Skip to main content
Agent skill

Flake

Track Remotion CI flakes in issue #8375, increment repeated signatures, discover failed PR checks when no PR is given, and rerun flaky GitHub Actions jobs.

Get started

Get the complete skill folder, including its scripts and reference files.

When to use this skill

Use Flake when an AI agent needs a reusable SKILL.md workflow for this job: Track Remotion CI flakes in issue #8375, increment repeated signatures, discover failed PR checks when no PR is given, and rerun flaky GitHub Actions jobs.

When not to use it

Skip Flake when the task is outside the coding category, or when a more specific skill in this directory already covers the same workflow with clearer triggers.

How to install

Where will you use it?

Install scope
Claude Code folder setup
mkdir -p ~/.claude/skills/flake
# Save SKILL.md to ~/.claude/skills/flake/SKILL.md
  1. Claude Code personal install: place the skill and its bundled files at ~/.claude/skills/flake/SKILL.md. For a shared project, commit the folder at .claude/skills/flake/ instead.
  2. Current Claude Code detects SKILL.md changes in watched directories during the session. If you just created a top-level skills directory, run /reload-skills. Check that the skill is listed before testing its trigger.
  3. For Claude chat or Cowork in the Desktop app, manage uploaded skills in Customize > Skills; copying files into a local Claude Code folder is not a Desktop install. Other agents may use different install locations.

Full install guide for Claude, Cursor, and Codex →

Try a starter prompt

After installing the skill, paste this starter prompt into your agent to verify it triggers on a small task.

Starter prompt
Use the Flake skill for this task: Use Flake when this documented workflow matches the task. Confirm any required inputs or files before making changes.

What supports this recommendation?

70/100 provisional content score. This measures documentation, practical detail and source signals. It is not a measured task success rate or a security certification.

Documentation
76/100
Practical detail
68/100
Source evidence
63/100
Source signals
70/100
Model compatibility evidence
  • claudeuntested

    No model-specific signal or recorded compatibility test was found.

  • chatgptuntested

    No model-specific signal or recorded compatibility test was found.

  • geminiuntested

    No model-specific signal or recorded compatibility test was found.

  • copilotuntested

    No model-specific signal or recorded compatibility test was found.

  • llamauntested

    No model-specific signal or recorded compatibility test was found.

  • perplexityuntested

    No model-specific signal or recorded compatibility test was found.

  • mistraluntested

    No model-specific signal or recorded compatibility test was found.

  • grokuntested

    No model-specific signal or recorded compatibility test was found.

Inferred compatibility is based on content signals. Check the app, its available tools and setup requirements before running a skill.

What this skill does

# Flake

Use this skill when a Remotion CI check fails and looks flaky, or when asked to run `/flake`.

Tracker issue: https://github.com/remotion-dev/remotion/issues/8375

## Goal

Record flaky CI failures in issue #8375 and get the affected suite running again.

For every confirmed or likely flake:

1. Identify the PR, workflow run, job, failed step, and concise failure signature. 2. Update issue #8375: - If the signature already exists, increment its `Count`. - If it is new, insert a row with count `1`, workflow/job, first-seen date, and evidence URL. - Sort all rows by numeric `Count` in descending order after every addition or increment, preserving the existing order for equal counts. 3. Rerun the flaky job or failed jobs.

## Find The Failure

Prefer `gh` because the GitHub app may not expose all Actions logs or write permissions.

If the user gave a PR:

```bash gh pr view <pr-number> --repo remotion-dev/remotion --json number,title,headRefName,statusCheckRollup gh pr checks <pr-number> --repo remotion-dev/remotion --watch=false ```

If the user gave a workflow run or job URL:

```bash gh run view <run-id> --repo remotion-dev/remotion --json databaseId,attempt,displayTitle,event,headBranch,headSha,status,conclusion,workflowName,createdAt,updatedAt,url,jobs gh api repos/remotion-dev/remotion/actions/jobs/<job-id>/logs ```

If there is no PR context, inspect recent open PRs and recent failed runs:

```bash gh pr list --repo remotion-dev/remotion --state open --limit 30 --json number,title,headRefName,statusCheckRollup gh run list --repo remotion-dev/remotion --workflow "Install and Test" --status failure --limit 20 --json databaseId,displayTitle,event,headBranch,headSha,conclusion,createdAt,url,workflowName ```

For rerun attempts, inspect the original attempt before deciding whether a failure was flaky:

```bash gh api repos/remotion-dev/remotion/actions/runs/<run-id>/attempts/1/jobs --paginate gh api repos/remotion-dev/remotion/actions/jobs/<failed-job-id>/logs ```

## Classify A Flake

Treat a failure as a flake when at least one is true:

- The same run, job, or failed step succeeded after a rerun without code changes. - The failure is infrastructure/network related, such as `fetch failed`, connection resets, package manager download failures, cache service failures, or runner provisioning errors. - The test failure is a timeout, race, browser disconnect, port collision, or intermittent external service failure, and the PR changes do not plausibly touch that area.

