FE-1431: Build the plugin authoring surface (cycle two) - #9348
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
1 Skipped Deployment
|
PR SummaryLow Risk Overview The rendered condition-4 teaching layer ( CI prune now copies Reviewed by Cursor Bugbot for commit a48e4c2. Bugbot is set up for automated code reviews on this repo. Configure here. |
github-actions
Bot
added
area/infra
The app's condition-5 test executes the context-owned runner, so package-isolated CI must copy that non-workspace path. Co-authored-by: Cursor <cursoragent@cursor.com>
Optional slot predicates keep kind-indexed prompts live only for the missing fact they address, while load-time checks reject predicates outside the plugin contract. Co-authored-by: Cursor <cursoragent@cursor.com>
Reject dead predicates that no plugin kind can demand and resolve review findings in the adjacent tests. Co-authored-by: Cursor <cursoragent@cursor.com>
One semantic slot can now accept either a numeric grade or a spelled-out structure, preserving calendars and qualitative objective rules without duplicate rows. Co-authored-by: Cursor <cursoragent@cursor.com>
Generic quantity and observed-practice methods now render only for plugin contracts that demand compatible precision, keeping shared defaults relevant without plugin overrides. Co-authored-by: Cursor <cursoragent@cursor.com>
Fill the ADR's missing selection and guidance, resolve the pressure-review contradictions, and make both plugin cells add specific content without repeating repertoire defaults. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
0b4754b to
a48e4c2
Compare
e12163a to
c191c78
Compare
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit a48e4c2. Configure here.
| - **No bare why** — Never ask "why do you do it this way?" as the primary probe; experts cannot report the basis of practised judgment on demand. Ask for an occasion and for what was attended to. | ||
| - **Mean or tail** — Before eliciting any quantity, ask whether what matters is the typical case or the bad one — a mean or a tail. The answer decides whether a single figure, a range, or a spread is being asked for. | ||
| - **Quantiles, never three points** — For anything that varies, ask "typically?", then "one time in ten, worse than?", then "one time in ten, better than?". Never ask for minimum, most likely, and maximum — the three-point habit yields overconfident answers. If a min/mode/max triple arrives unprompted, ask the confidence question and record whether the middle value is a mode or a mean. | ||
| - **Quantiles, never three points** — Ask first for a lower and upper plausible value, then a typical value, then how confident the expert is that reality falls inside the interval. Never ask for minimum, most likely, and maximum — the three-point habit yields overconfident answers. Record which protocol was used rather than silently mixing interval-first and typical-first answers. |
There was a problem hiding this comment.
Contradictory spread elicitation guidance
Medium Severity
The rewritten Quantiles, never three points technique now teaches an interval-first plus confidence protocol, while Precision words / PRECISION_LADDER still defines spread as typical plus one-in-ten tails. The interviewer is given two incompatible accounts of the same demanded grade, so spread elicitation and labeling can diverge from what completion expects.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit a48e4c2. Configure here.
1 participant
🌟 What is the purpose of this PR?
Cycle two of the plugin authoring loop: both plugins and the repertoire are rewritten against the cycle-one review findings, the two key shapes that review demanded are changed (pattern slots and the precision list), and the key catalogue is declared frozen under ADR-0007 decision 9 because the cycle changed no key.
Draft: opened so the stack is visible end to end. The description is completed when the branch is tied off.
🔗 Related links
🚫 Blocked by
ln/fe-1505-timing; lands after it.🔍 What does this change?
Pattern slots scoped to failing slots with wildcard validation; alternative precision demands; repertoire teaching conditioned on plugin precision; the cycle-two content pass, review resolution, freeze, and its evidence readout; the Brunch evaluation runner kept in pruned app jobs.
Pre-Merge Checklist 🚀
🚢 Has this modified a publishable library?
This PR:
📜 Does this require a change to the docs?
The changes in this PR: