You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
[bug] Stage readout in PROJECT.md goes stale after the last gate pass, and late asks skip change-intake #767
Field report (Greenfield build, auto mode, 2026-11-13; relayed by the maintainer, anonymised).
What happened:
After the build was finished, PROJECT.md still said "Stage P — needs attention", because gate-check was not re-run after the final passes. Anyone reading the register saw a project that never started.
Later asks from the user (e.g. "add an admin account") went straight into a numbered script without going through change-intake. The change log stayed empty, so the decision register no longer explains the built app.
Asked for: the stage line refreshes itself, or is clearly marked as stale once newer exec rows exist, so the build can't end with a wrong readout. And a cheap nudge when an exec script has no matching change-intake row after Stage 5 started.
Related: a second field report (not yet filed) where a whole multi-app build ran through exec.sh while the stage machine stayed at the scaffold baseline; bin/gate-check.sh "Current stage" line; skills/change-intake-detail.md.
Filed by the triage desk from a field report. Generated by Claude Code
Triage: DISCUSS track. instrument gap, verified by inspection Root cause / Problem:
Stale stage line.
## Current stage is written only by bin/gate-check.sh:4342-4418, on a full or --html run, a query for exactly that stage, or a PASSing --closeout (:4192).
exec.sh never touches PROJECT.md, and status.sh:56 echoes the line as it is.
No change-intake nudge.exec.sh links a change request only when --cr is passed (:324-334, :1121-1150), and says nothing when it isn't. The gate-check Change log: line (:3485-3530) counts only the rows that exist, so 0 rows reads PASS. Value: M. Every build that ends without a final gate-check hits it, and ships a wrong register readout plus an empty change log at hand-over. Risk of fixing: L–M.
Option 1 adds a WARN suffix. status.sh, token-burn.sh:353 and obligation-check.sh:182 match ^\*\*Stage, so they must still parse the line.
Option 1: per gate-check or status run, one awk over BUILD-LOG (est. <50 ms).
Option 2: per exec row. Avoid it.
Option 3: per gate-check run (~50 ms); 0 per row.
None can turn red. Needs clarification: no Learning value: learned rule in skills/change-intake.md: after Stage 5, an exec row with neither --cr nor a build-plan row is an unlogged ask. Proposed fix / Options:
A staleness marker shown when read. gate-check (every run) and status.sh print Current stage line is from <date>; N exec rows since — re-run bin/gate-check.sh, and gate-check's own rewrite appends · last exec <date>. All modes; risk L; cost ~0.
exec.sh rewrites the stamp after each row. That adds a new writer of PROJECT.md and a per-row cost. Not recommended.
The Change log: line (warn-only) lists the BUILD-LOG scripts since Stage 5 that have no cr: and are not named in architecture/build-plan.md. All modes with a plan; risk L. Proof: tests/wave2/test-stage-header.sh (a stamp older than the last BUILD-LOG row warns) and tests/wave2/test-change-log.sh (an exec row with no cr: and no plan row is listed). Both fail on master. Effort: M. Recommendation: DISCUSS, options 1 + 3; both are checks at read time with no per-row cost. Decision needed: is a warn-only marker enough, or should Stage 6 / --closeout refuse while the readout is older than the last exec row? Labels: kind/bug, enhancement, size/s (kind/process proposed, not synced)
Reply go (FIX) or go option N (DISCUSS). Fix PRs still get independent review and your merge.
Fix PR opened: #777 (options 1 and 3, warn only). The "Current stage" line now carries the date of the last exec it saw. gate-check and status.sh warn when exec rows have landed since. Under the Change log line, a WARN lists the exec scripts that have neither a cr: tag nor a build-plan row. No verdict or exit code changes. Waiting for an independent review. #776 also edits status.sh, so a composition check is needed.
Activity
MendixMau commentedon Nov 16, 2026
Triage: DISCUSS track. instrument gap, verified by inspection
Root cause / Problem:
## Current stageis written only bybin/gate-check.sh:4342-4418, on a full or--htmlrun, a query for exactly that stage, or a PASSing--closeout(:4192).exec.shnever touchesPROJECT.md, andstatus.sh:56echoes the line as it is.(derived by gate-check on <date>)stamp with newerdocs/BUILD-LOG.mdrows, so a build that stops running gate-check freezes at the first readout. In the field report that readout was Stage P FAIL, which traces back to [bug] Auto interview mode: the required form of an assumed intake answer is hard to find, costing a gate-check round trip #766.exec.shlinks a change request only when--cris passed (:324-334,:1121-1150), and says nothing when it isn't. The gate-checkChange log:line (:3485-3530) counts only the rows that exist, so 0 rows reads PASS.Value: M. Every build that ends without a final gate-check hits it, and ships a wrong register readout plus an empty change log at hand-over.
Risk of fixing: L–M.
status.sh,token-burn.sh:353andobligation-check.sh:182match^\*\*Stage, so they must still parse the line.Process cost (Process-cost guard: a PR that adds a check, gate, obligation or test step must state its time cost per build row, under a budget #687):
None can turn red.
Needs clarification: no
Learning value: learned rule in
skills/change-intake.md: after Stage 5, an exec row with neither--crnor a build-plan row is an unlogged ask.Proposed fix / Options:
Current stage line is from <date>; N exec rows since — re-run bin/gate-check.sh, and gate-check's own rewrite appends· last exec <date>. All modes; risk L; cost ~0.PROJECT.mdand a per-row cost. Not recommended.Change log:line (warn-only) lists the BUILD-LOG scripts since Stage 5 that have nocr:and are not named inarchitecture/build-plan.md. All modes with a plan; risk L.Proof:
tests/wave2/test-stage-header.sh(a stamp older than the last BUILD-LOG row warns) andtests/wave2/test-change-log.sh(an exec row with nocr:and no plan row is listed). Both fail on master.Effort: M. Recommendation: DISCUSS, options 1 + 3; both are checks at read time with no per-row cost. Decision needed: is a warn-only marker enough, or should Stage 6 /
--closeoutrefuse while the readout is older than the last exec row?Labels: kind/bug, enhancement, size/s (kind/process proposed, not synced)
Reply go (FIX) or go option N (DISCUSS). Fix PRs still get independent review and your merge.
Generated by Claude Code
MendixMau commentedon Nov 16, 2026
Fix PR opened: #777 (options 1 and 3, warn only). The "Current stage" line now carries the date of the last exec it saw. gate-check and status.sh warn when exec rows have landed since. Under the Change log line, a WARN lists the exec scripts that have neither a
cr:tag nor a build-plan row. No verdict or exit code changes. Waiting for an independent review. #776 also editsstatus.sh, so a composition check is needed.Generated by Claude Code
fix(gate-check,status): mark a stale Current stage readout and list u…