Gsx Uat Newtest

gsx-uat-newtest

Codex Adapter

In Codex, translate this Claude wrapper instead of rewriting it:

• $ARGUMENTS = text after $gsx-*.
• /gsd: ... or Skill("gsd:", args="...") = run .codex/skills/gsd-/SKILL.md with args as {{GSD_ARGS}}.
• AskUserQuestion: only map to request_user_input if your Codex build truly renders an interactive picker; otherwise (the usual Codex case) do not force it. Before any choice among options/plans, first print each option's core content in your reply, then list the choices as a numbered Markdown list and stop and wait for the user's text reply. Never self-answer a prompt the user never saw.
• Claude tool names are intent labels; use equivalent Codex tools.
• Preserve wrapper boundaries; if GSD owns edits, do not edit source here.

Append one new Test to the ## Tests section of a Phase's HUMAN-UAT.md / UAT.md, bump updated: + ## Summary, then commit the UAT file. Only touches the UAT file — no source, no STATE.md, no execution; the only thing it commits is that single UAT file. To run tests use gsx-uat-autorun / /gsd:verify-work.

0 — Parse args (Test description + optional Phase number)

$ARGUMENTS is free text (description and/or Phase number, or neither).

• Phase number: Phase-prefixed or bare 1–2 digit token (4 / 04 / Phase 4), or a decimal inserted phase (10.1). Keep decimals as-is; pad bare integers to 2 digits. Multiple candidates / ambiguous → don't guess, AskUserQuestion.
• Test description: remaining text → raw material for step 3.

Phase number blank → infer: (1) Phase under discussion in this conversation; (2) else .planning/STATE.md Last completed phase; (3) still not unique → AskUserQuestion (header Phase).

Test description empty → AskUserQuestion (header Test description) for one sentence covering "what to test / what to expect".

No .planning/ → not a GSD project; tell user and stop.

1 — Locate the UAT file

ls .planning/phases/{NN}-*/{NN}-*UAT.md

• 0 matches → stop; tell user "Phase {NN} has no UAT.md / HUMAN-UAT.md; run /gsd:verify-work {NN} first. This skill won't generate the file."
• Both {NN}-UAT.md + {NN}-HUMAN-UAT.md → prefer HUMAN-UAT.md; note choice in report.
• 1 match → target.

2 — Detect existing format + next number

Read content under ## Tests; match the new Test to the existing style.

Standard structure:

### N. 
expected:           # single-line; or `expected: |` + 2-space-indented block for multi-line
result: pass | issue | pending | skipped | blocked
note:               # optional, only when result != pending

• Title anchor ### N. (trailing English period).
• expected: single vs multi: grep existing for expected: |.
• result: set: pass / issue / pending / skipped / blocked (some files also partial). New Test defaults pending.

Rules:

• ## Tests has ### N.+expected: entries → follow that structure + single/multi-line style.
• ## Tests absent/empty → default structured single-line; first Test is Test 1 (step 5 adds the heading if missing).
• Free-form report style (### 2.1 起栈-type headings, no ### N.+expected:+result:, e.g. Phase 14) → fallback: do NOT force the template; tell user "free-form report UAT — append manually or migrate to structured first" and stop.

Next number: grep all ### N. lines, take max integer N, new = N+1. All Tests (any result) count; numbers never reused, monotonic in time order.

3 — Smart-parse + prompt for missing required fields

Extract from the description (mark unknowns "ask"):

| Field | Clues | Required? | |-------|-------|-----------| | title | one-sentence "what object + what action" | Required (ask if missing) | | expected | expected/observable acceptance point ("expected:", "期望：", "should see") | Required (ask if missing) | | result | "pass", "already pass", "pending", "failed issue" | Default pending if missing | | note | actual evidence; only meaningful when result != pending | Optional |

One pass: pack ALL missing required fields into a single AskUserQuestion (max 4 questions, required-priority order). Don't ask per-field; if all present, don't ask.

result candidates (fixed): pending (not yet run) (Recommended) / pass / issue (found a problem) / skipped (can't determine) / blocked (env/dependency unavailable).

When asking expected, push observable assertions (e.g. "endpoint X returns 200 + R.data has vcId", "/console/users shows paginated table w/ column Y") not "should work"; give a short example in the question.

