CI Diagnostics
Testing & Qualityci-diagnosticsCI Diagnostics is an Agent Skill for testing & quality published in the timeplus-io/proton repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Diagnose Proton CI failures and performance comparison results from GitHub checks and uploaded reports. Make sure to use this skill whenever CI checks are mentioned, a PR has red or failing checks, the user pastes a CI URL, or asks about test failures in the pipeline, even if they just ask 'why is CI failing'.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/timeplus-io/proton/blob/HEAD/.claude/skills/ci-diagnostics/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/ci-diagnostics, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
CI Diagnostics
Inputs
- PR number or URL
- optional check name substring
- optional direct report URL
First step: collect check status
For a PR:
gh pr view "$PR" --json title,body,url
gh pr checks "$PR"
REPO=$(gh repo view --json nameWithOwner --jq .nameWithOwner)
SHA=$(gh pr view "$PR" --json commits --jq '.commits[-1].oid')
gh api "repos/$REPO/commits/$SHA/status"
Use the commit status payload to find:
- failing or pending contexts
target_urllinks for uploaded HTML reports, raw logs, or performance artifacts
Proton report layout
CI uploads reports under:
<pr-number>/<commit-sha>/<normalized-check-name>...
The normalization logic is defined in:
Normalize a check name with lowercase and replacements for spaces, (, ), and ,.
Failure triage workflow
- List failing contexts from
gh pr checksor commit statuses. - Open each
target_urlreport first. - If the report is sparse, inspect the linked raw log.
- For test reports, summarize:
- failing test names
- first common error signature
- whether the failure looks deterministic, flaky, infra, or environment-specific
- Map failures back to touched areas in the diff.
Performance comparison workflow
Performance comparison artifacts upload:
report.htmlall-queries.htmlall-query-metrics.tsvqueries.rep- images/flamegraphs
If you have a report.html URL, inspect sibling artifacts by replacing the filename in the same prefix.
When reviewing perf results:
- start from the summary in
report.html - inspect
all-query-metrics.tsvfor the biggestclient_timeregressions - distinguish broad regressions from a few outlier queries
- correlate with touched execution paths, joins, windows, aggregations, or storage reads
Output expectations
Always report:
- failing checks
- best report/log URL for each failing check
- likely failure class: code bug, flaky test, infra, dependency, or timeout/resource limit
- smallest next debugging action
For performance changes also report:
- whether the regression is broad or narrow
- the most affected workload family
- whether more local benchmarking is needed before code changes