Every skillMọi skill
Every hs:* skill is prose guidance — a structured prompt that tells Claude Code what to do. Invoke any with /hs:<name>. This is the exhaustive list, generated straight from each skill's SKILL.md; the curated catalog groups the most-used ones with how to pick.Mỗi skill hs:* là hướng dẫn prose — một prompt có cấu trúc cho Claude Code biết phải làm gì. Gọi bằng /hs:<tên>. Đây là danh sách đầy đủ, sinh trực tiếp từ SKILL.md của mỗi skill; catalog tuyển chọn nhóm các skill hay dùng kèm cách chọn.
UtilitiesTiện ích (2)
hs:eli5
Explain something so the reader actually gets it — the last answer restated plainly, or a topic, error, or snippet pitched at a named reader.
hs:handoff
Create a concise, redacted conversation handoff for a fresh agent session. Use when switching context, ending a work session, or preserving decisions and blockers.
OtherKhác (64)
hs:advise
Pressure-test an idea, plan, or design one question at a time, then give an honest verdict — what to do, what to avoid, better alternatives, trade-offs. Use for a second opinion or a sanity check before planning.
hs:bakeoff
Empirical bake-off — run 2-4 candidate probes on one mechanical metric, pick the winner by numbers or hand to a human when inside the noise band. Use when reasoning can't separate measurable options.
hs:bootstrap
Bootstrap a new project from scratch -- clarify requirements, init git, create doc structure, delegate to hs:plan + hs:cook. Use when starting a new repo or resetting a project's SDLC foundation.
hs:brainstorm
Brainstorm solutions with honest trade-off analysis — ideation, architecture decisions, technical debate, feasibility exploration. Use before committing to an approach.
hs:cleanup
Remove files an over-install left behind, safely: auto-clear pristine version-dropped files, and decide modified ones interactively (keep or change). Use after upgrading the harness or when verify reports orphans.
hs:code-review
Review code with technical rigor — bugs, regressions, security. Supports pending changes, PR number, commit hash, codebase scan. Verdict in review-decision.json; failing review blocks pr/ship/deploy. Use when code needs a rigorous pre-merge review.
hs:codemap
Set up, refresh and check this repo's code map (codemap.yaml + snapshots) that backs the code facts plan, cook, fix and review read. Use when hs-run offers a code map, or to see or repair it.
hs:coding-agent-orchestration
Coordinate several coding agents across one workflow — who plans, implements, reviews, tests; how work splits and hands off. Use when a task needs more than one agent, or the user asks which should own a step.
hs:compound
Compound the harness's own self-knowledge — telemetry, formalization candidates, backlog, a completeness critic — into ONE ranked self-improvement proposal carried across runs. Advisory only; the forward-looking twin of hs:insights.
hs:context-engineering
Manage the context budget — check limits, optimize tokens, coordinate subagents, debug context failures. Use when context is nearly full, designing multi-agent, or deciding what to load or drop.
hs:contract-test
Validate a live contract by running the real probe — call the API/CLI, check status, exit code and payload — then anchor the result as manual evidence. Use after building an endpoint or command, before ship.
hs:cook
Execute an approved plan phase by phase — TDD red→green, generate verification/review-decision artifacts, trace every step. Use when a human-approved plan is ready to implement.
hs:critique
Multi-lens adversarial critique — fan independent lenses at an artifact, consolidate into one ranked verdict. Advisory by default; gate mode writes a machine verdict; optional --loop refute cycle.
hs:debug
Systematic debugging with root-cause analysis before fixing. Use when something fails or behaves unexpectedly and the cause is not yet known.
hs:deploy
Deploy a project to any platform, target auto-detected from config files and docs/deployment.md. Use when the user wants a project released to a hosting platform.
hs:diagram
Compile typed JSON IR (architecture, dataflow, lifecycle, sequence, workflow) into a validated interactive system map -- SVG, fragment, or self-contained HTML. Use for a topology or state-machine diagram whose data must be checked before it renders.
hs:discover
Shape an ambiguous problem into a discovery brief for hs:plan — research + brainstorm chain -> direction summary, trade-offs, open questions.
hs:docs
Analyze codebase and manage project documentation — init, update, summarize. Use when docs need to be created, refreshed, or audited.
