case is more dangerous than any blocker: a blocked PR at least shows an X
or a disabled merge button, while a staled-approval PR reads green, mergeable
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
and "waiting on the human" over code no reviewer has seen.
2026-07-20 17:27:06 +00:00
`UNKNOWN` mergeability is deliberately **not** treated as a conflict. GitHub
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
reports it for about a minute after every merge while it recomputes, and
2026-07-20 17:27:06 +00:00
flapping every open PR through `blocker:conflict` on each merge would be worse
than the bug this fixes. A failed read of either branch fact degrades to the
same "do not know" value, for the same reason.
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
An *unfinished* round still yields to an explicit human request — a maintainer
pulling a PR to themselves early is a deliberate act. `MISSING` (nobody has
reviewed yet) and `STALE` (everyone reviewed something else) are different
facts and are treated differently.
2026-07-18 17:34:34 +00:00
## Cross-cutting (PRs and issues)
| Label | Color | Meaning |
|---|---|---|
| `stale` | `#B60205` | No activity for 48h. Sweep-managed, never hand-applied. `state:building` + `stale` is precisely a forgotten draft. |
| `blocked` | `#6A737D` | Waiting on another PR or issue to land first. Quiet *legitimately* — the staleness sweep skips it. |
| `release` | `#0E8A16` | Release flow, versioning, and packaging work. |
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
| `merge-next` | `#0E8A16` | Head of the merge queue — **merge this one next** . Queue order is *intent* (which PR lands first, given how they conflict), so the reconciler never sets it: you or the agent maintaining the queue do. The reconciler only **clears** it, the moment the PR stops being something a human could merge — so it cannot go stale the way `state:needs-human` did. |
2026-07-18 17:34:34 +00:00
## Scope — which surface? (PRs and issues, any number)
All scopes share one calm color, `#C5DEF5` — scopes locate, states alert.
| Label | Covers |
|---|---|
| `scope:cli` | `bin/box` — the command surface itself |
| `scope:installer` | `install.sh` , the versioned install layout, upgrade/uninstall |
| `scope:host` | `host/` — setup-host, teardown, the firewall and isolation stack |
| `scope:tiers` | the restricted tier — grant/revoke, multi-user semantics |
| `scope:templates` | `templates/` — the box seeds |
| `scope:drill` | `drill/` — the rehearsals, doctor, RUNS.md |
## Issue types
`bug` , `enhancement` , `documentation` — issues only. PRs carry their type in
the conventional title (`feat:`, `fix:` , `docs:` ), so typing a PR with a label
would just say the same thing twice, drifting apart eventually.
## Maintenance
fix(labels): sweep on `labeled` so the handoff is immediate
A review landing was never a trigger for the labels workflow, so the exact
moment `state:needs-human` became true — the third bot approving — fired
nothing, and the label waited on the `*/15` cron. That cron does not run at
its declared rate: measured across box, rig and cast over a two-hour window
on 2026-07-20, one scheduled run each against the eight `*/15` implies.
The obvious fix does not work. There is no `pull_request_review_target`, and
on fork PRs — all of them here — `pull_request_review` runs with a read-only
token and cannot label anything.
So the handoff wakes the sweep itself:
- `pull_request_target` also fires on `labeled`/`unlabeled`
- the author sets `state:needs-human` at handoff, as the third act after the
round summary and the review request
The author's own label write fires the sweep that validates it — an
optimistic write, not a transfer of ownership. The reconciler confirms or
corrects it seconds later, and the cron falls back to a last resort. It
cannot loop: the reconciler writes with GITHUB_TOKEN, which does not create
workflow runs; agent writes use a PAT, which does.
`labels-reconcile.sh` is unchanged — it already recomputes every open PR
from scratch on every run, which is what makes the optimistic write safe.
