borinquenkid opened a new pull request, #15977:
URL: https://github.com/apache/grails-core/pull/15977

   ## Summary
   
   Docs/skills-only changes (no build or runtime code touched) improving how AI 
coding agents work in this repo. Every change is grounded in actual source 
investigation, not assumed — several corrected findings that turned out to be 
wrong on first pass (see commit messages for the specifics).
   
   - Fixed a dangling `codebase-memory` skill reference in `AGENTS.md` pointing 
at a file that never existed
   - Replaced the hardcoded skill-to-task mapping in `AGENTS.md` with 
directory-based discovery — skills are found via their own front-matter 
`description`, not a hand-maintained list that drifts out of sync every time a 
skill is added or removed
   - Added an optional `paths:` front-matter field so module-scoped skills are 
discovered structurally (by the files being touched), not just by 
description-matching
   - Added `hibernate-developer`-style skills for `grails-data-mongodb`, 
`grails-forge` (a Micronaut application, not Grails — corrects a root rule 
that's actually inverted there), `grails-gradle`, and `build-logic` — four 
real, previously-undocumented subsystems, each verified against actual source 
and, where relevant, actual test runs
   - Added scoped `AGENTS.md` files for `grails-forge/`, `grails-gradle/`, and 
`build-logic/` correcting root rules that don't transfer to those subprojects 
(e.g. `@GrailsCompileStatic` doesn't apply to any of them; 
`grails-gradle`/`build-logic` share root's dependency versions while 
`grails-forge` doesn't)
   - Linked `grails-developer` to `grails-test-examples/`, the ~50 real 
functional test apps that had zero skill coverage despite several sharing a 
name with a framework-internals module (e.g. `grails-test-examples/hibernate7/` 
vs. `grails-data-hibernate7/`)
   - Made adversarial self-review a required PR-checklist step, framed around 
what every contributor can actually do (a fresh-context, explicitly 
refute-first review pass) rather than requiring a second AI vendor
   - Documented that Codecov's underlying coverage data (JaCoCo) is fully 
available locally without a CI round-trip — it was already produced by the 
build, just never connected to "Codecov" by name anywhere
   - Added a `diff-coverage-check` skill computing real diff coverage (only the 
lines you changed, not whole-class coverage) entirely locally by 
cross-referencing JaCoCo XML against `git diff` — zero new build dependency, 
verified end-to-end against a real mixed-coverage diff
   - Added a `worktree-hygiene` skill for `.claude/worktrees/` sprawl, using 
PR/remote state as the staleness signal rather than commit age (a worktree can 
go idle between review rounds without being abandoned)
   - Added a `migration-scoping` skill for pre-work architectural triage, 
grounded in a real four-branch sprawl case (from the Hibernate7/GORM registry 
migration) found and cleaned up during this work
   - Defaulted PR guidance to a single large PR over a reviewability-stack, 
since GitHub's Copilot reviewer silently skips PRs over ~300 files and a 
stack's sub-PRs never get their combined integration reviewed as a unit
   
   ## Attribution
   
   The skill packaging format (`SKILL.md` front-matter, 
discovery-by-description convention, cross-agent `compatibility` field) follows 
the [Agent Skills Specification](https://github.com/anthropics/skills) — 
Anthropic's public spec and reference implementation for portable, cross-agent 
skill packages. Nothing in this PR is Claude-specific; every skill's 
front-matter lists `opencode, claude, grok, gemini, copilot, cursor, windsurf` 
as compatible tooling.
   
   ## Test plan
   
   - [x] Every class name, file path, task name, and build command cited in 
these docs was checked against actual source (`grep`/`Read`), not assumed
   - [x] Real test runs verified non-obvious claims where feasible — e.g. ran 
`:grails-i18n:test` to confirm the `diff-coverage-check` skill's script 
produces correct output against a real mixed-coverage diff, ran 
`grails-forge`/`grails-gradle`/`build-logic` builds to confirm real commands
   - [x] Discovery mechanism (`.agents/skills/*/SKILL.md` front-matter scan) 
tested against the full current skill set after each addition
   - [x] No `build.gradle`, `settings.gradle`, or source file was modified — 
this PR cannot change build or runtime behavior


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to