Index of every page on this site: llms.txt
All skills
e2e-inc avatar

/babysit

@a5d1788
by e2ee2e-inc/e2e14k stars
782
Run it yourself

Drive an open e2e-inc/e2e PR through conflicts, review bots, and CI until it is green with every thread handled, then label it Ready for Human Review. Use after opening a PR, or when asked to babysit, watch, monitor, check on, or get a PR green.

Use this Skill: https://skilld.dev/gh/e2e-inc/e2e/babysit

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesreview-triage.md

≈1.3k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Review triage

How to decide on a review finding, whoever posted it. Bots come and go and change behavior; this file names none of them. Learn who is reviewing a PR from pr-status.ts (thread and comment authors, check names), not from a list.

Every finding

  1. Read the code it points at, and its callers, before deciding.
  2. Check the claim against the current head. Findings go stale: a later commit may already fix it, or it may cite lines that moved.
  3. Reproduce it when you can: a failing unit test, an integration fixture project, or a testbed page. A finding you reproduced is a fix; one you could not reproduce needs a disproof from the code, not a shrug.
  4. Decide: fix, dismiss with a disproof, or escalate.

Fix

A plausible defect in code this PR adds or changes: wrong behavior, a security invariant break, a race, a broken contract. Fix it in this PR with a test that fails without the fix, reply with the commit, resolve.

Dismiss with a disproof

Reply with the one fact that disproves it, then resolve. Common shapes:

  • Pre-existing. git blame or git log -L shows the flagged lines predate the branch. A real, cheap issue gets a follow-up, not scope creep.
  • Misread library semantics. Cite the library source or docs (Playwright, the AI SDK, agent-device, node).
  • Already guaranteed. Name the shared helper, type, closed schema, or e2e/engine contract that makes it impossible.
  • Intentional tradeoff. Link the PR body line, the commit body, or the AGENTS.md rule that states it.
  • A tool rule that contradicts a repo rule in AGENTS.md. Cite the repo rule.

Escalate, never dismiss alone

The security invariants in AGENTS.md (secrets, fill authorization, tool call authorization, navigation policy, sessions, report escaping), the public contract (dist/*.d.ts, tests/types/sdk-types.ts, e2e/engine), wire schemas, peer ranges and release config, and CI gating (filters.yml, required checks). Fix it, or reply, leave the thread unresolved, and name it in the handoff.

Stop the loop

A reviewer that answers every fix with a narrower edge case in the same function is hardening a heuristic, not finding bugs. After two rounds on the same function, reply that the remaining cases are out of scope and name them in the handoff.

Patterns reviewers keep catching here

Code patterns, not bot behavior. The ship-pr self-review checks these before any reviewer does.

  1. Deadline arithmetic: a short remaining budget skips work that would have succeeded, or an expired deadline accepts a transitional or empty screen.
  2. One failure masquerading as another error code: a store outage read as a stale recording, a swipe timeout read as LOCATOR_NOT_FOUND, an operation timeout read as STEP_TIMEOUT.
  3. Secrets reaching an artifact, download, session envelope, report, or model input through a path the redaction does not cover.
  4. Trace cache keys or entries not changing when an input does (launch arguments, permissions, unique() values, URL encoding), so replay picks the wrong action; recordings evicted by a failure that was not theirs.
  5. In-page code trusting the page: app overrides of JSON.stringify, Function.prototype.toString, prototypes, or instanceof across frames.
  6. Hidden nodes, frames, or options entering an observation, or a visibility filter that one query path applies and another skips.
  7. String edge cases: substring vs exact match (Row 12 vs Row 120), whitespace-only values collapsed away, Unicode slicing, BOMs, filename byte limits.
  8. Reruns and repeats: --last-failed, --repeat-each, and retries losing evidence, or a folded report and its GitHub comment disagreeing.
  9. A value the code accepts that the wire schema rejects, or the reverse.
  10. Process lifecycle: a signal, timeout, or cleanup path that leaves a process, handler, or terminal state behind.
  11. Platform leaks: macOS-only tools in an Android path, POSIX paths or symlinks on Windows.
  12. A new build input or benchmark dependency missing from .github/filters.yml, so the benchmark suites skip a change they cover.
  13. Core naming an engine noun (Web, browser, page, route, playwright) outside the init presets.
  14. A scripted rename of our API spelling (a fixture name, a call shape) that also rewrote another library's code: a Playwright page.getByRole in a migration sample, a raw Page from surfaceOf. Grep the diff for the old spelling on receivers that are not ours.
  15. Negative-only security tests: dropping a result or reason makes absence and size checks pass. Assert the expected redacted value exists before checking that the secret is absent or the value is bounded.

Keeping this file useful

Add a dismissal shape or a code pattern only when it is about the code or the repo, never about a specific bot. When the same pattern gets caught twice, it belongs in a lint rule, a type, or a test, not here: propose one, and delete the line once it lands.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    The skill is a PR management tool that interacts with GitHub via standard CLI tools (git, gh). It includes robust security practices, such as explicit warnings against following instructions in PR comments and using secure methods to handle untrusted text to prevent command injection.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: MEDIUM · 1 issue

Signed by skilld at a5d1788. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 weeks ago
metadata
{
  "internal": true
}

README badge

README badge for e2e-inc/e2e/babysit