Claude Skills by hmj1026

Project initialization inventory (one-time), for first onboarding a project or rebuilding cache after structural changes. Not for: day-to-day development (read cache directly), finding specific files (use dhpk-codebase-exploration). Output: project map with entrypoints + test map + next steps.

Uncommitted code risk assessment with breaking change detection, blast radius analysis, and scope metrics. Use when: evaluating PR risk, pre-commit risk check, large refactoring review. Not for: security vulnerabilities (use /codex-security), code correctness (use /codex-review-fast). Output: 3-dimension weighted score + risk level + gate.

Root-cause investigation workflow for unexpected behavior, test failures, performance regressions, data inconsistencies, and cross-layer data-flow tracing. Use when users ask to investigate or trace a bug, regression, flaky test, or root cause (including 測試失敗、效能異常、調查 Bug、追蹤資料流). Not for direct implementation when the root cause is already confirmed, ordinary code review, or pure feature work. Output: a phase-based evidence report, confirmed root cause or explicit blocked state, solution optio...

Scans installed skills to extract cross-cutting principles and distill them into rules (append, revise, or create rule files), for periodic rules maintenance, after installing new skills, or when a dhpk-skill-stocktake surfaces patterns that should be rules. Not for: creating skills (use dhpk-create-skill), one-off rule edits, code review. Output: candidate report (principle + evidence + verdict) for user approval.

Security review (OWASP Top 10), codex-free. Use when: security audit, dependency vulnerability check, security-sensitive changes. For a Codex-driven audit, use /dhpk:codex-security instead. Not for: code review (use dhpk-change-review), test review (use dhpk-test-review). Output: security findings + audit report.

Audits dhpk usage evidence in the current user home, correlates session failures with installed agents and package versions, verifies candidate defects, and prepares deduplicated GitHub issue evidence. Not for: remote fleet scans or automatic source changes. Output: redacted report, verified findings, and confirmed issue handoff. Explicit invocation only.

Validate skill quality against routing, progressive loading, and verification criteria. Use when: auditing skills, checking skill health, reviewing skill design. Not for: code review (use dhpk-change-review) or doc review (use dhpk-doc-review). Depth split: dhpk-skill-health-audit = structural lint of one skill · dhpk-skill-quality-judge = deep rubric of one · dhpk-skill-stocktake = batch-audit of many. Output: health report with per-skill ratings + Gate.

Use when deeply evaluating one SKILL.md or package against official design practice with a scored rubric. Not for structural lint, batch stocktake, or creating/refactoring a skill. Output: per-dimension grade, critical issues, and prioritized improvements.

Search existing local, marketplace, GitHub, and web skill sources before creating a new skill. Use when: the user wants to create, build, fork, or find a skill for a workflow. Not for: authoring a skill (use dhpk-create-skill), auditing skill quality (use dhpk-skill-health-audit). Output: ranked candidate table + use/fork/create recommendation.

Batch-audit installed Claude skills and commands for quality, overlap, and staleness via subagent evaluation. Use when: auditing many skills at once, periodic stocktake, Quick Scan or Full Stocktake mode. Not for: linting one skill's structure (use dhpk-skill-health-audit), a deep rubric grade of one skill (use dhpk-skill-quality-judge), authoring (use dhpk-create-skill). Depth split: dhpk-skill-health-audit = structural lint of one skill · dhpk-skill-quality-judge = deep rubric of one · dhpk...

Framework-agnostic test-driven development guidance for behavior-first unit and integration tests, test scaffolds, and minimal RED-GREEN-REFACTOR changes. Use when building a feature or fixing a bug test-first, writing a test scaffold, or reviewing test seams and mocks. Not for: Playwright journey authoring, pure documentation or rename work, or replacing dhpk's tdd-guide dispatch. Output: a failing-test proof, minimal implementation/test change, scoped verification, and a concise TDD report.

Tech spec generation and review. Use when: designing features, writing specs, spec review. Not for: requirements analysis (use /dhpk:deep-analyze), implementation (use dhpk-adaptive-dev-workflow in feature mode), architecture advice (use dhpk-codex-architect). Output: numbered tech spec document.

Test coverage review via Codex MCP. Use when: reviewing test sufficiency, identifying coverage gaps, test quality audit. Not for: generating tests (use /dhpk:codex-test-gen), code review (use dhpk-change-review). Output: coverage analysis + gap report.

