
















@@ -16,8 +16,11 @@ Hard limits:
1616- Do not finish with tiny, cropped-wrong, off-bottom, or sidebar-heavy GIFs.
1717- Do not invent a generic proof. The proof must match the PR behavior.
1818- Do not force GIFs for internal-only, workflow-only, test-only, docs-only, or
19- otherwise non-visual PRs. A no-visual-proof manifest is a successful outcome
20- when GIFs would be misleading.
19+ otherwise non-visual PRs. A no-visual-proof manifest is a successful workflow
20+ outcome when GIFs would be misleading, but it is not proof that the PR passed.
21+- Keep public-facing manifest summaries short and user-domain. Do not mention
22+ harness internals, mock-provider limits, secret/trust boundaries, local paths,
23+ transcript seeding, or workflow implementation details in the summary.
21242225Inputs are provided as environment variables:
2326@@ -42,9 +45,10 @@ Required workflow:
4245 before/after. If it does not, write
4346`${MANTIS_OUTPUT_DIR}/mantis-evidence.json` with `comparison.pass: true`, no
4447 artifacts, and a summary that starts with
45-`Mantis did not generate before/after GIFs because`. Include the concrete
46- reason in the summary. Use this manifest shape and do not create worktrees
47- or start Crabbox for this case:
48+`Mantis did not generate before/after GIFs because`. Include a short
49+ public reason, such as `the PR changes internal session bookkeeping rather
50+than Telegram-visible behavior`. Use this manifest shape and do not create
51+ worktrees or start Crabbox for this case:
48524953```json
5054 {
@@ -73,6 +77,14 @@ Required workflow:
7377 }
7478```
757980+ If the PR appears visual but proof is blocked by Telegram Desktop session
81+ state, authorization, credentials, Crabbox, or another capture-infrastructure
82+ issue, do not describe it as a no-visual PR. Write a manifest with
83+`comparison.pass: false`, skipped lanes, no artifacts, and a summary that
84+ starts with `Mantis could not capture Telegram Desktop proof because`. The
85+ publisher will keep that out of PR comments so the failure stays in the
86+ workflow logs and artifacts.
87+76884. Decide what Telegram message, mock model response, command, callback, button,
7789 media, or sequence best proves the PR. Use `MANTIS_INSTRUCTIONS` as extra
7890 maintainer guidance, not as a replacement for reading the PR.
@@ -134,4 +146,6 @@ Expected final state:
134146`Main` and `This PR`.
135147- No-visual-proof manifests contain no artifacts and have `comparison.pass:
136148true`.
149+- Capture-infrastructure failure manifests contain no artifacts and have
150+`comparison.pass: false`.
137151- The worktree can be dirty only under `.artifacts/`.
此内容由惯性聚合(RSS阅读器)自动聚合整理,仅供阅读参考。 原文来自 — 版权归原作者所有。