hs:docs-build
Render docs (md + frontmatter) + _index + _present into public/ (showcase HTML, theme and effects kept) and dist/ (flat-md, pdf, excel). Pure rendering — content lives in the md files; a legacy merged showcase.yaml still works.
hs:docs-scaffold
Generate a doc skeleton from a template (frontmatter + headings + > TBD) for the capability-driven required set. No-clobber. Never invents content.
hs:docs-standardize
Validate structural docs (frontmatter + graph invariants) → artifact JSON → gate. Finds gaps against the templates. Does NOT edit content and does NOT judge quality.
hs:drawio
Generate polished, editable .drawio diagrams — branded shape libraries (AWS/Azure/GCP/Cisco/K8s/UML/ER), swimlanes, strict geometry, PNG/SVG/PDF export. Opens in draw.io desktop and diagrams.net.
hs:eval-bootstrap
Bootstrap a full evals/ framework for THIS repo — deterministic scorer, ground truth, optional LLM judge, CI, API-key-free. Two human gates.
hs:find-skills
Locate and route to the correct hs:* skill — analyze intent, query the hs plugin registry, return the exact invoke command. Use when unsure which skill fits or to browse the full hs:* catalog.
hs:fix
Fix bugs, test failures, and CI/CD failures with an evidence-based workflow. Use when there is a concrete bug, a clear error, or a red test.
hs:gemini
Delegate an advisory or coding job to the gemini partner lane — research, review, adversarial review, or a worktree-staged code task. Use for a second-engine pass whose output stays provenance-stamped and separate.
hs:git
Git operations with conventional commits. Use for commit, push, PR, merge. Auto-split commits by type/scope. Scan for secrets before committing.
hs:goal
Prepare a built-in /goal run at authoring time — interview the objective into a NEW self-contained goal.md, arm the autonomy bell, scaffold the cycle dir — then hand the loop over. Use for an unattended objective-until-met run.
hs:goal-warmup
Outcome-locked preflight before a long autonomous run: interview to an approved Outcome Contract, plan, review, whole-plan preflight matrix, then Ready/Blocked handoff. Never auto-starts the run.
hs:harness-creator
Create new harness primitives (hook, rule, schema, data, script, agent) — not skills. Use to add a hook/script/rule/schema/agent to the harness. To create an hs:* skill, use hs:skill-creator.
hs:insights
Read-only usage insights from harness telemetry — hot vs never-used skills, workflow chains, gate-block patterns — with proposed optimizations. Advisory only; never mutates. Pairs with hs:setup to act on one.
hs:loop
In-session self-optimization loop — N iterations against a measurable metric, learns from git history, auto-keep/discard changes. Use for mechanical metrics (coverage, bundle size). In-session only.
hs:mermaidjs
Text-based Mermaid.js v11 diagrams INLINE in markdown -- flowchart, sequence, class, ER, state, gantt, architecture -- renders natively on GitHub/GitLab without a separate file.
hs:partner
Delegate an advisory or coding job to the ccs partner lane — a second full-Claude session under a named provider, provenance-stamped with its own model and cost. Use for a second-brain pass before merging.
hs:plan
Create a verified implementation plan — research, constraint-scan, phase design, red-team, and validate before cook. Use when a real feature or refactor needs a verified plan before any code is written.
hs:port
Extract, compare, port, or adapt a feature from another repo into this project. Analysis + plan only — hands off to hs:cook.
hs:predict
5 expert personas independently debate a proposed change before implementation, catching architectural, security, performance, and UX risks early. Use before large features or high-risk changes.
hs:preview
Explain a change or architecture with a diagram when visuals are clearer than prose — flow/architecture, before-after, sequence; output to docs/ or plans/.
hs:problem-solving
Structured unblocking — identify the block type, choose the right technique, reframe before resuming implementation. Use when stuck in a loop, complexity escalates, or a hypothesis fails 3+ times.
hs:prompt
Write one optimized, ready-to-paste prompt from a rough idea, or fix, adapt, or split an existing one. Harness-aware for internal targets. Not for general chat, coding, or doc writing.
hs:release
Cut a harness release: cut version, push tag, pack dist/ for verification (release.yml publishes it). Owner-only; pauses before every outward action. Use when cutting a new harness version.
hs:remember
Capture what this session actually learned — decisions, non-obvious facts, correcting feedback — into the DEC ledger, auto-memory, or BACKLOG. You approve every entry before it is written.