Route code-exploration requests to gitnexus, cx, claude-mem, Read, or Grep. Use when locating a symbol, tracing callers or impact, exploring an unfamiliar module, planning a rename/refactor, checking change scope, or recalling a past decision. Not for plain-text/log/doc search, non-code questions, or a tool already chosen. Output: one primary tool, exact command or MCP call, and a named fallback.

Adaptive delivery workflow for substantial changes. Use when: a feature, bugfix, refactor, security, perf, or OpenSpec change needs one classification into Feature Delivery, Bug Investigation & Fix, or Lightweight Maintenance, with branch behavior and gates before heavy context loads. Not for tiny edits, investigation already underway, code review, or apply-ready OpenSpec tasks. Output: workflow classification + branch artifacts + delivery-loop gate checklist.

Full-stack diagnostic for agent and LLM applications — audits the 12-layer agent stack for wrapper regression, memory pollution, tool-discipline failures, hidden repair loops, and rendering corruption. Use when: shipping an agent/LLM feature, an agent degrades after adding wrapper/memory/tool layers, the same model works in the playground but breaks in your wrapper, or debugging agent behavior >15 min with no root cause. Not for: general code debugging, code review, or security scanning (use ...

Delegates git smart commit to agy-cli''s dhpk-git-smart-commit skill, offloading commit batching to Antigravity. Not for: a single simple commit, or when agy-cli is unavailable. Output: grouped commits created by agy-cli plus a git log summary.

Use when offloading a precise self-contained mechanical task to the agy CLI when available. Require target files, intent, and verification command; keep output in the agy-fast-worker subagent. Not for ambiguity, conversation-dependent work, or default fast-worker use. Output: verified edited files or honest BLOCKED evidence.

Code review using Codex MCP. Use when: PR review, code audit, or a second opinion on changes. Not for: document review (use dhpk-doc-review), security-specific audit (use dhpk-security-review), or test coverage review (use dhpk-test-review). Output: severity-ranked findings with file:line evidence, a reviewer-degradation state, and a fail-closed merge gate.

Claude Code config health check + plugin sync: audits .claude/ structure, checks naming, verifies hook setup, detects plugin version drift, and syncs installed assets. Not for: skill quality (use dhpk-skill-health-audit), code review (use dhpk-change-review). Output: health report + fix recommendations.

Trigger: explicit provider-qualified CLI dispatch-context construction and launcher execution. Avoid: direct adapter invocation or implicit role, mode, path, or authority derivation. Output: immutable dhpk.cli.context.v1 evidence or BLOCKED diagnostics.

Internal Codex and AGY CLI transport that validates an already-attested request and produces a contained terminal receipt; it never selects roles or accepts direct user invocation.

Explore an unfamiliar codebase with a focused symbol/flow trace, optionally run an independent second perspective, or request a depth-controlled explanation. Use when: tracing execution, understanding architecture, or diagnosing a code path. Not for: change review, security audit, or implementation. Output: evidence-backed flow findings with explicit gaps.

Codex architecture consulting, including bounded adversarial option convergence. Use when: designing features, evaluating architecture, comparing options, or running independent proposal and critique rounds. Not for: implementation (use dhpk-codex-implement), code review (use dhpk-change-review). Output: architecture advice + design recommendations.

Use when CODEX=on and a self-contained bulk task or blind second opinion should go to the GPT-5.6 family through one-shot codex exec: gpt-5.6-sol/high for read-only or gpt-5.6-luna/xhigh for workspace-write. Not for context-dependent or iterative work, or structured MCP codex-* review loops. Output: the bounded, redacted codex exec result.

Implement features via Codex MCP. Use when: writing new code from specs, implementing features, Codex-driven development. Not for: code review (use dhpk-change-review), architecture advice (use dhpk-codex-architect). Output: implemented code + review loop.

Composer-published package contract review. Use when: deciding semver, auditing public API or composer.json, validating Laravel package discovery, or coordinating a package release. Not for: everyday application code, language idioms, or Laravel-specific authoring mechanics. Output: a contract verdict, audited manifest, or release gate backed by the package-contract reference.

