FN-6229: enforce symptom verification in triage and review

Require bug-class specs and reviews to prove original symptoms are reproduced and fixed.

- Add Symptom Verification guidance to triage prompt templates and self-review rules.
- Teach reviewer prompts to REVISE missing symptom verification or green-build-only acceptance.
- Document symptom-based acceptance alongside Surface Enumeration and cover the prompt contracts with tests.

Files changed:
 AGENTS.md                                      |  1 +
 docs/testing.md                                | 12 +++++++++++
 packages/engine/src/__tests__/reviewer.test.ts | 29 +++++++++++++++++++++++++-
 packages/engine/src/__tests__/triage.test.ts   | 23 +++++++++++++++++---
 packages/engine/src/reviewer.ts                |  2 ++
 packages/engine/src/triage.ts                  | 20 ++++++++++++++++++
 6 files changed, 83 insertions(+), 4 deletions(-)

Fusion-Task-Id: FN-6229

Fusion-Task-Lineage: 01ad963f-c62b-410f-8571-57bc68307288
This commit is contained in:
gsxdsm
2026-06-11 00:04:39 -07:00
parent d68fe3e9f8
commit 1f366c8de1
6 changed files with 83 additions and 4 deletions

View File

@@ -147,6 +147,7 @@ Concrete examples:
- **Dependency correctness:** [Dependencies exist and are appropriate?]
- **Testing requirements:** [Real automated tests required, not just typechecks?]
- **Surface enumeration:** [For bug-fix specs and UI-affordance add/remove specs, is \`## Surface Enumeration\` present and does it enumerate the relevant providers/bridges/execution paths, desktop + mobile breakpoints/platforms, empty/undefined/duplicate/populated states, and shared hooks/components/modules/helpers? For UI-affordance add/remove tasks, also verify: (a) the spec searches for ALL components rendering the affordance, not just the one the user pointed at; (b) the spec explicitly addresses leftover shells after removal across desktop and mobile breakpoints. Missing or incomplete coverage is a blocking REVISE.]
- **Symptom verification:** [For bug-class/bug-fix specs only, is \`## Symptom Verification\` present and complete with **Original symptom**, **Exact reproduction**, and **Assertion it is gone**? A bug-class spec whose final verification only checks green build/tests without reproducing the original failure and asserting it no longer occurs is a blocking REVISE under FN-5893. Missing, empty, or incomplete \`## Symptom Verification\` is a blocking REVISE for bug-class specs; feature/docs/non-bug specs are not required to carry it.]
- **Documentation completeness:** [Must Update / Check If Affected sections present?]
- **Dangling task-document references:** [No \`.fusion/tasks/<id>/<file>\` path is cited in Context, Steps, or File Scope unless the file exists or is explicitly created as a \`(new)\` artifact in this spec. References to nonexistent task-local artifacts are a blocking REVISE.]
- **Sizing & review level:** [Size and review level appropriate for the work?]
@@ -202,6 +203,7 @@ Do NOT demand function-level implementation checklists.
When reviewing tests, check that they verify observable behavior and regression risk (not only implementation trivia).
Flag REVISE when key edge cases or failure modes for changed behavior are untested.
For bug fixes, apply FN-5893 strictly: if the regression test only reproduces the reported case instead of asserting the invariant across the spec's \`## Surface Enumeration\` surfaces, issue REVISE. Use the motivating recurrences (FN-5787/FN-5789/FN-5803, FN-5797/FN-5875/FN-5919, and FN-5751) as concrete examples of why repro-only coverage is insufficient.
For bug-class/bug-fix specs, also enforce symptom-based acceptance: if the spec is missing \`## Symptom Verification\`, leaves it empty/incomplete, lacks **Original symptom**, **Exact reproduction**, or **Assertion it is gone**, or its final verification only checks green build/tests without reproducing the original failure condition and asserting it no longer occurs, issue REVISE. Do not require \`## Symptom Verification\` for feature/docs/non-bug specs.
For UI-affordance add/remove changes, apply the same surface-enumeration strictness: if the test only checks the single surface the user reported instead of all enumerated surfaces, issue REVISE. For UI-affordance removals, require coverage/evidence that empty button shells, orphaned click targets, now-unused wrappers, and dangling aria-labels are cleaned up across desktop and mobile breakpoints; FN-6115/FN-6118/FN-6123 is the motivating recurrence.
## Worktree Boundary Review