The `scope` job is skipped on label events, where no path can have changed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 19:44:07 +00:00
State labels are machine-owned, with exactly one exception. Every state above
is derivable from GitHub's own facts — the draft flag, requested reviewers,
2026-07-18 18:31:51 +00:00
review states, push timestamps — so the labels workflow
([.github/workflows/labels.yml](.github/workflows/labels.yml)) recomputes the
fix(labels): sweep on `labeled` so the handoff is immediate
A review landing was never a trigger for the labels workflow, so the exact
moment `state:needs-human` became true — the third bot approving — fired
nothing, and the label waited on the `*/15` cron. That cron does not run at
its declared rate: measured across box, rig and cast over a two-hour window
on 2026-07-20, one scheduled run each against the eight `*/15` implies.
The obvious fix does not work. There is no `pull_request_review_target`, and
on fork PRs — all of them here — `pull_request_review` runs with a read-only
token and cannot label anything.
So the handoff wakes the sweep itself:
- `pull_request_target` also fires on `labeled`/`unlabeled`
- the author sets `state:needs-human` at handoff, as the third act after the
round summary and the review request
The author's own label write fires the sweep that validates it — an
optimistic write, not a transfer of ownership. The reconciler confirms or
corrects it seconds later, and the cron falls back to a last resort. It
cannot loop: the reconciler writes with GITHUB_TOKEN, which does not create
workflow runs; agent writes use a PAT, which does.
`labels-reconcile.sh` is unchanged — it already recomputes every open PR
from scratch on every run, which is what makes the optimistic write safe.
The `scope` job is skipped on label events, where no path can have changed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 19:44:07 +00:00
state and reconciles labels statelessly, on PR events (label changes included)
plus a 15-minute cron. A hand-moved label is a lie waiting to happen; the
workflow asserts the effective state instead.
The exception is `state:needs-human` , which the author sets at handoff
([CONTRIBUTING.md](CONTRIBUTING.md), step 6). That is an optimistic write, not
a transfer of ownership: because `pull_request_target: labeled` wakes the
workflow, the author's own label write fires the sweep that validates it, and
a handoff that had not earned the label is corrected within seconds.
It exists because the wake signal was missing. There is no
`pull_request_review_target` — on fork PRs, which is all of them here,
`pull_request_review` runs read-only and cannot label anything — so the moment
the label becomes true, the third approval landing, fired nothing at all. What
was left was the `*/15` cron, and GitHub deprioritises short intervals hard
enough that the delivered rate is closer to hourly. The label could therefore
lag the round it described by hours, worst on the quietest repo: every sweep
reconciles the whole board, so a busy repo stays fresh by piggybacking on
unrelated PR events, while a quiet one depends on the cron most and receives
it least. `scope:` labels on PRs are applied from the changed
2026-07-18 18:31:51 +00:00
paths by actions/labeler ([.github/labeler.yml](.github/labeler.yml));
[CONTRIBUTING.md ](CONTRIBUTING.md ) says who sets what.
The same workflow bootstraps the taxonomy: a manual dispatch creates any
missing label idempotently. To create them by hand (needs push access):
2026-07-18 17:34:34 +00:00
```sh
gh label create "state:building" --color FBCA04 --description "PR is a draft — the coding agent is still building" --force
gh label create "state:bots-reviewing" --color 1D76DB --description "Waiting on the bot reviewers to finish the round" --force
gh label create "state:addressing" --color D93F0B --description "All bots reviewed — coding agent owes the single reply + fixes" --force
2026-07-20 17:27:06 +00:00
gh label create "blocker:conflict" --color B60205 --description "Does not merge — the branch conflicts and the agent owes a rebase" --force
gh label create "blocker:ci-red" --color B60205 --description "A check is failing — the agent owes a fix (not a rebase)" --force
gh label create "blocker:unrequested" --color E99695 --description "Somebody still owes a verdict and nobody was asked for one" --force
# retired — the reconciler strips it; delete it once no PR carries it
# gh label delete "state:needs-rebase"
gh label create "state:needs-human" --color 8250DF --description "No blockers, all bots approve — waiting on the human reviewer" --force
fix(labels): state:needs-human means a human could merge it right now
decide_state() derived state from three inputs -- draft flag, requested
reviewers, submitted reviews -- and read NOTHING about mergeability or checks.