Create, update, or scan per-task request tickets for progress tracking. These are date-prefixed non-lifecycle docs under requests/, NOT feature-level requirements (use /dhpk:deep-analyze for those). Use when: tracking task progress, updating completion status, scanning incomplete requests, checking request status dashboard. Not for: feature-level problem-space analysis (use /dhpk:deep-analyze for 1-requirements.md lifecycle doc), tech specs (use dhpk-tech-spec), code implementation (use dhpk-...

Creates a new Claude Code skill, or refactors an existing one to dhpk conventions (lean SKILL.md + progressive disclosure, trigger-rich description, validation gate), from a request to "create a skill", "make a skill", "refactor this skill", or to capture a workflow as a reusable skill. Not for: writing a slash command (use command-creator) or auditing skill quality only (use /check-skill). Output: a validated skill package with an explicit diff summary and routing/verification evidence.

Compares Claude-first configuration across Codex, Antigravity, AGY native plugins, and Cursor to align cross-platform skills, commands, agents, hooks, or orchestration. Produces reviewable plans, OpenSpec tasks, dry-run/apply reports, and a configured-platform PASS/FAIL/BLOCKED validation gate (NOT_CONFIGURED/SKIP_INCOMPATIBLE stay visible without failing it; legacy PARTIAL is a deprecated compat field). Not for reverse sync, single-platform edits, or missing Claude source.

Cross-project deploy file-list generator (schema=v1). Use when: building a deploy/release checklist from git history, listing which files to ship for a deploy, 部署清單, deploy list/checklist. Scope via --anchor (in-code markers), --deploy-commits (pinned commits), or whole base..head; filters dev-only paths (tests, docs, CI, AI harness) and groups by ecosystem preset (php-yii, laravel, node, python, generic, or a custom preset). Not for: plain commit lookup (git log), raw diffs (git diff), or ac...

Document review via Codex MCP. Use when: reviewing .md docs, tech spec audit, document quality check. Not for: code review (use dhpk-change-review), test review (use dhpk-test-review). Output: 5-dimension rating table + gate.

End-of-task self-check for substantive repository changes. Use after the final Edit/Write before replying or smart-commit, especially for source, SQL/repository/migration, auth/crypto/money/upload, high-volume-table, frontend/template-script, harness, policy, or documentation changes. Not for trivial typo-only edits or research/planning with no pending commit. Output: a gated wrap-up reply with changed files, verification, and risks/open questions.

Execution-policy router for dhpk software-engineering work. Use when choosing a task flow or OpenSpec route at kickoff, checking post-edit reviewer obligations, recovering from an anti-loop stop, or shaping a blocked/completion reply. Not for: substantial-change classification (use dhpk-adaptive-dev-workflow), an investigation already underway (use dhpk-root-cause-investigation), pure code tracing, or mid-workflow execution. Output: one next route plus the required gates and completion shape.

Feasibility analysis from first principles. Use when: evaluating solutions before dhpk-tech-spec, comparing approaches, risk assessment. Not for: implementation (use dhpk-adaptive-dev-workflow in feature mode), architecture advice (use dhpk-codex-architect). Output: quantitative comparison + recommendation.

Feature verification (READ-ONLY, P0-P5). Use when: verifying feature behavior after deployment, validating API responses, diagnosing production issues, post-deploy smoke test. Not for: modifying data (use dhpk-adaptive-dev-workflow in feature mode), code review (use dhpk-change-review), writing tests (use dhpk-tdd-workflow), security audit (use dhpk-security-review). Output: a P0-P5 verification report with pass/fail evidence per check.

Git history investigation. Use when: tracking code changes, finding where bugs were introduced, root cause analysis. Not for: code exploration (use dhpk-codebase-exploration), issue analysis (use dhpk-issue-analyze). Output: history trace + root cause report.

將雜亂的 git 變更依邏輯分群、逐批提交,適用於需要把多項變更拆成多個邏輯一致 commit 的情境。Not for: 單一小變更或只需要一個 commit（直接 commit 即可）。Output: 可直接執行的分組提交計畫與 commit 指令。

Run GitNexus CLI commands (analyze, status, clean, wiki, list). Use when: indexing or re-indexing a repo, checking index freshness, generating a wiki, listing indexed repos. Not for: querying the graph (use dhpk-gitnexus-exploring), tool/schema reference (use dhpk-gitnexus-guide). Output: the npx gitnexus command to run + next step.

