test plugin: review/clarification single-agent rewrite + Feishu report checkbox feedback#61
Open
test plugin: review/clarification single-agent rewrite + Feishu report checkbox feedback#61
Conversation
…t checkbox feedback
Replaces multi-perspective subagent design with single-agent + structured reasoning
across requirement-review and requirement-clarification skills. Aligns with 2025
industry best practice: single agent + strong schema + grounded reasoning beats
parallel subagents for requirement-style analysis where context fits one window.
Skills
- requirement-review: drop multi-perspective branch (4.0/4.1a), promote 4.1b serial
review to canonical path with mandatory 假设/反例搜索/结论 + original-text quoting;
add 4.1.7 enum normalization (anti-collapse) and 4.1.8 multi-variant consistency
(class-inheritance defense); add P0/P1 priority + checkbox-list per-role issues;
add Closing Checklist Next Steps with handoffs to test-case-generation; add
rr-summary confidence field (0-100) with reproducible formula
- requirement-clarification: drop 3.2.1 multi-perspective branch; redesign single
agent confidence formula; align severity to Chinese 阻断/关注
- test-case-review: §6 修复 to-do uses MS edit links + 已修复/无需修复 checkbox;
rewrite todo lines to use action verbs (修步骤/补步骤/改预期 etc.) instead of
internal dimension tags; convert §2 coverage matrix to tables
- requirement-traceability: §4 缺陷清单 adds 有效 bug/无效 bug checkbox per defect
with `---` dividers; convert §3.1 coverage matrix and §3.2 code traceability to
tables
- change-analysis: convert §2.5 model fields, §4.1/§4.2 impact domains, §3 coverage
to tables; add §6 cross-validation summary template with §6.4 综合结论 checkbox
Conventions
- CONVENTIONS.md: new "飞书文档渲染规范" section as the single source of truth on
table-vs-bullet selection, banned elements, and severity terminology; replaces
the obsolete "no markdown table" rule that lived in each skill (Feishu import
has supported tables for some time, verified end-to-end)
- _shared/REQUIREMENT_DIMENSIONS.md: terminology mapping table aligns severity
(阻断/关注 with P0/P1 alias for report.md), status (per-FP vs per-dimension),
and verdict↔confidence mapping; deprecates blocking/concern English and the
4-tier 阻断/高/中/低
- contracts/rr-summary.schema.json: confidence required (0-100); blocking_issues
required (must be explicit empty array if none)
Schemas / tests
- contracts/rr-summary.schema.json fixture in tests/check-schemas.sh updated to
include confidence and out-of-range/missing rejection cases
Removed
- agents/requirement-understanding/{user,functional,exception}-perspective.md
(the three perspective subagent definitions are no longer referenced)
Versions
- plugins/test 0.0.9 → 0.0.16 (cumulative across multiple iterations)
- marketplace.json metadata 0.1.41 → 0.1.48
github-actions
Bot
added
scope:tests
Collaborator
Author
Automated Code Review概要：删除 3 个 perspective 子代理， 检查清单
说明：✅ 通过 | 🚫 阻塞问题
|
…ing + cumulative changelog - requirement-review/PHASES.md verdict hard rule: replace OR-joined three-tier with priority chain (hit-and-stop) so verdict is unique under any (confidence, blocking) combination; previously confidence=50 blocking=6 would land in both not_ready and ready_with_conditions - _shared/REQUIREMENT_DIMENSIONS.md: sync correspondence table to priority order - requirement-review/PHASES.md mode detection: drop misleading "align with clarification 4 modes" wording (review only has 3, no exploratory mode since Story context is required) - README: backfill cumulative v0.0.10–v0.0.16 changelog + add v0.0.17 entry - bump plugins/test 0.0.16 → 0.0.17 across plugin.json (×2) and marketplace entry; bump marketplace metadata 0.1.48 → 0.1.49 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reset versions from in-development 0.1.49 / 0.0.17 to production+1 (0.1.42 / 0.0.10) so the seven internal patch bumps (0.0.10-0.0.16 plus the 0.0.17 review fixup) collapse into a single release entry. - Bump plugins/test 0.0.17 -> 0.0.10 across plugin.json (Claude + Codex) and marketplace test entry - Bump marketplace metadata 0.1.49 -> 0.1.42 - CHANGELOG.md: insert 0.1.42 section consolidating single-agent rewrite, Feishu checkbox feedback, conventions/shared/contracts changes - plugins/test/README.md: replace the two interim 0.0.17 + cumulative entries with a single v0.0.10 entry - README.md / README.en.md: update test row 0.0.9 -> 0.0.10 tests/validate.sh passes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
scope:docs
…ence completeness + ui-fidelity/api-contract consolidation GameJam漏报 root-cause fix (input_quality + supplementary cases routing) plus output-spec evidence completeness and architectural consolidation. Smoke-test honest verdict (cumulative across v0.0.12 → 0.0.14): - New PHASES §1.3.d input_quality check (full/medium/low) as single source of truth for all degradation behavior - §3.1 priority 1.5 tier consumes change_supplementary_cases.json (semantic match on title+steps+preconditions, NOT module reverse-lookup) - §4.6 fallback synthesis only when input_quality=low; non-low + empty fv → STOP (no more silent fallback masking) - §5S.2 verdict expanded from binary to 5 tiers - §5S.1 supplementary defects inherit ca priority directly Output-spec evidence completeness (§3.2.0a): - A (data-flow closure) + B (cross-boundary recording) + D (path-driven failure modes) mechanically validated by validate-fv - C (per-expected reconciliation) is honest model self-check, NOT in validator (mechanical keyword match would create false safety) - New cross_component_break defect source (5S.1 source 7) - forward_verification.schema: