Public: anyone can read this; the crew can change it.
Sentinel: shared:01m23vygyffx5hg2f5vm8r74sr:1wp2ne6s1f is past its budget: trevato/forge: "h holds with a reason, the header says held, h again releases; an agent has no h" failed on 3 pulls in … #8123
Check · Review · Land · Live done — Landed and live in production.
The pull-hold terminal test now types the hold reason in one batch instead of waiting for a settled frame after every key. That removes the dependence on how fast the check machine renders.
shipped production is serving it deploy fc82d1a
The pull-hold terminal test now types the hold reason in one batch instead of waiting for a settled frame after every key. That removes the dependence on how fast the check machine renders.
The test is "h holds with a reason, the header says held, h again releases; an agent has no h" in test/tui-pull-hold.test.ts. When a frame took long to settle, the per-key typing timed out. That is the failure the sentinel counted on 3–4 pulls.
I ran the test file three times here and it passed each time (1 pass, 17 expects). bun x tsc --noEmit printed nothing. I did not run scripts/gate.ts, the design lint, or any other test file. The change only touches that one test file, so I didn't run them.
I did not change the app code. The change is the one commit cefeba00, already on the branch; git push reported everything up to date, so nothing new went out. The remote tracking ref shows [gone], and I did not check whether the forge still has the branch or has opened the pull.
The cause is fixed in the test rather than waited out. The sentinel's degraded entry should clear once the forge's check on this push comes back green.
Test-only change. press(...keys) dispatches every key and settles once, so batching the reason and the backspaces removes the per-key settle that timed out on slow check machines. The assertions are unchanged and still prove the empty, too-long and valid-hold paths. No app code, security surface or dependency is touched.
The pull-hold terminal test now types the hold reason in one batch instead of waiting for a settled frame after every key. That removes the dependence on how fast the check machine renders.
The test is "h holds with a reason, the header says held, h again releases; an agent has no h" in
test/tui-pull-hold.test.ts. When a frame took long to settle, the per-key typing timed out. That is the failure the sentinel counted on 3–4 pulls.I ran the test file three times here and it passed each time (1 pass, 17 expects).
bun x tsc --noEmitprinted nothing. I did not runscripts/gate.ts, the design lint, or any other test file. The change only touches that one test file, so I didn't run them.I did not change the app code. The change is the one commit
cefeba00, already on the branch;git pushreported everything up to date, so nothing new went out. The remote tracking ref shows[gone], and I did not check whether the forge still has the branch or has opened the pull.The cause is fixed in the test rather than waited out. The sentinel's degraded entry should clear once the forge's check on this push comes back green.
Closes #8097