Trace bugs and errors through the GitNexus knowledge graph. Use when: debugging a failure, tracing an error to its origin, finding callers of a suspect symbol, investigating unexpected behavior. Not for: general architecture understanding (use dhpk-gitnexus-exploring), change blast-radius (use dhpk-gitnexus-impact-analysis), CLI/index ops (use dhpk-gitnexus-cli). Output: a root-cause hypothesis with the call chain + files to inspect.

Explore and understand a codebase via the GitNexus knowledge graph. Use when: asking how code works, understanding architecture, tracing execution flows, or exploring unfamiliar code. Not for: blast-radius of a change (use dhpk-gitnexus-impact-analysis), bug tracing (use dhpk-gitnexus-debugging), CLI/index ops (use dhpk-gitnexus-cli). Output: a code-flow explanation grounded in graph queries + the files to read.

Reference for GitNexus MCP tools, resources, and graph schema. Use when: asking what GitNexus tools exist, how to query the graph, or for MCP resource / schema reference. Not for: CLI commands (use dhpk-gitnexus-cli), task flows (use dhpk-gitnexus-exploring / dhpk-gitnexus-debugging / dhpk-gitnexus-impact-analysis / dhpk-gitnexus-refactoring). Output: the right tool / resource / skill to use next.

Assess the blast radius of a code change via the GitNexus graph. Use when: asking what breaks if you change X, who depends on a symbol, or for a pre-commit safety check. Not for: understanding how code works (use dhpk-gitnexus-exploring), bug tracing (use dhpk-gitnexus-debugging), performing the rename/refactor (use dhpk-gitnexus-refactoring). Output: a depth-ranked dependents list + risk level.

Safely rename, extract, split, move, or restructure code using the GitNexus graph + coordinated rename. Use when: renaming a symbol across files, extracting a module, splitting a function/service, moving code. Not for: understanding code (use dhpk-gitnexus-exploring), pre-change risk only (use dhpk-gitnexus-impact-analysis), bug tracing (use dhpk-gitnexus-debugging). Output: an ordered refactor plan + verified multi-file edits.

Audits Claude Code context-window token consumption across agents, skills, MCP, rules, and CLAUDE.md; produces prioritized token-savings fixes. Use when: context feels bloated, after adding components, or checking headroom. Not for: trimming harness structure (use dhpk-harness-revise), backfilling .claude/ (use dhpk-harness-fill), or checking hook/plugin health (use dhpk-claude-health). Output: budget report + ranked fixes.

Explore-driven meta-workflow SSOT: 5-phase parallel inventory → fill in .claude/ infrastructure. Backfills .claude/skills/agents/rules + per-layer CLAUDE.md in one shot (≤3 Explore agents in parallel per round) when explicitly invoked via /dhpk:dhpk-harness-fill. Not for: single-file patches (use Edit), specific symbol lookups (use cx definition / gitnexus_impact), PR review (route through the sentinel chain to the matching reviewer agent). Output: list of created / modified files + conventio...

Trims and validates the project harness (.claude/ or .codex/) for a harness audit request, after major rule additions, or as periodic maintenance. Detects the active environment and ensures consistency, hygiene, and trigger preservation. Not for: business code review. Output: inventory snapshot, gap list with severity, proposed fixes, validation results.

GitHub Issue and PR review thread deep analysis with Codex blind verdict. Use when: analyzing issue root cause, classifying problems, investigation planning, triaging PR review comments for actionability. Not for: fixing bugs (use dhpk-adaptive-dev-workflow in bug mode), code exploration (use dhpk-codebase-exploration). Output: classified analysis + verdict assessment + investigation strategy.

Cross-version Laravel package authoring. Use when: designing a service provider, facade, version-conditional binding, publishable artifact, discovery contract, or package compatibility surface. Not for: application code, one-major API details, Composer release hygiene, or deep Testbench matrix mechanics. Output: an authoring decision or package design with provider, facade, publishing, discovery, and compatibility gates.

Orchestra Testbench matrix mechanics for Laravel packages spanning multiple majors. Use when: choosing Testbench pins, defining PHP × Laravel cells, building the shared TestCase, provisioning per-cell databases, running prefer-lowest, or validating package discovery. Not for: single-major packages, application test suites, Laravel package authoring outside the matrix, or semver decisions. Output: a per-cell CI matrix, TestCase design, and evidence-backed test gate.