Combined with the `if requested "$HUMAN"` short-circuit at the top of its
precedence, the label was sticky: once the maintainer was requested, the PR
read state:needs-human through conflicts, through red CI, through a
force-push that staled every approval. Nothing demoted it.
Observed twice in one afternoon, in two different shapes. Three PRs sat at
state:needs-human while CONFLICTING for hours -- the board inviting a merge
GitHub had already disabled. And #119, after a rebase, read MERGEABLE, four
green checks, state:needs-human, with ZERO reviews bound to its head: every
visible signal saying "merge me" over a tree no reviewer had seen. That second
shape is the dangerous one, because unlike a conflict nothing on the page
contradicts it.
The rule the label now keeps: state:needs-human means a human could merge this
RIGHT NOW, so anything making that false outranks the request that put it
there.
CONFLICTING or failing checks -> state:needs-rebase (new; the agent's to fix)
approvals staled by a push -> state:addressing (nobody reviewed this tree)
An UNFINISHED round still yields to an explicit human request -- a maintainer
pulling a PR to themselves early is deliberate, and MISSING (nobody has
reviewed yet) is a different fact from STALE (everyone reviewed something
else). That distinction is why the two are handled in different arms rather
than collapsed.
UNKNOWN mergeability is deliberately NOT treated as unmergeable: GitHub
reports it for about a minute after every merge while it recomputes, and
flapping every open PR through needs-rebase on each merge would be worse than
the bug being fixed. A failed read of either fact degrades to the same "do not
know" value, for the same reason -- an API hiccup must not relabel the board.
Also adds merge-next, because a correct needs-human still does not say WHICH
PR to merge first, and order matters when they conflict through CHANGELOG.md.
Queue order is intent, so the reconciler never sets it; it only CLEARS it the
moment the PR stops being mergeable-by-a-human -- precisely the staleness that
made needs-human untrustworthy.
Both live shapes are pinned in test/labels-reconcile.sh (19 -> 29 fixtures),
including that UNKNOWN does not trigger needs-rebase and that a draft outranks
a conflict. Proven non-vacuous: dropping the mergeability arm fails 4
assertions, dropping the STALE precedence fails 2, restoring returns 29/0.
Closes #136
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 15:18:49 +00:00
gh label create "merge-next" --color 0E8A16 --description "Head of the merge queue — merge this one next (set by hand/agent, cleared here)" --force
2026-07-18 17:34:34 +00:00
gh label create "stale" --color B60205 --description "No activity for 48h — needs a poke (sweep-managed)" --force
gh label create "blocked" --color 6A737D --description "Waiting on another PR or issue to land first" --force
gh label create "release" --color 0E8A16 --description "Release flow and version/packaging work" --force
gh label create "scope:cli" --color C5DEF5 --description "bin/box — the command surface" --force
gh label create "scope:installer" --color C5DEF5 --description "install.sh, versioned installs, upgrade/uninstall" --force
gh label create "scope:host" --color C5DEF5 --description "host/ — setup, teardown, firewall, isolation stack" --force
gh label create "scope:tiers" --color C5DEF5 --description "restricted tier — grant/revoke, multi-user" --force
gh label create "scope:templates" --color C5DEF5 --description "templates/ — the box seeds" --force
gh label create "scope:drill" --color C5DEF5 --description "drill/ — rehearsals, doctor, RUNS.md" --force
2026-07-18 18:04:49 +00:00
# delete is not an upsert: a label that is already gone exits non-zero. Swallow
# that, so this block converges on re-run instead of erroring after first success.
for L in duplicate invalid question wontfix "help wanted" "good first issue"; do
gh label delete "$L" --yes 2>/dev/null || true
done
2026-07-18 17:34:34 +00:00
```
