Developer-side AI tooling under .claude/ for catching weak/missing tests pre-submit

[PaulMRamirez]

Hi @LeStarch following up on what I showed you last night. Putting this here for broader CCB visibility before any code lands.

What this is

A proposed .claude/ directory containing Claude Code skills, subagents, and slash commands that help contributors catch weak or missing tests in their own PRs before maintainer review. Two main capabilities:

• /fprime-test-check — audits the diff. Every new FPP command, event, telemetry, param, input port, and output port is checked for a corresponding sendCmd_, ASSERT_EVENTS_, ASSERT_TLM_, paramSet_, invoke_to_, or ASSERT_from_ reference under test/ut/. A quality pass scans for weakening patterns (DISABLED_, GTEST_SKIP, removed/commented assertions), missing doDispatch() on async paths, count-only checks without value verification, and tautological asserts.
• /fprime-pr-prep — one-shot pre-submit gauntlet: build, unit tests, coverage, quality, with a copy-pasteable readiness summary.

It is developer-side only — intended as a complement to the existing Copilot review agent at .github/agents/fprime-code-review.agent.md, not a replacement, and not touching CI.

Where to look

• Proposal branch: https://github.com/PaulMRamirez/fprime/tree/proposal/claude-dev-tooling/.claude
• Diff vs devel: devel...PaulMRamirez:fprime:proposal/claude-dev-tooling
• AI disclosure: see .claude/README.md (https://github.com/PaulMRamirez/fprime/blob/proposal/claude-dev-tooling/.claude/README.md).
  TL;DR — authored with Claude Code, model Claude Opus 4.7; AI-drafted via iterative prompting, human-reviewed, lightly tested by me with Claude Code only (not exercised through Copilot).

Why a Discussion, not a PR

CONTRIBUTING.md is explicit that non-trivial work needs a CCB-approved issue first, and this is cross-cutting tooling rather than a
localized code change. I'd rather get the framing right before asking for review time.

Open questions for the CCB

1. Does F´ want this in-tree at all? Plausible homes I see: nasa/fprime/.claude/, an fprime-community/fprime-claude add-on, or under
   fprime-tools.
2. Copilot parity. The team uses Copilot, not Claude Code. Should an equivalent set of artifacts (or a shared rubric consumed by both)
   be a prerequisite to anything landing?
3. Maintenance ownership. The FPP-to-test mapping needs to track F´ as it evolves. Who keeps it honest?
4. Relationship to the existing Copilot review agent. Shared rubric, or independently maintained?

Ask

I'm not asking for a merge. I'm asking whether this direction is wanted at all, and in what form. Happy to file an issue, narrow the
scope, or move it to a community repo based on CCB input.

[PaulMRamirez]

You probably don't need the README as this was just to describe what this is for now. Also we'd want to put this in the more generally picked up place so people could inherit this when they checked out the repository and using other things than Claude Code (e.g. GitHub Copilot, Devin, etc.)

[thomas-bc]

I like where this is going and agree that we need to start getting on top of those things. The F Prime team has been bombarded with AI-authored input for a while, and it's becoming a struggle to keep up with it.

For your questions:

1. Yes we want it!
2. Yes we should make the setup as tool-agnostic as possible. Should at least support the following: Claude Code, Copilot, Windsurf
3. I don't know what this means. Maintainers can keep these files up-to-date if that's the question. Maintainers should fully review all agents/skills that go in as a first pass, even if that means limiting the number of skills/agents. And then we can keep iterating.
4. I don't fully understand the question, but yeah Copilot is kinda there, they should ideally share the agents definitions.

[thomas-bc]

In progress: #5163