Skip to content

test+docs: drift guards batch — grammar completeness, docs sync, surface-authority honesty (architecture review F1/F6/F9) - #53

Closed
silversurfer562 wants to merge 1 commit into
claude/forms-cleanup-batchfrom
claude/forms-drift-guards
Closed

test+docs: drift guards batch — grammar completeness, docs sync, surface-authority honesty (architecture review F1/F6/F9)#53
silversurfer562 wants to merge 1 commit into
claude/forms-cleanup-batchfrom
claude/forms-drift-guards

Conversation

@silversurfer562

Copy link
Copy Markdown
Member

Summary

Batch C of the 2026-08-20 dynamic-forms architecture review. Stacked on #52 (base: claude/forms-cleanup-batch) — merge that first, then retarget/merge this.

The review's central ruling was to ACCEPT the ~19-file per-construct cost as the honest price of four surfaces (a construct means something different on each one) and pin it with drift catchers instead of abstractions. This batch is those pins.

Changes

  • tests/test_grammar_completeness.py (F1 pin): every QuestionType member must carry a row in two completeness tables (widget collect mode + a canonically wrong-shaped answer), and each of the four surfaces must emit construct-specific output for it — widget data-collect modes, AskUserQuestion expansion counts (derived from the shared helpers), elicitation-schema property shapes, the markdown reply skeleton's answer shapes, and validator rejection by field name. A construct wired into only three surfaces, or a 16th type added without updating the tables, fails red instead of degrading silently.
  • tests/test_docs_drift.py (F6): README's spelled-out construct count must track the enum (it rotted by hand once — 543a7a0), README/SKILL.md must describe every construct/question type, and every MCP tool or x_to_y library function the skill names must actually exist.
  • docs/adding-a-construct.md: the 19-touchpoint checklist with file anchors, PRs feat: ranking construct — grammar member #7 (0.6.0) #24/feat: assumption-review construct — grammar member #8 (0.6.0) #25 as worked examples, and the rejected alternatives (base class, entry points, codegen) recorded so they stay rejected.
  • Surface-authority honesty (F9): select_form_surface's docstring and the README now state that in the shipped plugin the router is advisory — the agent's MCP tool choice is the effective decision (skill prose ladder), the router runs after the fact for telemetry agreement, and its answer is binding only for library consumers routing their own calls. The markdown-surface gap in its range is noted for the 0.7.0 follow-through (Batch D).

Not in this batch (release-gated → 0.7.0, Batch D)

SKILL.md step 4 still instructs library-only markdown calls a plugin-only (uvx) install cannot make (F3). That fix is either two thin MCP tools or an honest degradation paragraph — a chair decision on the 0.7.0 train.

Verification

  • Full suite: 715 passed.
  • Mutation check: changing README's "eight constructs" fails test_readme_states_the_real_construct_count red.

🤖 Generated with Claude Code

@silversurfer562
silversurfer562 force-pushed the claude/forms-drift-guards branch from 548cdfe to 912368f Compare August 20, 2026 11:57
@silversurfer562
silversurfer562 force-pushed the claude/forms-cleanup-batch branch from e47d35b to 614cb2b Compare August 20, 2026 13:13
…ace-authority honesty (architecture review F1/F6/F9)

The F1 accept-and-pin ruling made executable: a grammar-wide
completeness test (every QuestionType must have table rows and
construct-specific output on all four surfaces), docs drift tests for
the hand-maintained README/SKILL grammar docs, the adding-a-construct
touchpoint checklist, and advisory-status honesty on
select_form_surface.

Mutation-checked: changing README's construct count fails red.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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