No related issue: split out of #5191 after a repro pass on planning 0.46.0 (origin/main at 4d1d9d8).
Summary
Finding 8 of #5191, reported against planning 0.42.0: After the session revised a recommendation in response to an own answer, the page kept showing the user's original text as the decision until the user re-accepted, and the later accept event carried the stale own text.
Fix
Decide whether the surface or the session owns clearing the stale own decision, then implement.
Verification
Repro pass on origin/main, playwright-cli driving Chromium against a copy of plugins/planning/surface/, plus round.sh, watch.sh and curl. Verdict: reproduces.
Surface code is unchanged on origin/main (planning 0.46.0). f9fcaf9d3 changed only SKILL.md, context/loop.md, context/surface.md and tests (rule R-K, context/surface.md:190). Live: own event seq 26 "what are the patterns?" on Z3, then revise Z3 --rec ... --seq 26: /api/state responses.Z3 still own with the same text, no terminal; the card chip still reads "Own answer"; #note still holds the text; Ctrl+Enter accept posted seq 27, kind accept, text "what are the patterns?". Code that sets a decision aside exists only for a user hold (round.py:592-617). R-K tells the session to mirror a revised decision with record-terminal; that path was not exercised, so this claims only that the surface code does nothing on its own. Related: #5009 (closed).
Related
Refs #5191. #4653 tracks the commitment and export gaps behind findings 9, 17 and part of 18.
No related issue: split out of #5191 after a repro pass on planning 0.46.0 (origin/main at 4d1d9d8).
Summary
Finding 8 of #5191, reported against planning 0.42.0: After the session revised a recommendation in response to an
ownanswer, the page kept showing the user's original text as the decision until the user re-accepted, and the lateracceptevent carried the stale own text.Fix
Decide whether the surface or the session owns clearing the stale own decision, then implement.
Verification
Repro pass on origin/main,
playwright-clidriving Chromium against a copy ofplugins/planning/surface/, plusround.sh,watch.shandcurl. Verdict: reproduces.Surface code is unchanged on origin/main (planning 0.46.0).
f9fcaf9d3changed onlySKILL.md,context/loop.md,context/surface.mdand tests (rule R-K,context/surface.md:190). Live: own event seq 26 "what are the patterns?" on Z3, thenrevise Z3 --rec ... --seq 26:/api/stateresponses.Z3stillownwith the same text, noterminal; the card chip still reads "Own answer";#notestill holds the text; Ctrl+Enter accept posted seq 27, kindaccept, text "what are the patterns?". Code that sets a decision aside exists only for a user hold (round.py:592-617). R-K tells the session to mirror a revised decision withrecord-terminal; that path was not exercised, so this claims only that the surface code does nothing on its own. Related: #5009 (closed).Related
Refs #5191. #4653 tracks the commitment and export gaps behind findings 9, 17 and part of 18.