beomeodev

webapp-testing

Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.

codebase-snapshot

Agent-facing codebase exploration and SYSTEM_MAP maintenance skill. Use when starting a new task, creating or refreshing docs/SYSTEM_MAP.md, checking whether stored architecture knowledge is stale, or recording invariants, risk areas, and verification commands with git metadata. Structural facts (file lists, call relationships, hot paths) are answered by the Graphify code graph when graphify-out/graph.json exists — this skill maintains the curated prose the graph cannot express, with rg/find/git as the no-graph fallback.

Web & Frontend Listed

design-baseline

Generates a minimal, coherent, ready-to-ship design foundation (docs/design/DESIGN.md + tokens.css + a neutral base stylesheet) the first time a Feature creates a UI surface and no design baseline exists yet. Use when a Feature's in-scope deliverables include a web UI and there is no design input from the user — produces real default values (palette, type scale, spacing, WCAG floors), not design-judgment prose. Later UI work reads the generated tokens instead of hardcoding new values.

Code & Development Listed

git-worktrees

Full lifecycle discipline for isolated git worktrees — setup (detect existing isolation, prefer native tools like EnterWorktree, ignore-check before project-local creation, per-worktree dependency install, clean test baseline) and teardown (merge-before-remove-before-delete ordering, provenance-based cleanup, never remove from inside the worktree). Use when starting feature work that needs isolation from the current workspace, when running parallel Features (see parallel-features), or when finishing a branch and deciding merge/PR/keep/discard.

ops-debugging

Symptom-indexed playbook for deployment/environment debugging — SSH tunnels, reverse proxies, TLS, container lifecycle, secret rotation, and real-time API integration failures that "work locally, break deployed." Every entry traces to a real diagnosed incident (probe -> resolution), stated at the failure-class level so it generalizes across projects. Use when a deploy-only symptom, a "works here but not there" report, or an environment-boundary failure needs the fastest-discriminating probe, not generic debugging methodology (that lives in /ms.fix's Hard-Bug Discipline).

parallel-features

SPECTER-specific methodology for running two or more independent Feature cycles concurrently in separate git worktrees — eligibility test (DAG independence + plan-time file-overlap check), per-worktree vs shared-state rules, union-merge setup for append-shaped workflow files, and merge-back sequencing (rebase + gate re-run for the second merge). Use when the Feature Map DAG shows independent branches and the user wants to run them in parallel, or when deciding whether two Features are safe to parallelize at all.

prd-authoring

PRD-authoring skill that co-writes a Product Requirements Document with the user through a one-question-at-a-time interview, tagging gaps inline (assumption/open question) and producing output that /ms.featuremap and /ms.checklist can consume directly — explicit commitments, GEARS-compatible acceptance criteria, and a Testing Decisions section. Use when no PRD exists yet and the user wants to co-author one before /ms.pre-specter, or when extending an existing PRD with a new `## PRD Amendment N` section for /ms.expand.

session-handoff

Compose the minimal paste-ready resume block when a session ends with work remaining — context threshold approaching, a Feature finished while others remain, the user declares the session over ("세션 종료", "내일 이어서", "오늘은 여기까지"), or /ms.fin publishes with unfinished Features. Enforces Diet Constraints against handoff bloat. Trigger symptoms — load this skill when you are ABOUT to write a long end-of-session summary, inline today's debugging history "so nothing is lost", or list every lesson learned in the final message.

specter-agent-protocols

Canonical external-agent protocols shared by the dual-agent SPECTER commands (/ms.agent-verify, /ms.verify, /ms.analyze, /ms.review, /ms.codex-checklist, /ms.expand) — session-level preflight, single-agent degrade rule, report-write/salvage protocol, re-round convergence caps, and the auditor bias-prevention doctrine (context isolation, evidence-cited verdicts, UNVERIFIED marking, grade-down-on-doubt, defect-claim symmetry) plus the verification-report structure (Claim/Evidence/Baseline/Gaps/Residual-risk). Commands reference this file instead of restating the mechanics; each command keeps only its own report paths and station-specific invariants inline.

spike

Sanctioned throwaway exploration OUTSIDE the gated SPECTER workflow — time-boxed, never merged, no TAG/gate ceremony, and the only durable artifact is a findings note that feeds a PRD Amendment (/ms.expand) or a fix decision. Use when the user wants to test feasibility, compare approaches, or probe an unknown ("실험해보자", "프로토타입", "검증만 해보고 버리자", "될지 한번 쳐보자") before committing to spec ceremony — NOT for building anything intended to ship.

testing-skills-with-subagents

TDD applied to skill/process documents — pressure-test a discipline-enforcing skill with subagents BEFORE deploying it, by running realistic multi-pressure scenarios without the skill (RED, capture rationalizations verbatim), with it (GREEN), then closing each loophole (REFACTOR) until agents comply under maximum pressure. Use when creating or editing any gate/discipline skill in this repo — especially before /ms.sync broadcasts the edit to every registered project repo. Skip for pure reference skills with no rules to violate.

Data & Documents Listed

transcript-mining

Efficient, bounded mining of Claude Code session transcripts (~/.claude/projects/*/*.jsonl) for workflow audits, usage-pattern analysis, and skill distillation — JSONL anatomy, proven extraction queries (tool-use counts, duplicate reads, token aggregation, command sequences), the scan-only-named-files rule, and the content-vs-mechanics separation rule. Use when auditing workflow health, quantifying token waste, reconstructing what happened across sessions, or distilling incident knowledge from transcripts into a skill.

trust-validation

Comprehensive code quality validator enforcing TRUST 5 principles - Test-First (≥85% coverage with pytest/vitest), Readable (≤700 SLOC production files / tests no limit, ≤10 complexity), Unified (strict type checking with mypy/tsc), Secured (OWASP Top 10 implementation patterns + trivy/bandit scanning, cryptographic best practices, input validation, authorization checks, adversarial abuse-case review for IDOR/multi-tenant isolation), Trackable (complete TAG anchor chain integrity SPEC→TEST→CODE) with detailed compliance reports and remediation guidance. Use when validating quality before merging PRs, running quality gate checks, checking release readiness, validating TRUST compliance, or performing comprehensive code reviews