Split from #195, fix 2. Diagnose before patching — the first deliverable is a measurement, not a
code change.
On 2026-08-29, every architect message to an opencode builder was held with
no-composer-marker, meaning resolveBottomAnchorRegion could not bound a composer at all.
Holds of 3.5m, 8m and 12m; the 8-minute one never delivered. claude and codex builders in the same
workspace, same messages, delivered on the first attempt every time.
no-composer-marker is returned in exactly two cases (render-gate.ts): no row matches
rulePattern, or the row directly above the rule does not match bodyPattern.
OPENCODE_PROFILE (gate-profiles.ts:136) anchors on literal glyphs:
rulePattern: /^\s*╹▀{5,}/,bodyPattern: /^\s*┃/,topEdgePattern: /^\s*$/,minContentRows: 3,maxLookback: 20,Its comments record these as measured across seven states (idle, draft, mid-turn, dialog, boot,
/ picker, @ picker), so they were right when written. They are exact glyphs from one opencode
release, and opencode has shipped since.
Task 1 — capture, and report before changing anything
Dump real frames from a live opencode builder in at least: idle prompt, mid-turn, and with a
dialog open. Diff each against the four patterns above and report which pattern fails on which
frame.
Do not skip to a patch. Drift is the architect's hypothesis, explicitly not verified — nobody
has captured a current frame. If the patterns still match, the cause is elsewhere and this issue
needs re-diagnosing rather than a profile edit. Say so plainly if that is what you find; a wrong
diagnosis confirmed cheaply is a good outcome here.
Task 2 — fix what the capture actually shows
Only after task 1. If the profile drifted, correct it against the captured frames.
Task 3 — make the next drift loud
This is the part worth more than the fix. A profile that silently stops matching costs hours of
undelivered mail before anyone connects the two.
Commit the captured frames as fixtures and add a test that fails when a fixture stops resolving to
a bounded composer. Then an opencode release that changes the box breaks CI instead of breaking
delivery.
The fixture test must assert its own reach: it has to fail if pointed at a fixture directory
that does not exist or is empty. A zero-fixture run that reports success is the same defect one
level up, and this repo has hit that shape repeatedly (#190, #193, #194).
Consider doing the same for claude and codex while the harness is in hand — they work today, and
nothing would tell us when they stop.
Scope
Do not touch commands/spawn.ts or commands/spawn-worktree.ts — builder/spir-146 has
substantial changes there and will conflict.
Related: #195 (parent), #196 (sibling), #190.
Split from #195, fix 2. Diagnose before patching — the first deliverable is a measurement, not a
code change.
On 2026-08-29, every architect message to an opencode builder was held with
no-composer-marker, meaningresolveBottomAnchorRegioncould not bound a composer at all.Holds of 3.5m, 8m and 12m; the 8-minute one never delivered. claude and codex builders in the same
workspace, same messages, delivered on the first attempt every time.
no-composer-markeris returned in exactly two cases (render-gate.ts): no row matchesrulePattern, or the row directly above the rule does not matchbodyPattern.OPENCODE_PROFILE(gate-profiles.ts:136) anchors on literal glyphs:Its comments record these as measured across seven states (idle, draft, mid-turn, dialog, boot,
/picker,@picker), so they were right when written. They are exact glyphs from one opencoderelease, and opencode has shipped since.
Task 1 — capture, and report before changing anything
Dump real frames from a live opencode builder in at least: idle prompt, mid-turn, and with a
dialog open. Diff each against the four patterns above and report which pattern fails on which
frame.
Do not skip to a patch. Drift is the architect's hypothesis, explicitly not verified — nobody
has captured a current frame. If the patterns still match, the cause is elsewhere and this issue
needs re-diagnosing rather than a profile edit. Say so plainly if that is what you find; a wrong
diagnosis confirmed cheaply is a good outcome here.
Task 2 — fix what the capture actually shows
Only after task 1. If the profile drifted, correct it against the captured frames.
Task 3 — make the next drift loud
This is the part worth more than the fix. A profile that silently stops matching costs hours of
undelivered mail before anyone connects the two.
Commit the captured frames as fixtures and add a test that fails when a fixture stops resolving to
a bounded composer. Then an opencode release that changes the box breaks CI instead of breaking
delivery.
The fixture test must assert its own reach: it has to fail if pointed at a fixture directory
that does not exist or is empty. A zero-fixture run that reports success is the same defect one
level up, and this repo has hit that shape repeatedly (#190, #193, #194).
Consider doing the same for claude and codex while the harness is in hand — they work today, and
nothing would tell us when they stop.
Scope
Do not touch
commands/spawn.tsorcommands/spawn-worktree.ts—builder/spir-146hassubstantial changes there and will conflict.
Related: #195 (parent), #196 (sibling), #190.