返回 CodeWhale
SKILL.md
1 ---
2 name: codew-release-qa-sweep
3 description: "Use before claiming Codewhale release work is done: run the full gate sweep and list the manual QA targets."
4 ---
5
6 # Codewhale Release QA Sweep
7
8 Run this before claiming any Codewhale release work is "done." A green automated
9 gate sweep plus the three manual QA targets is the evidence bar. No sweep, no
10 "done" — report exactly what was run and the result of each step.
11
12 ## When to use
13
14 - Before telling Hunter (or a PR thread) that release work is complete or
15 merge-ready.
16 - After harvesting/landing PRs into the release branch, before the publish boundary.
17 - When verifying a release candidate on the **real** landing branch
18 (e.g. `<release-branch>`), which is often local-only.
19
20 ## Automated gate sweep
21
22 Run from the repo root, in order. Stop on the first failure and report it.
23
24 ```bash
25 # 0. Confirm you are on the real release head, not a main-based assumption.
26 git branch --show-current # expect e.g. <release-branch>
27 git status --short # working tree should be clean
28
29 # 1. Formatting + stray whitespace/conflict markers
30 cargo fmt --all --check
31 git diff --check
32
33 # Core npm workspace tests plus the shared web gate (install web dependencies first).
34 # check:web checks committed facts before regeneration, then docs, design tokens,
35 # lint, TypeScript and the production build. These do not replace the Rust gates.
36 npm test && npm run check:web
37
38 # 2. Library/protocol/cli/flow/state tests, locked
39 cargo test -p codewhale-config -p codewhale-protocol -p codewhale-cli \
40 -p codewhale-workflow -p codewhale-state --locked
41
42 # 3. TUI test binaries, locked
43 cargo test -p codewhale-tui --bins --locked
44
45 # 4. TUI debug build, locked
46 cargo build -p codewhale-tui --locked
47
48 # 5. Release build for the shipped binaries, locked
49 cargo build --release --locked -p codewhale-cli -p codewhale-tui
50
51 # 6. Version-drift gate (workspace ↔ npm ↔ Cargo.lock ↔ changelog ↔ README)
52 ./scripts/release/check-versions.sh
53
54 # 7. Binary smoke
55 ./target/release/codewhale --version
56 ```
57
58 If you are validating a PR for landing, also test mergeability against the
59 **actual** release head, never the main-based clean flag:
60
61 ```bash
62 git merge-tree $(git merge-base <release-branch> <pr-head>) <release-branch> <pr-head>
63 ```
64
65 A PR that is clean against `main` can still conflict with the release branch.
66
67 ## Manual QA targets
68
69 Unit/build gates do not cover the live TUI. Exercise all three and record what you saw:
70
71 Use the built binary with a sealed local home and loopback fixtures, then
72 exercise the relevant scenarios below in an actual terminal. Record dimensions,
73 inputs, visible state, and side effects. Do not substitute a full-screen
74 assertion harness for looking at and using the product.
75
76 1. **Six-worker fanout liveness (#3216/#2211).** Spawn 6 sub-agents. Confirm
77 typing, render, cancel, and the workbar stay live throughout, and that **Esc
78 cancels mid-fanout** (prompt interrupt, not a wedged ~24s burst or freeze).
79 For the Windows Terminal retest path from #3289, start in plan mode, add
80 follow-up input to the plan, press Esc, switch to yolo/accept flow, trigger
81 at least two auto/Fleet worker spawns, and keep typing/cancel/mode-switch
82 checks live for several minutes. Attach logs if the freeze reproduces.
83 2. **Multi-terminal route isolation (#3227).** Open multiple terminals on
84 distinct provider/model routes. Confirm zero cross-terminal contamination and
85 no provider+model mismatch — each terminal honors its own route.
86 3. **Running-turn input contract (#3203).** During a busy turn, confirm Enter
87 queues a typed follow-up, the preview advertises **Enter send now**, and an
88 empty Enter promotes the oldest queued follow-up. Confirm Ctrl+Enter steers
89 typed text directly, Shift+Enter inserts a newline, and Ctrl+G/Ctrl+S only
90 stash drafts.
91
92 ## Reporting format
93
94 Report a checklist: each command, pass/fail, and the salient output line
95 (test counts, the `--version` string, `check-versions.sh` verdict). For manual
96 QA, state what you actually observed per target, citing the issue number. If a
97 step was skipped or could not be run (e.g. no display for TUI QA), say so
98 explicitly — do not imply coverage you do not have.
99
100 ## Red flags / don't
101
102 - Don't claim "done," "passing," or "merge-ready" without the evidence above.
103 Assertions without command output are not acceptable.
104 - Don't trust the main-based mergeability flag for a release branch; use
105 `git merge-tree` against the real head.
106 - Don't skip the manual TUI targets because the build is green — the freeze,
107 route-mismatch, and steering regressions live in the runtime, not the gates.
108 - Don't tag, publish, create a GitHub Release, push artifacts, or merge/close
109 any PR or issue without Hunter's explicit approval. A green sweep is readiness
110 evidence, not permission.
111 - Never harvest/close from a PR title or label alone — review from code, tests,
112 comments, and checks.
113 - When the sweep clears a harvested PR, preserve contributor credit: cherry-pick
114 keeps the original author, otherwise add `Co-authored-by: Name <email>` and
115 `Harvested-from: PR #N by @handle` so the auto-close-at-main workflow credits
116 the contributor.
117 - Keep any contributor-facing comment positive and crediting; gates stay
118 dry-run/advisory unless Hunter approves enforcement.
119
119 lines MARKDOWN