Do not record deterministic type errors, lint errors, snapshots, or test assertions caused by changed source behavior unless there is evidence they pass on retry.

## Signature Rules

Use a stable signature that groups repeats but stays specific enough to act on:

- Include package or suite when available, for example `@remotion/it-tests#testssr`. - Include test name when available. - Include the root error text, stripped of timestamps, paths, random IDs, ports, line numbers, and ANSI codes. - Include workflow/job separately in the tracker row, not inside the signature unless needed to disambiguate.

Good examples:

- `` `oven-sh/[email protected]` failed with `TypeError: fetch failed` `` - `` `delayRender()` timeout while running `Render video with browser instance not open` in `@remotion/it-tests#testssr` ``

## Update The Tracker

Read the issue body:

```bash gh issue view 8375 --repo remotion-dev/remotion --json body --jq .body ```

Edit the full issue body, preserving the introduction, all existing entries, and the `Workflow` section. Keep every flake row in one uninterrupted table under `## Flakes`, before `## Workflow`. Blank lines terminate Markdown tables: never put blank lines between the header, separator, or data rows, and never append rows below the `Workflow` section or at the end of the issue.

Write the replacement Markdown with real newline characters, not literal `\n` text. When using `gh`, always save it to a temporary file and use `--body-file`; do not pass multiline Markdown inline.

Table format:

```md | Count | Signature | Workflow / job | First seen | Evidence | | ----: | -------------------- | ------------------ | ---------- | --------------------------------------------------------- | | 2 | `repeated signature` | `Workflow` / `Job` | YYYY-MM-DD | https://github.com/remotion-dev/remotion/actions/runs/... | | 1 | `signature` | `Workflow` / `Job` | YYYY-MM-DD | https://github.com/remotion-dev/remotion/actions/runs/... | ```

Keep exactly five cells per row. Escape literal pipes as `\|`, including inside inline code, and use `<br>` for line breaks within cells instead of real newlines.

When incrementing an existing row:

- Change only `Count` unless the evidence link is dead or less useful than the new one, then move the row to its new sorted position. - Keep `First seen` as the earliest known date.

When inserting a row:

- Use the GitHub Actions job URL when possible. - Use the current date from the log timestamp, not the local machine date, if they differ. - Insert it into the existing table in descending numeric `Count` order, after existing rows with the same count.

Before publishing, check that all existing entries are retained, each row has five cells, counts are sorted numerically from highest to lowest, and no blank lines or non-table content interrupt the table. Preserve all counts, first-seen dates, and evidence except for the intended updates.

Publish the prepared file:

```bash gh issue edit 8375 --repo remotion-dev/remotion --body-file /path/to/tracker.md ```

Verify after editing:

```bash gh issue view 8375 --repo remotion-dev/remotion --json body --jq .body ```

Confirm the saved body matches the prepared Markdown and still has one complete, uninterrupted table with descending counts. Inspect the rendered issue (or GitHub-rendered HTML) to ensure every entry appears inside the five-column table; a successful API response alone does not verify the layout.

## Rerun

Prefer rerunning only failed jobs:

```bash gh run rerun <run-id> --repo remotion-dev/remotion --failed ```

If only one known flaky job needs a rerun and the run is complete:

```bash gh api --method POST repos/remotion-dev/remotion/actions/jobs/<job-id>/rerun ```

If a flaky suite is still in progress and waiting would block the task, abort that workflow run and rerun failed jobs after cancellation is visible:

```bash gh run cancel <run-id> --repo remotion-dev/remotion gh run watch <run-id> --repo remotion-dev/remotion --exit-status gh run rerun <run-id> --repo remotion-dev/remotion --failed ```

If `gh run watch` exits non-zero because cancellation made the run fail, continue to rerun failed jobs.

## Report Back

Summarize:

- Tracker issue row added or incremented. - Rerun action taken, including run or job URL. - Any remaining running checks that were intentionally left alone.

Intended uses

  • Use Flake when this documented workflow matches the task.

Implementation notes

  • Review the source instructions and adapt inputs before running the workflow.

Try it, then tell us how it went

A copied prompt does not tell us whether the task worked. Report your result after checking the output. We collect your tool and outcome, never your files or prompt.