원클릭으로 Manus에서 모든 스킬 실행

skill-studio-write

스타2

포크0

업데이트2026년 5월 26일 22:40

Write, adapt, and validate Cursor skills and packs in felipeblassioli/agent-skills. Covers greenfield skill authoring through Socratic discovery, distilling reference material into a skill, scaffolding a pack under `packs/<name>/`, evaluating an external skill candidate before import, decomposing a Claude-style plugin into Cursor-native artifacts, and running the eval and comparison loop that proves an authoring change improved the artifact. Use when the user wants to create or refactor a skill or pack, needs help with `SKILL.md`, `metadata.json`, `pack.json`, or pack-bundled skills, or wants to decide whether content should live in a skill, pack, rule, hook, subagent, or docs. Invoke explicitly via `/skill-studio-write`. Do not use for portfolio overlap audits (use `skill-studio-audit`), governance or registry version bumps (use `skill-studio-maintain`), or skill install and sync (use the repo scripts directly).

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

felipeblassioli

felipeblassioli/agent-skills

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

관련 직업SOC

SOC 직업 분류 기준

소프트웨어 개발자컴퓨터 및 수학직·SOC 15-1252

파일 탐색기

42 개 파일

SKILL.md

readonly

이 저장소의 다른 Skills

같은 저장소

promote-check

felipeblassioli/agent-skills

Promotion preflight gate for the agent-skills marketplace. Run before promoting a skill from a sandbox plugin to an official one, or before opening a promotion PR, to get a go / no-go verdict. Audit-first: it chains the deterministic skill-audit signals, marketplace/registry alignment, and version + CHANGELOG presence, and BLOCKS (no-go) on any unresolved finding, citing the failing gate. User-invoked only. For fixing what it surfaces use skill-studio:skill-enhance; for the release wiring use repo-governance:skill-maintainer.

2026-07-062

smart-prompt

felipeblassioli/agent-skills

Shapes a loose intent into a well-formed prompt for a Claude Code session that can gather context, use tools, and ask questions. Matches the intent to a validated archetype, fills its slots (objective, context to gather, anchors, clarify-first questions, acceptance criteria, verification, output contract), routes the tier via model-recommender, and self-checks the result with prompt-audit. Use when you have a rough idea of what to ask an agent and want it reshaped to hit a specific goal, when you type /smart-prompt with a loose instruction, or to turn "read X and study Y to figure out Z" into an executable agentic prompt.

2026-07-052

prompt-audit

felipeblassioli/agent-skills

Adversarially reviews a draft prompt against a versioned rule set to find what will make it underperform — archetype mismatch, wrong prescription level for the target model, the Fable reasoning-echo trap, negatives that should be positives, unstated expectations, rejected params, and more — then returns tagged findings (rule_id + severity) plus a rewritten, --fix-style prompt. Use when asked to review / audit / critique / improve a prompt, check whether a prompt is good, tighten a system prompt, or before shipping a prompt to a model.

2026-07-052

model-recommender

felipeblassioli/agent-skills

Recommends which Claude model and effort level to run for a task. Classifies the task into an archetype → tier + effort using the rubric in ~/.claude/model-profiles.md, then resolves today's model string from that file's tier_to_model. Use when deciding which model to use, routing a task or phase to a model, picking an effort level, choosing between Opus/Sonnet/Haiku/Fable, or when a task spans plan/implement/review phases that each want a different model.

2026-07-052

skill-audit

felipeblassioli/agent-skills

Audit a Claude skill for quality, context-efficiency, and best-practice compliance — classify its archetype, score the four principles and instruction craft, and return a report — or review a PORTFOLIO of skills for overlap and trigger collisions. Use when someone asks to audit, review, grade, "lint", or sanity-check a skill or plugin's quality; asks "does this skill follow best practices / is it well-scoped / is the description good enough to trigger / do these skills overlap or collide"; or wants a governance sweep across a set of skills. Read-only: it diagnoses and recommends, it does not edit. Do not use to create or scaffold a skill (use skill-create), to improve a skill or run eval/benchmark loops (use skill-enhance), or for versioning/release/CHANGELOG governance (use repo-governance:skill-maintainer).

