ENH: Add fix-nightly-warnings skill and CDash query scripts

[blowekamp]

Summary

Adds a GitHub Copilot agent skill for the workflow of fixing ITK nightly build warnings reported on CDash, along with two helper Python scripts for querying the CDash GraphQL API.

Changes

• SKILL.md — Skill definition describing the step-by-step procedure: fetch nightly build warnings from CDash, identify warning patterns, apply minimal source fixes, build affected modules to verify, and open a draft PR.
• list_nightly_warnings.py — CLI script to list recent CDash nightly builds and their warning counts via the GraphQL API.
• get_build_warnings.py — CLI script to fetch and display detailed warning messages for a specific CDash build ID via the GraphQL API.

Notes

No production source changes; purely development tooling.

[blowekamp]

I have had some success with this skill. Adding the scripts is a way to reduce the number of requests needed for the AI, and improve efficiency. I have am running low on available requests with the initial development of this skill.

Someone else is welcome to further refine this workflow.

[dzenanz]

Looks good on a glance.

[thewtex]

Very cool!!

We may want to set up a GitHub Agentic Workflow to run this on a regular basis.

[hjmjohnson]

@blowekamp I have some improvements ready for this PR (typo fixes, ThirdParty filtering, MSVC warning support, a triage_nightly.py entry point, etc.).

To let me push directly to your PR branch, could you enable "Allow edits from maintainers"? You can do this by:

1. Scrolling to the bottom of this PR page
2. Checking the box "Allow edits and access to secrets by maintainers"

(This option is in the right sidebar under the PR description when you're editing the PR, or at the bottom of the PR creation/edit page.)

Once enabled, I'll be able to git push to your fork's skill_fix_nightly_warnings branch directly. Alternatively, I can close this PR and open a new one from upstream/skill_fix_nightly_warnings where I've already pushed the changes — let me know your preference.

======
FYI: I asked claude to make separate commits with recommendations to make this skill more automated and efficient:

https://github.com/InsightSoftwareConsortium/ITK/tree/skill_fix_nightly_warnings

The 10 commits should probably be squished, but they seem reasonable for draft consideration.

[blowekamp]

That option is already enabled.

[greptile-apps]

Greptile Summary

This PR adds a GitHub Copilot agent skill (fix-nightly-warnings) with three Python helper scripts for querying the CDash GraphQL API, triaging ITK nightly build warnings, and automating the fix-and-PR workflow. There are no production source changes.

• triage_nightly.py line 116: the empty-builds JSON branch emits \"warnings\": [] while the normal path emits \"warnings_by_flag\": [...] — any agent or jq pipeline reading .warnings_by_flag will silently get null instead of an empty list.

Confidence Score: 4/5

Safe to merge after fixing the JSON key inconsistency in triage_nightly.py.

One P1 defect: the empty-builds branch in triage_nightly.py emits "warnings" instead of "warnings_by_flag", which silently breaks any consumer (agent or jq pipeline) that reads the normal key. The remaining findings are P2 style/quality issues. Per the confidence guidance, a P1 finding keeps the score at 4.

triage_nightly.py — fix the JSON key mismatch in the early-return branch (line 116).

Vulnerabilities

No security concerns identified. The scripts only perform read-only queries against a public CDash GraphQL API endpoint; no credentials are handled, no user data is processed, and no code is executed from remote sources.

Important Files Changed

Filename	Overview
.github/skills/fix-nightly-warnings/SKILL.md	Agent skill definition for the nightly-warning triage workflow; minor: Quality Checks section uses bare [] instead of - [ ], so checkboxes won't render correctly on GitHub.
.github/skills/fix-nightly-warnings/scripts/get_build_warnings.py	Paginated CDash GraphQL fetch for per-build warnings; pagination, exclusion filters, and output modes are all correct.
.github/skills/fix-nightly-warnings/scripts/list_nightly_warnings.py	Lists CDash builds with warnings/errors; logic for nightly-start window, sorting, and --new-warnings-only baseline heuristic is correct.
.github/skills/fix-nightly-warnings/scripts/triage_nightly.py	Orchestrates list + fetch into a deduplicated triage report; has a P1 JSON key inconsistency ("warnings" vs "warnings_by_flag" in the empty-builds branch) and silently swallows per-build fetch errors with no diagnostic output.

Sequence Diagram

sequenceDiagram
    participant Agent
    participant triage_nightly.py
    participant get_build_warnings.py
    participant CDash GraphQL

    Agent->>triage_nightly.py: run (--type, --since, --json)
    triage_nightly.py->>CDash GraphQL: project builds query (QUERY_TEMPLATE)
    CDash GraphQL-->>triage_nightly.py: edges [{node: {id, name, warnings, errors}}]
    loop for each build
        triage_nightly.py->>get_build_warnings.py: fetch_entries(build_id, WARNING, limit)
        get_build_warnings.py->>CDash GraphQL: build warnings query (paginated)
        CDash GraphQL-->>get_build_warnings.py: warning entries
        get_build_warnings.py-->>triage_nightly.py: (build_meta, entries)
        triage_nightly.py->>triage_nightly.py: deduplicate by (sourceFile, flag)
    end
    triage_nightly.py-->>Agent: JSON {builds_inspected, builds, warnings_by_flag}

Loading

_{Reviews (1): Last reviewed commit: "STYLE: Address review comments on fix-ni..." | Re-trigger Greptile}

[greptile-apps]

Inconsistent JSON key in empty-builds case

When no builds are found, the JSON output uses the key "warnings", but the normal execution path uses "warnings_by_flag". Any downstream consumer (e.g., an agent or jq pipeline) that reads .warnings_by_flag will get null for the empty case instead of [], silently producing incorrect behaviour.

Suggested change

	if not builds_with_warnings:
	if args.json_output:
	print(json.dumps({"builds_inspected": 0, "warnings": []}, indent=2))
	else:
	print("No builds with warnings found.", file=sys.stderr)
	return
	if not builds_with_warnings:
	if args.json_output:
	print(json.dumps({"builds_inspected": 0, "warnings_by_flag": []}, indent=2))
	else:
	print("No builds with warnings found.", file=sys.stderr)
	return

[greptile-apps]

Silent failure when all warning-fetches fail

fetch_entries calls sys.exit(2) on any network error. Catching SystemExit swallows those errors, so if every per-build fetch fails the script exits with code 0 and produces an empty result — the agent will interpret this as "no warnings" and skip the triage step entirely. At minimum, a warning should be printed for each skipped build so the caller can tell whether the result is complete.

Suggested change

	try:
	_meta, entries = fetch_entries(
	str(build_id), "WARNING", args.limit_warnings
	)
	except SystemExit:
	continue
	try:
	_meta, entries = fetch_entries(
	str(build_id), "WARNING", args.limit_warnings
	)
	except SystemExit:
	print(f"Warning: failed to fetch warnings for build {build_id} ({build_name}), skipping.", file=sys.stderr)
	continue

[greptile-apps]

Checklist items won't render as checkboxes

GitHub Flavored Markdown requires the - [ ] syntax for interactive checkboxes. The bare [] used here will render as plain text, not tick-boxes.

Suggested change

	[] All targeted warnings are fixed
	[] No new warnings or errors introduced
	[] Changes are limited to the files affected by the warnings
	[] Commit message clearly describes the fix and references the CDash issue if applicable
	- [ ] All targeted warnings are fixed
	- [ ] No new warnings or errors introduced
	- [ ] Changes are limited to the files affected by the warnings
	- [ ] Commit message clearly describes the fix and references the CDash issue if applicable