hxsk

Plugin hxsk from marketplace hexoskeleton in SukbeomH/HExoskeleton. Commands read from the repo's manifest — 1 month ago.

Skills (24)

• arch-review

  Reviews architecture for circular imports, layer violations, boundary breaches and design-doc contradictions, recalling past architecture decisions from memory. Use before merging structural changes or when asked for an architecture or design review (아키텍처 검토, 레이어 위반, 순환 의존성, 설계 모순).

• clean

  Lints and formats shell scripts with shellcheck and shfmt and fixes the findings. Use before running or committing shell scripts, or when asked to lint, format or clean up scripts (린트, 포맷팅, 스크립트 정리).

• cleanse-memory

  Finds and, after human confirmation, deletes .hxsk/memories entries that contradict a Ground Truth source, within an explicitly scoped cleanup. Use only when the user explicitly asks for it (/hxsk:cleanse-memory <ground-truth> <scope-tags>, optionally --dry-run).

• codebase-mapper

  Analyzes an existing codebase and writes ARCHITECTURE.md and STACK.md covering structure, dependencies, patterns, integrations and technical debt. Use when onboarding to a codebase, before planning or refactoring, or when asked to map the structure or tech stack (코드베이스 분석, 구조 파악, 기술 부채).

• commit

  Creates conventional commits from staged changes: runs qlty/shellcheck pre-commit checks, proposes logical splits and writes phase-scoped messages. Use when committing changes or asked to write or split a commit (커밋 생성, 커밋 메시지, 커밋 분할).

• context-health-monitor

  Detects context rot (3 failed attempts, a repeating approach, context usage above 60%) and dumps session state to .hxsk/STATE.md for recovery or a fresh session. Use when debugging keeps failing, work goes in circles, or the context window is filling up (컨텍스트 상태, 세션 덤프, 3-strike).

• create-pr

  Pushes a branch and opens a pull request with gh after A-E quality checks, a lessons-learned review and a size check (under 500 lines of production code). Use when local changes are ready to ship for review (PR 생성, 풀 리퀘스트, PR 올리기).

• debugger

  Systematic root-cause debugging with hypothesis testing, one-variable experiments and debug-memory recall; stops and records state after 3 failed fix attempts. Use for bugs, errors or unexpected behavior (버그 디버깅, 오류 원인, 에러 추적).

• define-term

  Registers and merges project glossary terms stored as term-definition memories under .hxsk/memories/term-definition/, with human confirmation on conflicts. Use when a domain term needs a canonical definition or the user invokes /hxsk:define-term (용어 정의, 용어 등록).

• dispatcher

  Splits a PLAN.md or SPEC.md into a master issue and wave-ordered work issues under .hxsk/issues/, dispatches them to subagents in isolated git worktrees and merges the results. Use when planned tasks should run in parallel waves (병렬 실행, wave 실행, 이슈 분할).

• doc-lint

  Finds broken relative links in the project's markdown files with a bundled script, then checks document claims (paths, commands, counts) against the actual project. Use after modifying markdown, before a PR, or when links or structure look inconsistent (문서 정합성, 깨진 링크).

• empirical-validation

  Requires executed evidence (command output, exit codes, test results, API responses, screenshots) before any claim that work is complete, fixed or passing. Use before declaring done or whenever tempted to say it should work (경험적 검증, 증거 기반 확인).

• executor

  Executes PLAN.md tasks one at a time with atomic commits, deviation rules, checkpoints and an execution summary. Use when a PLAN.md exists and implementation should start (플랜 실행, 계획 실행, execute plan).

• handoff

  Wraps up a session: commits the work, runs the tests and stores a handoff memory and updates .hxsk/SESSION_HANDOFF.md so the next session or agent can resume. Use when a session ends, work pauses, or another agent will continue (세션 종료, 핸드오프, 인수인계).

• hxsk-init

  Initializes HExoskeleton (HXSK) in the current project: creates the per-project .hxsk/ working state (SPEC, STATE, CURRENT, SESSION_HANDOFF, VERIFICATION, PATTERNS, memories/) from the plugin templates without overwriting existing files, and adds an HXSK pointer block to AGENTS.md. Use when setting

• impact-analysis

  Traces the dependents of a file before it is modified and computes an impact score that decides whether approval is needed. Use before modifying any existing file, especially shared code (영향 분석, 의존성 분석).

• memory-protocol

  Stores and recalls project knowledge as typed markdown memories under .hxsk/memories/ (architecture decisions, root causes, patterns, deviations, session summaries) with the bundled md-store/md-recall scripts. Use at session start, after an architecture decision or root-cause finding, at session end

• plan-checker

  Validates PLAN.md before execution across six dimensions: requirement coverage, task completeness, dependencies, key links, scope sanity and verification derivation. Use after planning and before execution (플랜 검증, 계획 점검).

• planner

  Turns SPEC.md phase goals into a PLAN.md of 2-3 goal-backward tasks, each with exact file paths, actions, verify commands and done criteria, grouped into waves. Use when a SPEC.md exists and work needs planning (플랜 작성, 계획 수립, 태스크 분해).

• pr-review

  Reviews a pull request through six personas (developer, QA, security, architecture, DevOps, UX), classifies blockers by severity and issues a merge verdict. Use when given a PR number or URL to review (코드 리뷰, PR 리뷰).

• refactor

  Removes code smells (long methods, duplication, magic numbers, nested conditionals, dead code) in small behavior-preserving steps guarded by tests. Use when smells are identified and tests or specs exist (리팩토링, 코드 정리).

• skill-testing

  Tests a skill TDD-style (RED/GREEN/REFACTOR) by running pressure scenarios without and with it and comparing rule violations, to prove it changes agent behavior. Use when creating or hardening a skill (스킬 테스트, 스킬 검증).

• verifier

  Verifies implemented work against the spec at three levels (exists, substantive, wired), scans for stubs, placeholders and anti-patterns, and appends a dated section to .hxsk/VERIFICATION.md. Use when code exists and phase completion must be confirmed (구현 검증, 완료 확인).

• write-report

  Writes a conclusion-first decision report comparing 3-5 solutions with weighted scoring, TCO and status-quo risk. Use for technology selection, vendor comparison or buy-vs-build decisions (솔루션 비교 보고서, 기술 선정).

Subagents (2)

• spec-reviewer

  Validates implementation against SPEC.md requirements — checks what was built matches what was requested.

• verifier

  Validates implemented work against spec requirements with empirical evidence. Use after execution to confirm phase goals are met.

Hooks (13)

• SessionStart

  "${CLAUDE_PLUGIN_ROOT}/hooks/session-start.sh"

• PreToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/file-protect.py"

• PreToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/read-before-edit.py"

• PreToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/write-guard.py"

• PreToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/bash-guard.py"

• PostToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/track-read-history.py"

• PostToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/auto-format.sh"

• PostToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/track-modifications.sh"

• PostToolUse

  "${CLAUDE_PLUGIN_ROOT}/hooks/track-modifications.sh"

• PreCompact

  "${CLAUDE_PLUGIN_ROOT}/hooks/pre-compact-save.sh"

• Stop

  "${CLAUDE_PLUGIN_ROOT}/hooks/post-turn-verify.sh"

• Stop

  "${CLAUDE_PLUGIN_ROOT}/hooks/stop-context-save.sh"

• Stop

  "${CLAUDE_PLUGIN_ROOT}/hooks/collect-rationalization.sh"