2026-07-032

gh-post-code-review

felipeblassioli/agent-skills

Post a structured code review to a GitHub PR via `gh` CLI (REST API fallback for multi-comment / inline-anchored reviews). Maps BLOCKER/HIGH/MEDIUM/LOW/QUESTION sections to the right review event and inline anchors, parses ` ```start:end:path ` code refs, and adapts tone to the PR author. Use when you have a written code review and want it posted to a specific PR URL or number. Do NOT use to author the review itself, triage existing comments, or open/merge PRs.

2026-07-032

name	skill-studio-write
description	Write, adapt, and validate Cursor skills and packs in felipeblassioli/agent-skills. Covers greenfield skill authoring through Socratic discovery, distilling reference material into a skill, scaffolding a pack under `packs/<name>/`, evaluating an external skill candidate before import, decomposing a Claude-style plugin into Cursor-native artifacts, and running the eval and comparison loop that proves an authoring change improved the artifact. Use when the user wants to create or refactor a skill or pack, needs help with `SKILL.md`, `metadata.json`, `pack.json`, or pack-bundled skills, or wants to decide whether content should live in a skill, pack, rule, hook, subagent, or docs. Invoke explicitly via `/skill-studio-write`. Do not use for portfolio overlap audits (use `skill-studio-audit`), governance or registry version bumps (use `skill-studio-maintain`), or skill install and sync (use the repo scripts directly).
disable-model-invocation	true

Skill Studio — Write

Authoring surface for Cursor skills and packs in this repository. One skill, six branches, shared principles and subagents. Pause for confirmation at each phase and prefer the smallest correct shape.

Applicability Gate

Apply this skill when ANY of the following are true:

The user wants to create or refactor a Cursor skill.
The user wants to create a Cursor pack under packs/<name>/.
The user provides reference material (docs, code, URLs) to package as a skill or pack.
The user wants a go/no-go review of an external skill candidate before importing it into skills/.
The user wants to decompose a Claude-style plugin or mixed plugin folder into Cursor-native artifacts.
The user wants evidence (an eval loop, blind comparison, or source audit) that an authoring change actually improved the artifact.

Do NOT apply when:

The user wants a portfolio overlap or compliance audit across many skills → use skill-studio-audit.
The user wants registry versioning, release artifacts, or governance work → use skill-studio-maintain.
The user only wants to install, sync, list, or version existing skills → use the repo scripts (scripts/skill-sync.sh, scripts/skill-version.sh, scripts/skill-import.sh).
The task is generic code implementation unrelated to skill or pack authoring.

Intent Router

Pick exactly one branch. Each branch lists the primary references and supporting templates/scripts to load on demand.

Signal	Branch	Primary references	Supporting assets
New skill, no reference dump; user wants Socratic intake	A. Greenfield skill	`references/greenfield-discovery.md`, `references/surface-selection.md`, `references/skill-archetypes.md`, `references/cursor-skill-standard.md`, `references/skill-quality-checklist.md`	`assets/templates/skill-contract.md`, `assets/templates/skill-archetypes/*`, `scripts/validate-skill.sh`
User provides reference material (`@docs`, code, URLs) for a skill	B. Skill from refs	`references/material-intake.md`, `references/skill-archetypes.md`, `references/cursor-skill-standard.md`, `references/skill-quality-checklist.md`	`assets/templates/skill-archetypes/*`, `scripts/validate-skill.sh`
User wants an installable pack under `packs/<name>/`	C. Pack from refs	`references/material-intake.md`, `references/pack-standard.md`, `references/pack-archetypes.md`, `references/pack-quality-checklist.md`	`assets/templates/pack/*`, `assets/templates/bundled-skill-artifact.fragment.json`, `scripts/validate-pack.sh`
Candidate skill in another repo or arbitrary path	D. External skill intake	`references/candidate-review.md`, `references/import-paths.md`	`scripts/inspect-candidate-skill.sh`, `assets/templates/skill-intake-report.md`
`.claude-plugin`, `.mcp.json`, mixed plugin tree	E. Claude-plugin adaptation	`references/source-decomposition.md`, `references/pack-standard.md`, `references/cursor-skill-standard.md`	`assets/templates/adaptation-report.md`, `assets/templates/bundled-skill-artifact.fragment.json`
User wants proof an authoring change improved the artifact	F. Eval / comparison loop	`references/eval-loop.md`	`scripts/bootstrap_skill_comparison.py`, `scripts/aggregate_benchmark.py`, `eval-viewer/generate_review.py`

