Skip to content

Forge wave closure, lean E2E guide, human walkthrough

Choose a tag to compare

@Haaaiawd Haaaiawd released this 04 May 01:37
· 10 commits to main since this release

/forge

  • Static code-reviewer at Step 1 §1.4 (pre-wave): after Wave is written and the Step 1 checkpoint is satisfied, before Step 2, review commits on the working branch vs PRD / ADR / System Design / 05_TASKS.md (default each new wave; documented skip: no reviewable implementation yet, or CODE_REVIEW_DISABLED_BY_USER). Wave-end §3.4.5 is removed from the flow in favor of this gate.
  • Normal vs AUTO: Normal mode—you still propose the wave (§1.1–1.3 format); each wave needs show → user approves → Wave + §1.4, then Step 2/3; the next wave repeats Step 1. AUTO—AUTO sign-off, assume the user may be away; no “confirm this mix / continue next wave?” prompts; stops only on the documented AUTO must-stop set plus Challenge Critical, §1.4 hard blocks, manual verification finals, and Critical/High from pre-wave review without an acceptable path.
  • Step 4 §4.0 hard gate: §1.4 done (auditable) or documented waiver; and all wave tasks finished including §3.4.6 on the last task (Closure A/B or skip) plus §3.5 / §3.6. Resume mid-wave: do not re-run §1.4 for the same wave unless records are missing—then return to Step 1.
  • Per-task flow: non–last-task → §3.4 → 3.5 → 3.6; last task → §3.4 → §3.4.6 → 3.5 → 3.6.
  • §3.4.6 wave-end E2E: unchanged orchestration—Closure A (§1.4 + §3.4, note UI risk) vs Closure B (e2e-testing-guide then browser with authorization); skip with one line if no E2E/manual in the wave; guide-only if no browser tools.
  • Copy: shorter Challenge preface in Step 0; Step 3 goal stays task-only; cadence detail in forge + code-reviewer templates.
  • End checklist matches §1.4 + Closure A/B (no legacy §3.4.6 scoring vocabulary).

e2e-testing-guide

  • Scope narrowed: the skill defines how to write the human-oriented testing guide plus the E2E Verification report skeleton; live browser sequencing is owned by /forge §3.4.6 (report first, then browser).
  • Removed from the skill: long “browser execution protocol”, per-step/per-journey multi-status matrices, JourneyScore, overall scoring, INCONCLUSIVE / BLOCKED / NOT_RUN as first-class verdict columns.
  • Three verdicts only for journey/step columns: PASS, PARTIAL_PASS, FAIL—semantics live in <!-- --> comments immediately above Required output in the skill (no separate scoring section).
  • Human walkthrough (required authoring): new section 「测试指南要像人类那样写」 / “The testing guide must read like a human walkthrough”—real entry points, read-then-click, full app chrome, primary + secondary affordances, common user combos, empty/one/many list expectations; Surface rows must say how a human discovers each entry (not bare routes).
  • Tables kept: RTM, Surface coverage, Journeys, Step breakdown, Findings / Coverage gaps / Recommendation (no JourneyScore / scoring summary columns).
  • EN template hygiene: valid YAML frontmatter and corrected /forge references where broken markdown had appeared.

Upgrade

npm install -g @haaaiawd/anws@latest
anws update