holmesrm88
- 8 skills
- 0 followers
- 3 hours ago last updated
- ▌ Harness · holmesrm88 bundleRun a disciplined TDD implementation loop against a Jira story or a small idea. Two phases — init decomposes the work into small, independently committable features and generates a deterministic pre/post test gate; code implements one feature at a time through a red-green-commit cycle with human approval at each step. Use when the user says "harness init", "harness code", "start the harness", "let's implement PROJ-1234", "break this story into features", or otherwise wants to begin or continue structured implementation work on a specific story. Also use to resume — the skill reads its own state from disk and picks up mid-feature. Do NOT use for deciding whether work needs planning (use bmad-intake) or for reviewing code that already exists (use java-code-review).
- ▌ Repo Recon · holmesrm88 bundleSystematically map an unfamiliar codebase and produce a written architecture brief covering build system, module topology, entry points, data layer, test posture, team ownership, and a risk register. Use this whenever the user is orienting to code they did not write — onboarding at a new job, inheriting a service, picking up a legacy repo, or asking things like "help me understand this codebase", "what does this service actually do", "give me the lay of the land", "where do I start", or "map this repo for me". Use it even when the request sounds casual or the user only points at a directory without saying the word "architecture". This skill is for orientation to a whole codebase or subsystem, NOT for critiquing the quality of a specific diff or class (use java-code-review for that).
- ▌ Bmad Intake · holmesrm88 bundleTriage a unit of development work before starting it — decide whether it warrants the BMad Method planning process, goes straight to implementation, needs codebase reconnaissance first, or is really a human conversation rather than a technical task. Then prime brownfield context and hand off to the installed BMad workflow. Use this at the START of any non-trivial development work — when the user says "I need to add/build/change X", "let's start on Y", "kick off BMad for Z", "should we plan this out", or hands over a ticket, epic, or feature request. Use it even when BMad is not mentioned, since deciding not to use it is a valid and common outcome. Do NOT use this for understanding an existing codebase with no change in mind (use repo-recon) or for evaluating code that already exists (use java-code-review).
- ▌ Review Panel · holmesrm88 bundleOrchestrate a panel of isolated review subagents over a completed story's full diff, then triage their findings into new harness features, tickets, or inline fixes. Runs security, ghost-test, clean-code, and correctness reviewers in parallel with no cross-talk, deduplicates and ranks what they find, and presents one batch for human approval. Use when all features for a story are committed and the work is heading toward a PR — "run the panel", "review the story", "we're ready for PR", "check PROJ-1234 before I push". Do NOT use for reviewing a single file or a work-in-progress diff (use java-code-review directly) or for implementing features (use harness).
- ▌ Dev Env Verify · holmesrm88 bundleVerify a story's acceptance criteria against the ephemeral dev environment created by its pull request — locate the environment URL, build an explicit check list from the acceptance criteria, execute the checks through the browser or the API, and report pass/fail with evidence. Use after a PR's build succeeds and before marking it ready for human review — "test the dev env", "verify PROJ-1234 in dev", "does the preview work", "check the deployed changes". Do NOT use for local testing before a PR exists, or for diagnosing a failed build (use gh-actions-triage).
- ▌ Story Estimator · holmesrm88 bundleEstimate story points for unpointed Jira stories ahead of a grooming session, calibrated against how the team's past work actually went rather than against generic heuristics. Also runs in calibration-only mode to learn a team's pointing patterns from completed stories and surface where estimates and reality diverge. Use when preparing for grooming, backlog refinement, or sprint planning — "point the backlog", "estimate these stories", "prep for grooming", "how long will this take", "calibrate the estimator". Do NOT use to evaluate people's performance or throughput; this skill operates at team level only and refuses individual analysis.
- ▌ Java Code Review · holmesrm88 bundleReview Java code against a senior-engineer checklist covering correctness, resource handling, concurrency, exception discipline, API design, persistence, and security, and return prioritized findings with file locations and concrete fixes. Use this whenever the user asks for review, critique, feedback, or a second opinion on Java or Kotlin code — a pull request, a diff, a single class, a method, or "does this look right?" — and also proactively before the user submits Java code they have just written or asks whether it is ready to ship. Use it even when the request is casual ("take a look at this", "anything wrong here?"). This is for evaluating specific code that already exists, NOT for mapping an unfamiliar codebase (use repo-recon) and NOT for writing new code from scratch.
- ▌ Gh Actions Triage · holmesrm88 bundleDiagnose a failed GitHub Actions run — fetch the failed-step logs, find the actual error in the noise, classify it as a real failure, a flake, base-branch breakage, config drift, or infrastructure, and propose a specific fix. Use whenever CI is red — "the build failed", "CI is broken", "why did the pipeline fail", "check the Actions run", "my PR won't build" — or after pushing when a run needs checking. Do NOT use for local test failures that never reached CI, or for reviewing code quality (use java-code-review).