Surface-only question ("should this be a skill, rule, hook, subagent, or pack?") → load references/surface-selection.md first; route afterwards.

Shared Principles

These apply to every branch.

Context architecture first. Treat SKILL.md (or pack.json) as the hot path. Push detail into references/, assets/, and scripts/.
One-Hop Rule. Every reference must be linked directly from SKILL.md. No multi-hop chains.
Smallest correct shape. Prefer a single repo-root skill over a pack unless runtime assets are required. Prefer pack-bundled skills over copying guidance into rules.
Cheap-agent-first. Delegate read-only exploration, classification, and blind grading to subagents with model: fast / readonly: true. Reserve the main thread for design decisions.
Strict output shapes. Prefer JSON or constrained tables over chatty prose, especially for intake summaries and pack contracts.
The agent is already smart. Skip generic explanations. Only include domain-specific knowledge the agent does not have.
Confirmation before files. Do not write files until the current phase has been approved.

Branch Procedures

Each branch is a thin handoff to its references. Follow the numbered steps; load reference files only when needed.

A. Greenfield skill

Confirm surface selection (references/surface-selection.md).
Run Socratic discovery (references/greenfield-discovery.md); fill assets/templates/skill-contract.md. Pause for contract approval.
Choose the archetype (references/skill-archetypes.md).
Scaffold only justified files; keep SKILL.md ≤ 500 lines.
Run scripts/validate-skill.sh <skill-path> and walk through references/skill-quality-checklist.md. Pause for final sign-off.

B. Skill from reference material

Classify every source with references/material-intake.md. Pause for classification approval.
Choose archetype (references/skill-archetypes.md) and scaffold from assets/templates/skill-archetypes/.
Distill — high-signal content only; omit what the agent already knows.
Validate with scripts/validate-skill.sh and the quality checklist.

C. Pack from reference material

Classify sources with references/material-intake.md. Pause for classification approval.
Pick a pack archetype (references/pack-archetypes.md). Pause for scope approval.
Define the pack contract (artifact matrix, targets, profiles, install policy). Use references/pack-standard.md. Pause for scaffold approval.
Scaffold from assets/templates/pack/. For bundled skills, use assets/templates/bundled-skill-artifact.fragment.json and a pack-scoped skillId.
Add the entry to cursor-pack-registry.json (repo-only operation).
Run scripts/validate-pack.sh <pack-name> (wraps scripts/cursor-pack-verify.sh). Walk through references/pack-quality-checklist.md. Pause for final sign-off.

D. External skill intake

Run scripts/inspect-candidate-skill.sh <candidate-path> [expected-name].
Read the JSON output. Classify ready, adapt, or reject.
Apply references/candidate-review.md to check repo fit.
Choose the import path from references/import-paths.md.
Present assets/templates/skill-intake-report.md. Pause for approval.
Only after approval, copy/normalize into skills/<name>/, add metadata.json, and register in skill-registry.json.

E. Claude-plugin adaptation

Inspect the candidate root cheaply before reading many files.
Classify each file with references/source-decomposition.md (pack-runtime, skill-guidance, docs-reference, claude-only, ignore).
Recommend the smallest correct destination shape (pack only, repo-root skill only, pack with bundled skills, or pack/skill plus docs).
Call out MCP trust/portability concerns explicitly; default to example config with ${env:VAR} placeholders, never live mcp.json.
Present assets/templates/adaptation-report.md. Pause for approval.
If the user proceeds with import, route to Branch C or D as appropriate.

