Skip to content

docs(sdlc): ChatGPT composer confirmation and compatibility switch bundles - #143

Merged
IchenDEV merged 2 commits into
mainfrom
cursor/placeholder-aware-confirmation
Nov 3, 2026
Merged

IchenDEV merged 2 commits into
mainfrom
cursor/placeholder-aware-confirmation

Conversation

@IchenDEV

@IchenDEV IchenDEV commented Nov 3, 2026

Copy link
Copy Markdown
Owner

Outcome

Docs only: two SDLC bundles for the problems the owner reported after v0.0.62. No source, test, setting or release change.

  1. 2026-11-03-chatgpt-composer-confirmation (High): dictation into an empty ChatGPT composer lands but ends uncertain, because ChatGPT keeps its placeholder inside the field value and removes it on insert. Intent, spec and plan are approved; verification is a placeholder. The spec adds a second accepted value for an empty-looking web editor and a counts-only diagnostic (D5).
  2. 2026-11-03-opaque-target-compat-paste (High): a compatibility switch so dictation still reaches input boxes Utter cannot inspect (WeChat: 4 of 4 recorded deliveries are notDelivered). Intent is approved; spec, plan and verification are placeholders.

SDLC bundle and risk

  • Bundles: docs/sdlc/changes/2026-11-03-chatgpt-composer-confirmation/ and docs/sdlc/changes/2026-11-03-opaque-target-compat-paste/
  • Risk: High for both (core output path); this PR itself changes documents only.
  • Human decisions: the owner approved both intents, the first bundle's spec and plan, and the defaults named in them (chat, 2026-11-03). One approval was a bare "批准" to a compound question; how it was interpreted (D5 included, compat default off, quiet HUD) is written in each Approved-by line and can be reversed in review.
  • Evidence is shape-only (character counts per class, never text): evidence/batch0-shapes.txt and the read-only watcher source.

Verification

  • bash scripts/sdlc-checks.sh
  • bash scripts/ci-basic-checks.sh
  • swift test: not run, no code changed here. The implementation PRs will run it.
  • Release-style build: left to the PR checks.

Residual risk and rollback

  • The ChatGPT placeholder explanation is measured twice (2 of 2) but the placeholder text itself was not read. A second, different cause of uncertain (T3 Code with text before the caret) is observed once and is not fixed here.
  • WeChat's missing focused element is measured once; the probe is still collecting.
  • Rollback: revert the two docs commits; nothing else depends on them.

Reviewer focus

Whether the approval wording in each Approved-by line matches what you actually decided, and the stated limits of the evidence.

IchenDEV and others added 2 commits November 3, 2026 22:37
Intent, spec, plan and a verification placeholder with shape-only Batch 0 measurements. No code changes.

Co-authored-by: Cursor <cursoragent@cursor.com>
…ot inspect

Approved intent with evidence; spec, plan and verification are placeholders. No code changes.

Co-authored-by: Cursor <cursoragent@cursor.com>
@IchenDEV
IchenDEV marked this pull request as ready for review November 3, 2026 14:49
@IchenDEV
IchenDEV merged commit 0dd0e88 into main Nov 3, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant