Infrahub Issue Reporter

Overview

Help the user file a bug report or feature request against the correct Infrahub-ecosystem repository on GitHub. The skill does not auto-submit — it prepares a complete, sanitized issue and stops at the user's review gate. Submission only happens after the user explicitly approves both the content and the submission method.

The ecosystem has 11 candidate repos. Most users will not know which one owns their problem. The skill's job is to make a confident guess from local-repo context, confirm it with the user, search for duplicates, and produce a draft that matches the target repo's intake form.

When to Use

Trigger this skill when the user says things like:

• "I think I found a bug in Infrahub"
• "How do I report this issue?"
• "I'd like to request a feature"
• "Where should I file this?"
• "This isn't working — I want to open a ticket"

Do not trigger if the user is debugging locally and hasn't expressed an intent to file an issue. Filing is the action that distinguishes this skill from general troubleshooting.

Do not trigger directly when the friction is with an Infrahub skill's own guidance (a rule that's wrong, missing, or confusing) rather than the Infrahub product itself. That belongs to infrahub-reporting-skill-gaps, which checks the tracker for an existing report and decides bug vs. feature before handing a payload to this skill.

Skill-gap intake

This skill may also be invoked directly by infrahub-reporting-skill-gaps with a prepared payload: {type, title, body, searched, issue?}. For that caller, type is bug, feature, or docs gap, already decided by that skill's two-level triage.

The payload carries no repo, and routing is yours. That caller deliberately names no destination, so this table is the only place the mapping lives:

type	Repo
bug	opsmill/infrahub-skills
feature	opsmill/infrahub-skills
docs gap	opsmill/infrahub

A bug or a feature is about a rule file, which lives in the skills repo. A docs gap is about Infrahub's own documentation, which lives in the main repo, and the skills maintainers have no power to write it. Resolve the repo from this table before step 4, then continue as though step 3 had run.

Titles use bug(docs): for the docs-gap case, without a [skill] segment, since the originating skill means nothing to opsmill/infrahub's maintainers; see reference.md for that repo's own title notes and the bug:/feat: convention this pairs with.

searched names the repo the caller already searched at its own step 2, and issue is optional. When issue is present, the caller matched an existing issue: treat it as the target, render body as a comment, and post with gh issue comment at step 8 rather than gh issue create.

Step 4 depends on searched. Skip it when searched equals the repo you just resolved: the caller has already run that exact search, and repeating it risks a second answer that contradicts the one already written into body. Run it as normal when they differ, which is the docs-gap case: the caller searches opsmill/infrahub-skills before it knows the kind, but a docs gap routes to opsmill/infrahub, so that tracker has not been checked by anyone yet.

When invoked this way, classification (step 2) is already decided, skip it. Step 3's detection logic does not apply either; the table above replaces it, and there is nothing to confirm with the user since the kind was already established from the session's own evidence. Skip environment info gathering (step 5) too; it collects Infrahub server, SDK, and OS versions, which apply to neither a report about a skill's own guidance nor a report about missing documentation. Use the caller's title and body as the render for step 6 instead of a generic template. Continue through steps 7-9 as normal: review gate, submission method, confirm.

Workflow

Follow these steps in order. Stop at every user-gate step — never proceed past one without explicit confirmation.

1. Capture the problem

Ask the user to describe the problem in their own words if they haven't already. Don't probe yet — just listen. Take note of any product names, version numbers, error messages, or workflow context they mention; they feed every later step.

2. Classify

Decide whether this is a bug, feature request, or question/usage help. Use 1-2 follow-ups only if classification is ambiguous:

• "Did this work before and now doesn't, or is it something that's never existed?"
• "Do you have a reproduction case, or is this an intermittent observation?"
• "Is the desired behavior documented anywhere, or is this new functionality you're asking for?"

If the user describes a workaround they're already using, lean toward "feature request" (the bug has a workaround → real problem is the missing ergonomics). If the user describes a crash, wrong output, or 500-level error, lean toward "bug". If the user is asking how to do something, route them to Discord/discussions instead of filing an issue — don't open issues for support questions.

3. Route to a repo

Read reference.md for the full 11-repo registry, detection cues, and template availability. Apply this detection logic in order:

1. Read the local working directory for filesystem cues (infrahub.toml, pyproject.toml deps, galaxy.yml, .claude-plugin/plugin.json, etc.)
2. Cross-reference detected cues with what the user said. A user complaining about Ansible playbooks in a repo with no Ansible cues is still an Ansible issue.
3. Pick the most specific match.

The default-to-main rule: if detection is ambiguous, if multiple repos match equally, if the user expresses any uncertainty about which component is at fault, or if the symptom plausibly lives at the platform layer — default to opsmill/infrahub. The main repo can re-route an incoming issue to a sub-repo more easily than a sub-repo can re-route back. Erring toward main is cheaper than erring toward a wrong specialized repo.

Present your guess to the user with a one-sentence rationale and ask them to confirm or correct. Example:

"Based on infrahub-sdk in your pyproject.toml and the error mentioning InfrahubClient, this looks like an opsmill/infrahub-sdk-python issue. Confirm?"

4. Search for duplicates

Always search both bugs and features. Use the first available method:

1. gh CLI — gh search issues --repo <owner/repo> --state all "<keywords>". Pull keywords from the user's description plus any error message strings. Run a second pass with synonyms if the first returns nothing.
2. MCP GitHub server — if gh is not installed, use whichever GitHub MCP tool is available in the user's environment.
3. Manual — if neither is available, give the user the GitHub search URL and ask them to check.

Show the user the top 3-5 matches with title, state (open/closed), and a one-line excerpt. Ask:

"Is your issue covered by any of these? If yes, we'll add a comment instead of opening a new one."

If a match exists, switch to comment mode and choose one of two paths. State which one to the user:

1. Complement: this occurrence adds information the existing issue lacks, such as a different trigger, a second affected skill or component, or a new failure shape. Comment with the new information only; do not restate what the issue already says.
2. Plus one: this occurrence adds nothing new. Post a short comment recording another hit, so maintainers can gauge frequency, without repeating the existing description.

Either way, give the user the issue URL. Otherwise, proceed to draft a new issue.

5. Gather environment info (bugs only)

For features, skip this step. For bugs, collect only:

• The relevant product version(s) — e.g., Infrahub server version, SDK version, Ansible collection version, plugin version.
• Operating system + architecture (e.g., macOS 14 arm64, Ubuntu 22.04 x86_64).
• Python or runtime version if relevant to the component.

That's it. Do not collect file paths, project structure, logs verbatim, env dumps, docker compose ps output, or anything that could leak internal state. The environment-info-sanitization rule governs what is and is not allowed in the issue body — read it before composing the bug section.

Where to find versions:

Component	Command
Infrahub server	infrahubctl info (Infrahub Version line), or check image: tag in compose file
Python SDK	infrahubctl info (SDK Version line), or pip show infrahub-sdk
Ansible collection	ansible-galaxy collection list opsmill.infrahub
Nornir plugin	pip show nornir-infrahub
Helm chart	helm list -n <namespace>
infrahub-sync	infrahub-sync --version
infrahub-backup	infrahub-backup --version
MCP server	check MCP client config or pip show infrahub-mcp
VS Code extension	Extensions panel in VS Code
Skills plugin	.claude-plugin/plugin.json version field

If a version can't be determined cleanly, mark it unknown rather than guessing or running broader discovery commands.

6. Render the issue

If the target repo has a .github/ISSUE_TEMPLATE/ directory, fetch the matching template (bug_report.yml for bugs, feature_request.yml for features) and follow its form fields. Use:

gh api repos/<owner>/<repo>/contents/.github/ISSUE_TEMPLATE/bug_report.yml --jq '.content' | base64 -d

Repos that ship issue templates:

• opsmill/infrahub
• opsmill/infrahub-sdk-python
• opsmill/infrahub-ansible
• opsmill/nornir-infrahub

Repos without templates fall back to templates/bug.md or templates/feature.md in this skill.

Title conventions: use the form <area>: <imperative summary>, keep it under 72 characters, no trailing punctuation. Examples:

• SDK: InfrahubClient.create() raises on empty list payload
• Schema: hierarchical generic loses parent on rename
• Docs: clarify human_friendly_id behavior for inherited nodes

7. User review gate (mandatory)

Present the rendered issue to the user. Show:

• Target repo and URL
• Final title
• Full body (markdown)
• Whether this is a new issue or a comment on an existing one

Ask: "Does the content look right? Anything to add or remove before we proceed?"

Do not skip this step. Iterate with the user until they explicitly approve the content.

8. Pick a submission method

Once content is approved, ask the user how they want to submit. The commands differ for a new issue and a comment; pick the column that matches what step 7 showed:

Method	New issue	Comment on an existing issue
CLI	gh issue create --repo <owner/repo> --title "..." --body "..."	gh issue comment <number> --repo <owner/repo> --body "..."
MCP GitHub server	Use it if the user has one. Confirm the tool name before invoking.	Same, using the server's comment tool.
Manual	Print title and body as copy-paste-ready markdown, plus https://github.com/<owner>/<repo>/issues/new.	Print the body as copy-paste-ready markdown, plus https://github.com/<owner>/<repo>/issues/<number>.

A comment carries no title. If the render still has one, step 6 built the wrong artifact; go back rather than pasting a title into a comment body.

Never assume a default — always ask. After the user picks a method, execute it.

9. Confirm

After submission, give the user:

• The issue URL, or the comment's permalink when the submission was a comment (for gh/MCP submissions)
• A one-line summary of what was filed or commented
• A reminder that they can subscribe to the issue for updates

Rule Categories

Prefix	Category	Description
env	Environment info	What can and cannot appear in issue bodies

See rules/_sections.md for the index.

Supporting References

• reference.md — the 11-repo registry with detection cues, descriptions, and template availability. Read this in step 3.
• templates/bug.md — generic fallback bug template for repos without a .github/ISSUE_TEMPLATE/.
• templates/feature.md — generic fallback feature template.
• rules/environment-info-sanitization.md — what to redact before submitting. Security-critical; read this for every bug report.
• ../infrahub-common/rules/workflow-information-priority.md -- Skill content first; how to consult docs.infrahub.app on a genuine gap (e.g. deleting nodes)

Anti-patterns

• Submitting without explicit user approval. The user gate in step 7 is not optional. A surprise GitHub issue on a public repo is hard to unwind.
• Filing in a sub-repo when the symptom is ambiguous. When in doubt, file in opsmill/infrahub and let the maintainers re-route. Wrong sub-repo issues create more work for everyone.
• Filing a duplicate. If a search match exists, default to adding a comment with new info instead of opening a new issue.
• Pasting full logs or docker compose output. These leak hostnames, IPs, container names, and sometimes secrets. Stick to versions + OS for v1.
• Filing support questions as bugs. "How do I do X?" belongs in Discord or GitHub Discussions, not Issues.