4 — Duplicate-registration warning

Before writing, compare extracted title against existing ### N. headings. Significant overlap (≥3 effective word tokens vs one entry) → AskUserQuestion (header Duplicate registration):

• New Test — continue (Recommended) → step 5.
• Supplement to Test X → do NOT create a new entry; append a "Supplement ({YYYY-MM-DD}): …" line under that Test's note: (insert note: if absent). Then step 6; Summary unchanged (no new Test, counts unchanged).
• Don't register → stop.

5 — Write the new Test

Edit-append at the end of ## Tests (just before the next ## heading, or EOF). One blank line before/after. If ## Tests absent → add heading after frontmatter, before ## Summary. Chinese Markdown rules (space between Chinese & numbers; Chinese punctuation).

Single-line template:

### {N}. {title}
expected: {expected}
result: {result}

Multi-line template:

### {N}. {title}
expected: |
  {line 1}
  {line 2}
result: {result}

(2-space indent, consistent with file). When result != pending and user gave a note → add note: {note} (single) or note: | block.

Single vs multi selection:

• single sentence, 50% existing use expected: |, else single).
• contains newlines or >80 chars → force multi-line.
• ## Tests empty → default single-line.

6 — Bump frontmatter updated:

In the --- frontmatter, set updated: to current ISO-8601 UTC to the second (date -u +"%Y-%m-%dT%H:%M:%SZ", never hardcode).

• exists → Edit value.
• absent but frontmatter exists → insert below started:.
• no frontmatter → don't force-add; note in report.

7 — Update ## Summary counts

• total: → +1; absent → insert total: 1 at start of Summary.
• matching result count +1 (insert = 1 if absent): pending→pending:, pass→passed:, issue→issues:, skipped→skipped:, blocked→blocked:. Note name shifts: pass→passed: (past tense), issue→issues: (plural).
• absent field but total: exists → insert keeping order (total / passed / issues / pending / partial / skipped / blocked / open_gaps).
• preserve inline comments; increment number and append Test ref in comment (e.g. pending: 2 # Test 4 待跑；Test {N}（新增）).
• ## Summary absent → don't force-add; note in report.
• Step-4 "supplement" branch → total: and all counts unchanged.

7b — Commit the UAT file

Commit the edit — only the UAT file. Registering a Test is a standalone doc change, so give it its own atomic commit:

git add .planning/phases/{NN}-*/{NN}-*UAT.md
git commit -m "docs({NN}): UAT 新增 Test {N}（{title}）"

Scope git add to the exact UAT file path — never git add -A / git add .. Concurrent GSD work may have unrelated files staged; this skill commits only the UAT file, nothing else. This also covers the step-4 "supplement" branch (a supplement line was still written — there use message docs({NN}): UAT Test {X} 补充说明). Only skip the commit when nothing was written at all (step-4 "Don't register" → stop). Record the commit hash in the report.

8 — Summary report

## gsx-uat-newtest — Phase {NN} ({filename})
New Test registered:
- Test {N}: {title} · expected: {summary} · result: {result} · note: {summary or (none)}
Updated: `## Tests` appended · frontmatter `updated:` → {ts} · `## Summary` `total:` {old}→{new}, `{field}:` {old}→{new}
Committed as docs({NN}) {write-back commit hash}.
Next: gsx-uat-autorun {NN} (auto-run) · /gsd:verify-work {NN} (manual) · gsx-uat-newgap {NN} (register a Gap)

Guardrails

• Only the UAT file (*-UAT.md / *-HUMAN-UAT.md); never source / STATE.md / SUMMARY / .planning/quick/. The commit is scoped to the UAT file path alone (never git add -A), so it can never sweep up source or other staged files.
• No branch switch, no execution/verification. The only commit it makes is the UAT write-back itself.
• Numbers only increase, never reused (all results count): new = max(N)+1.
• Follow existing format, don't enforce: single stays single, multi stays multi; free-form report → fallback stop.
• Warn on duplicates; leave the decision to the user.
• One Test per run — re-invoke for more.
• Reply in Chinese with Chinese Markdown formatting.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: JinsYin
• Source: JinsYin/skills
• License: MIT

Install and usage instructions live in the source repository linked above.