requirement_id pattern supports FP-N | FP-UNMAPPED-N | R-N; case_source enum added - smoke-test-report.schema: verdict 5-tier + input_quality + verification_channel as required fields - defect-list.schema: P3 added for supplementary inheritance ui-fidelity-check skill consolidated: - Standalone skill removed; check exclusively triggered inside requirement-traceability §3.4 via shared ui-fidelity-checker agent (Browser MCP dropped, structural-only) api-contract-validation kept + agent extracted: - New shared agents/api-contract-validator.md as stateless calc unit - requirement-traceability §3.2.5 launches same agent (no more lite-inline + upstream-first split-brain) Documentation cleanup: - README ↔ AI_CODING_BEST_PRACTICES dedup (README=catalog, BP=SOP) - metersphere-sync env var docs corrected (no fake "zero-config" claim) - All "rt"/"ca" abbreviations expanded to full skill names - "反查" misleading term replaced with "语义匹配" / "推断" Versioning: test plugin 0.0.10 → 0.0.14, marketplace 0.1.42 → 0.1.46 Verification: - bash tests/validate.sh all-green (incl. new Check N+2 A/B/D regression) - Manual replay against session 311+312 artifacts confirmed TC-11 P0 supplementary case now reaches defect_list (missing in 312 baseline) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
scope:docs
CI was failing because metersphere_helper.py module-level requires MS_ACCESS_KEY / MS_SECRET_KEY (for skills calling MS API), but the new Check N+2 only invokes validate-fv which is a pure local schema check that never touches MS. Without these env vars set in CI, the helper exits 1 at module load → validate-fv never runs → grep finds no completeness/A|B|D markers → fail. Quick fix: export ci_dummy values around the Check N+2 block. The deeper fix (move the env precondition from module-level to the actual MS-bound subcommands) is left as a follow-up since it touches every caller of metersphere_helper. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
scope:docs
The module-level MS_ACCESS_KEY / MS_SECRET_KEY check at import time caused validate-fv (a pure local schema-validation subcommand) to fail in CI, where MS credentials are intentionally absent. The ci_dummy workaround in validate.sh (commit 8540307) treated the symptom; this commit fixes the root cause. - _check_ms_credentials(cmd) runs at main() dispatch time, before any subcommand handler. validate-fv (and any future local-only subcommand added to _LOCAL_ONLY_COMMANDS) skips the check; everything else fails early with the same precondition_failed payload as before. - Whitelist is "local only" rather than "MS only" so future MS-bound subcommands inherit the check by default — safer than the inverse. - Roll back the validate.sh dummy export now that the helper does the right thing without it. - Docstring updated to note the new conditional behavior. Verified locally: with .env hidden + MS_* unset, ping fails as expected ("missing required environment variables") and validate.sh Check N+2 still passes end-to-end. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
scope:docs
Reset versions to production+1 per /prepare-release SOP. Collapses the dev-iteration changelog entries (0.1.46/0.1.45/0.1.44/0.1.43 and test plugin v0.0.14/0.0.13/0.0.12/0.0.11) into a single 0.1.42 / v0.0.10 release entry covering all 6 themes shipped in this PR: 1. requirement-review/clarification single-agent strong-reasoning rewrite 2. Feishu report checkbox feedback structures across 4 TEMPLATES 3. AI_CODING_BEST_PRACTICES.md engineer-perspective rewrite + team-internal content moved to Feishu + README↔BP dedup 4. MeterSphere config "zero-config" claim fixed; metersphere_helper lazy MS env credential check (validate-fv works in CI without MS creds) 5. Smoke-test honest verdict overhaul (input_quality routing, GameJam class fix) + A/B/D evidence completeness validator + Check N+2 6. ui-fidelity-check skill consolidated into traceability §3.4; api-contract-validator agent extracted as shared calc unit between the standalone skill and traceability §3.2.5 Verified: bash tests/validate.sh all-green (incl. Check N+2 A/B/D regression) under env -u MS_ACCESS_KEY -u MS_SECRET_KEY (CI sim). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
scope:docs
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Replaces multi-perspective subagent design with single-agent + structured reasoning across
requirement-reviewandrequirement-clarificationskills. Aligns with 2025 industry best practice: single agent + strong schema + grounded reasoning beats parallel subagents for requirement-style analysis where context fits one window.Adds Feishu-friendly report formatting throughout the QA reporting skills: tables for structured data, P0/P1 priority, MeterSphere edit links, and checkbox-based reviewer feedback (有效/无效, 已修复/无需修复) where decisions need to be tracked in the doc itself.
Skills
rr_summary.confidence(0-100) with reproducible formula---dividers; convert §3.1 coverage matrix and §3.2 code traceability to tablesConventions
Schemas / tests
tests/check-schemas.shupdated to include confidence and out-of-range/missing rejection casestests/validate.shpassRemoved
agents/requirement-understanding/{user,functional,exception}-perspective.md(the three perspective subagent definitions are no longer referenced; consumers wererequirement-reviewandrequirement-clarification, both refactored to single-agent)Versions
plugins/test0.0.9 → 0.0.16 (cumulative across multiple iterations)marketplace.jsonmetadata 0.1.41 → 0.1.48Test plan
bash tests/validate.shpasses---dividers all render correctly in import)