hs:research
Verified technical research — pose a question, gather multiple sources, verify evidence, synthesize a report. Use before implementing or when evaluating technology or architecture.
hs:review-pr
Review a GitHub PR or GitLab MR (forge auto-detected) — diff, CI, correctness, security, breaking changes. Supports --fix and --reply. For small local diffs use hs:code-review instead.
hs:rops
Run a heavy build, test, benchmark, or container on a remote machine over ssh + rsync, from ANY repo — destination, excludes and aliases detected from the stack. Use when a run needs a faster or quieter host.
hs:rule-author
Author a per-repo review-rule override under docs/standards/ — detects conflicts with the shipped standards and records a reason for each. Use to tune, disable, or add a review rule without editing the shipped tree.
hs:scenario
Decompose a feature/code path across 15 dimensions to generate edge cases, risks, and test targets before implementation. Use for QA planning, risk discovery, and coverage audits.
hs:scout
Fast codebase exploration using parallel agents — find files, locate code, gather context before implementing or debugging. Output goes to plans/reports/.
hs:security-scan
Scan codebase for security issues — hardcoded secrets, dependency CVEs, injection/authz gaps, STRIDE+OWASP. Use before releases, after adding auth/payment/data, or for periodic audits.
hs:setup
Configure this project's harness — guard/stage policy, output language, code style, hook components — through the validated config CLIs. Use when onboarding a fresh install or changing a config decision.
hs:shape
Bridge an approved PO story spec into dev-task decomposition, roadmap and effort rollup, market-experiment specs, and the POC-gate loop. The BA half of the product pair; never mutates a PO story.
hs:ship
Gated ship pipeline: review PASS → verification PASS → human approval → push/pr. Highest-risk stage — all gates must be real. Use when a completed branch needs to become a PR.
hs:skill-creator
Create or update hs:* skills for the harness — SKILL.md, frontmatter, thin-core, references, validate via catalog.py. Use to create a new skill, refine triggers, or extend the harness.
hs:spec
Interview-driven product spec hierarchy (Vision/BRD/PRD/Epic/Story) with traceability, validation, and visualization. Use when a non-technical PO needs to capture or structure requirements without writing code.
hs:team
Orchestrate parallel Agent Teams — research, cook, review, debug with multiple independent teammates.
hs:tech-graph
Generate publish-grade static technical diagrams (SVG+PNG, 8 styles) — architecture, data flow, sequence, agent/memory, concept map — for embedding in docs or slides.
hs:techstack
Detect the repo's tech stack — languages, test command, package manager, CI — so the harness adapts instead of assuming pytest. Use before hs:test or hs:setup in a non-Python repo. Read-only.
hs:test
Run and validate tests for the current change — unit/integration profiles, concise QA report, 100% pass gate. Use when a change needs its tests run and a gate-ready verdict.
hs:threejs
Build 3D web experiences with Three.js. Use for WebGL/WebGPU scenes, GLTF models, animations, physics, VR/XR.
hs:triage
Orchestrate the defect lifecycle — reproduce, classify, and gate bugs via hs:scout→hs:debug→hs:fix→hs:test. Use when a bug, test failure, or unexpected behavior is reported.
hs:understand
Orchestrate codebase comprehension before touching code — chain hs:scout and hs:context-engineering to build a codebase map. Use before hs:plan on unfamiliar areas.
hs:web-testing
Web testing with Playwright, Vitest, k6. Use for browser-based test automation, flaky-test diagnosis, and web performance or accessibility checks.
hs:workflow-orchestrate
Design a spawn strategy for a delegated task — pick subagents vs Workflow vs Agent Teams, size and group the fan-out, batch-consolidate, early-write findings — then present it for approval before spawning. The delegation planner other skills call.
hs:worktree
Create, inspect, and clean up isolated git worktrees. Use for feature isolation, worktree health audits, and stale metadata cleanup.
workflow (2)
hs:dispatch
Decide who owns a piece of work before any of it starts — answer it directly, route it to one skill, or chain several — and set the quality gate its risk earns. Use when a request is multi-step, spans domains, or has no obvious owner.
hs:finetune
Route work that must improve a MEASURED result to the skill that owns that outcome, and hand over the bounded-iteration contract. Use when the goal is to make a number better by iterating rather than by one attempt.