GPT Review — 23-P3C1 rev6-restored and UX Notes rev2

Date: 2026-05-07
Reviewer: GPT-5.5 Thinking / Incomex Hội đồng AI
Reviewed:

• knowledge/dev/laws/dieu44-trien-khai/prompts/23-p3c1-iu-edit-draft-safe-functions-prompt.md rev6-restored
• knowledge/dev/laws/dieu44-trien-khai/design/23-p3c1-ux-state-foundation-notes.md rev2

Verdict

Do not dispatch yet. Rev7 required.

Opus successfully restored the P3C1 execution prompt and correctly patched UX notes to the per-actor last-view watermark model. The UX design direction is accepted.

However, the restored prompt has a serious shell-test blocker: many [ ... ] comparisons omit spaces around =, which makes tests pass incorrectly.

UX notes verdict

Accepted.

The per-actor watermark model is the right PG-native direction:

• no global last_viewed_at;
• state is per (draft_id, reviewer_ref);
• unread = comments newer than that actor’s watermark;
• latest 3–5 viewers are derived by ORDER BY last_viewed_at DESC LIMIT 5;
• actor card/GUC hook is deferred but recorded;
• no read/unread hacks in comment body or JSONB hot path.

This satisfies the User’s operational concerns: GPT, Opus, Gemini, etc. each keep their own viewed state and do not overwrite each other.

Prompt restore verdict

The full structure is restored:

• pre-read companion UX notes;
• preflight;
• four safe function bodies;
• T1–T21 tests;
• cleanup/report;
• hard boundaries.

But rev6-restored is not executable-safe yet due to shell assertion syntax.

Required rev7 fixes

P1 — BLOCKER: fix all shell [ "$X"="Y" ] comparisons

Rev6-restored contains many checks like:

[ "$T1"="no_change" ]
[ "$T2"="plan_ok" ]
[ "$T5_S"="draft_created" ]
[ "$SD"="t" ]

In POSIX shell, [ "$T1"="no_change" ] is a single non-empty argument, so it evaluates true regardless of the value. This causes false PASS.

Patch all such tests to include spaces around =:

[ "$T1" = "no_change" ]
[ "$T2" = "plan_ok" ]
[ "$T5_S" = "draft_created" ]
[ "$SD" = "t" ]

Also patch numeric/count comparisons:

[ "$DRAFT_NOW" = "$((DRAFT_BEFORE+2))" ]
[ "$DROP_FAIL" = "0" ]

Search and fix every pattern matching:

• "="$
• "="
• [ "$VAR"="..." ]
• [ "$(...)"="..." ]

P2 — Add a shell safety self-check

Add early in setup:

if [ "x"="y" ]; then
  echo "BUG: shell test operator spacing check failed";
  exit 2;
fi

But note this exact expression is intentionally bad and will return true in many shells. Better self-check:

if [ "x" = "y" ]; then
  echo "BUG: shell equality broken";
  exit 2;
fi

And add a comment:

# All test comparisons MUST use spaces around '=': [ "$A" = "$B" ]

Do not include the intentionally bad expression as executable code unless used only as a comment, because it would always fail the script.

P3 — Make T19 grantee privilege query quote role safely

Current:

SELECT has_function_privilege('$R','public.$SIG','EXECUTE');

If role contains special characters, this can break. Use psql variables:

CAN=$("${PSQL_NOSTOP[@]}" -v role="$R" -v sig="public.$SIG" -t -A -c \
  "SELECT has_function_privilege(:'role', :'sig', 'EXECUTE');" 2>/dev/null)

P4 — T20 should avoid raw address interpolation

T20 still contains:

SELECT (fn_iu_comment('$TEST_ADDR_A','x','x'))->>'guidance'

Patch to use -v addr_a, -v addr_b and :'addr_a', :'addr_b'.

P5 — Add report line for UX notes version read

In final report, include:

ux_notes_read=23-p3c1-ux-state-foundation-notes.md rev2
ux_state_hooks=timestamp,actor_card_guc_deferred,per_actor_watermark_deferred

This makes it explicit that the last-view model is captured but not implemented in P3C1.

Directive to Opus

Patch P3C1 prompt to rev7 with P1–P5.

Path:

knowledge/dev/laws/dieu44-trien-khai/prompts/23-p3c1-iu-edit-draft-safe-functions-prompt.md

Do not dispatch after patch. Return for GPT/User final review.

Hard boundaries

• No dispatch.
• No table DDL.
• No schema changes.
• No trigger changes.
• No gateway changes.
• No IU/UV writes.
• No vector mutation.
• No cleanup.
• No Pack 2C.

Summary

The UX-state design is now correct: per-actor last-view watermark, not global. The execution prompt is restored, but the shell test comparison bug is a real blocker because it can make failures look like PASS. Rev7 should fix comparison syntax, tighten remaining quoting, and record the UX notes hook in the report.