F. Eval / comparison loop

Bootstrap a workspace with scripts/bootstrap_skill_comparison.py (see references/eval-loop.md for the layout and JSON schemas).
Run candidate + baseline against shared evals.json prompts.
Grade with skill-creator-grader; compare blindly with skill-creator-comparator; audit structure with skill-creator-structural-auditor.
Aggregate with scripts/aggregate_benchmark.py; render the review UI with eval-viewer/generate_review.py.
Produce the recommendation with skill-creator-analyzer.

Subagent Delegation

Subagents shipped with this pack (packs/cursor-skill-studio/.cursor/agents/):

Subagent	Use when
`skill-creator-bootstrapper`	Branch C/E: mixed source tree needs an artifact matrix before scaffolding.
`skill-creator-structural-auditor`	Branch F: source-level comparison of two skills; pre-import shape audit (Branch D).
`skill-creator-grader`	Branch F: grade eval outputs against expectations.
`skill-creator-comparator`	Branch F: blind A/B comparison of outputs.
`skill-creator-analyzer`	Branch F: synthesize grades + comparisons + structural findings into the recommendation.
`skill-overlap-clusterer` / `skill-architecture-checker` / `skill-consolidation-advisor`	Owned by `skill-studio-audit`. Hand off if the user shifts from authoring to portfolio review.

Prefer subagents over inline work whenever the task is bounded, read-heavy, and would otherwise bloat the main context.

Repo-Only Operations

These steps only work inside felipeblassioli/agent-skills:

Registering a new skill in skill-registry.json.
Registering a new pack in cursor-pack-registry.json.
Importing an external skill via scripts/skill-import.sh.
Verifying a pack via scripts/cursor-pack-verify.sh.

If the user is in another workspace, stop and report the missing dependency rather than improvising.

Gotchas

metadata.json MUST exist alongside SKILL.md. The bundled scripts/validate-skill.sh reports a missing metadata.json as a hard failure, not a warning.
For pack-bundled skills, the skillId in pack.json MUST equal the name in SKILL.md, or cursor-pack-verify.sh rejects the manifest.
Pack-bundled skills are NOT auto-added to skill-registry.json. Adding one is a promotion (skill-studio-maintain Branch D) and flips version authority from the pack to the root registry.
Cursor user installs (--target=user) skip .cursor/rules/. Anything that must persist for user installs cannot live in rules/ — move it to a subagent, bundled skill, or guide.
disable-model-invocation: true does NOT shrink the description — it still ships and still counts against the listing budget. Keep the description tight even on explicit-only router skills.
Pack templates ship with mcpPolicy: "none" by default. If the pack legitimately needs MCP examples, set "example-only" — never write a live mcp.json from the install path.

Output Contracts

Branch	Final artifact to present
A, B	Filled `skill-contract.md`, file tree, validator JSON, checklist results.
C	Pack contract table, file tree, `cursor-pack-registry.json` diff, `cursor-pack-verify.sh` output.
D	`skill-intake-report.md` filled with classification, blocking issues, suggested destination.
E	`adaptation-report.md` with destination matrix, MCP classification, blocking concerns, smallest viable migration plan.
F	Analyzer JSON (`analysis.json`) with recommendation, evidence, residual risks.

skill-studio-write

Skill Studio — Write

Applicability Gate

Intent Router

Shared Principles

Branch Procedures

A. Greenfield skill

B. Skill from reference material

C. Pack from reference material

D. External skill intake

E. Claude-plugin adaptation

F. Eval / comparison loop

Subagent Delegation

Repo-Only Operations

Gotchas

Output Contracts

See Also

Skill Studio — Write

Applicability Gate

Intent Router

Shared Principles

Branch Procedures

A. Greenfield skill

B. Skill from reference material

C. Pack from reference material

D. External skill intake

E. Claude-plugin adaptation

F. Eval / comparison loop

Subagent Delegation

Repo-Only Operations

Gotchas

Output Contracts

See Also