Agent Skills

A collection of agent skills for Claude Code, Codex and other AI coding agents.

Install

# All at once

npx skills add sentimony/skills -a codex claude-code -y

# Or each separately

npx skills add sentimony/skills -s web-debug -a codex claude-code -y
npx skills add sentimony/skills -s vitest -a codex claude-code -y
npx skills add sentimony/skills -s typescript -a codex claude-code -y
npx skills add sentimony/skills -s echarts -a codex claude-code -y
npx skills add sentimony/skills -s scope-triage -a codex claude-code -y
npx skills add sentimony/skills -s plan-crafting -a codex claude-code -y
npx skills add sentimony/skills -s dashfix -a codex claude-code -y
npx skills add sentimony/skills -s negafix -a codex claude-code -y
npx skills add sentimony/skills -s commit-all -a codex claude-code -y
npx skills add sentimony/skills -s maintaining-agent-context -a codex claude-code -y
npx skills add sentimony/skills -s frontend-crafting -a codex claude-code -y
npx skills add sentimony/skills -s tdd -a codex claude-code -y
npx skills add sentimony/skills -s debugging -a codex claude-code -y
npx skills add sentimony/skills -s review-request -a codex claude-code -y
npx skills add sentimony/skills -s review-resolution -a codex claude-code -y
npx skills add sentimony/skills -s verification-gate -a codex claude-code -y
npx skills add sentimony/skills -s git-worktree-isolation -a codex claude-code -y
npx skills add sentimony/skills -s parallel-agents -a codex claude-code -y
npx skills add sentimony/skills -s inline-plan-dev -a codex claude-code -y
npx skills add sentimony/skills -s subagent-plan-dev -a codex claude-code -y
npx skills add sentimony/skills -s branch-finish -a codex claude-code -y
npx skills add sentimony/skills -s skill-crafting -a codex claude-code -y

Skills

Skill	Skill Version	Release	Description
web-debug	1.3.3	v1.26.0	Debug and verify local web apps via Playwright.
vitest	1.3.1	v1.21.0	Configure, write, debug, run, migrate, and audit Vitest tests for JavaScript/TypeScript projects.
typescript	1.4.1	v1.20.0	Configure tsconfig, diagnose compiler behavior, and audit or migrate TypeScript projects.
echarts	1.2.1	v1.22.0	Build, audit, style, debug, and optimize Apache ECharts visualizations in vanilla JS, React, or Vue.
scope-triage	1.0.5	v1.34.0	Classify request scope before design work, then route to direct implementation, a light spec, or a full design cycle.
plan-crafting	1.3.3	v1.34.2	Turn an approved design or settled requirements into a bite-sized, TDD-oriented implementation plan.
dashfix	1.2.2	v1.12.1	Ban typographic dashes in English text, check their form where a language's orthography requires them, audit a project, and score it 0-100.
negafix	1.2.2	v1.12.1	Ban negative parallelism ("it's not just X, it's Y"), audit prose for it, and score it 0-100.
commit-all	1.1.1	v1.16.0	Gather the working tree into a single commit on the current branch, on explicit /commit-all invocation only.
maintaining-agent-context	1.3.1	v1.24.0	Audit, restructure, and maintain a repository's agent instruction architecture for Claude Code and Codex.
frontend-crafting	1.3.1	v1.23.0	Create, redesign, review, and polish user interfaces with subject-driven design decisions and a verifiable quality gate.
tdd	1.0.4	v1.34.2	Drive behavior changes through valid RED, sufficient GREEN, and evidence-backed refactoring.
debugging	1.0.9	v1.33.1	Investigate bugs and unexpected technical behavior with a root-cause-first evidence workflow.
review-request	1.0.3	v1.27.0	Prepare and dispatch independent code review against requirements, exact scope, and the actual diff.
review-resolution	1.0.3	v1.28.0	Validate and resolve code-review findings with evidence, explicit dispositions, and proportional re-review decisions.
verification-gate	1.0.3	v1.29.0	Turn a completion claim into an evidence-backed verdict against the current tree.
git-worktree-isolation	1.2.0	v1.30.0	Select, detect, or create a safe isolated development workspace with explicit ownership, baseline, and handoff.
parallel-agents	1.0.3	v1.31.0	Prove work units independent, isolate mutable state, and dispatch one bounded parallel wave with reconciled results.
inline-plan-dev	1.0.3	v1.32.0	Execute an existing implementation plan inline in the current session, with plan-reality reconciliation, proportional verification, and durable resume.
subagent-plan-dev	1.0.3	v1.32.0	Execute an existing implementation plan through scoped subagents with risk-based dispatch, independent verification, and controlled escalation.
branch-finish	1.0.4	v1.34.1	Decide, execute and report the integration outcome for verified work, cleaning up only what is provably safe to remove.
skill-crafting	1.1.0	v1.38.0	Package-owned portable eval contract, validation, fresh-sandbox runner, and aggregation scripts with runtime adapters and isolated artifacts.

Development Workflow

scope-triage is the development entry point. It classifies the request and picks the minimum appropriate process, so a one-line configuration change and a new subsystem do not receive the same ceremony.

                         scope-triage
                              │
               ┌──────────────┼──────────────┐
               │              │              │
            Route A        Route B        Route C
         direct work     light spec      full design
               │              │              │
               │         direct work     plan-crafting
               │              │              │
               │              │        ┌─────┴─────┐
               │              │        │           │
               │              │   inline-plan   subagent-plan
               │              │      -dev           -dev
               └──────────────┴────────┬───────────┘
                                      │
                               implementation
                                      │
                         behavior change → tdd
                                      │
                 review-request (when warranted)
                                      │
             review-resolution (when findings exist)
                                      │
                              verification-gate
                                      │
               branch-finish (when lifecycle exists)
                                      │
                                   complete

When work fails unexpectedly, the diagnostic path runs alongside the main flow and returns to whatever workflow discovered the failure.

any development stage
        │
        ▼
     debugging
        │
        ├── browser evidence needed
        │          │
        │      web-debug
        │          │
        └──────────┘
        │
     root cause
        │
        ├── material redesign needed → scope-triage
        │
        ├── behavior correction → tdd
        │
        ▼
       fix
        │
        ▼
return to calling workflow

Four properties of this workflow are worth stating explicitly.

• tdd is used within implementation, whenever behavior changes. It is a micro-cycle inside a task rather than a sequential phase that follows implementation. Both execution modes and direct implementation invoke it at the moment a behavior change is written.
• Review is proportional; verification is authoritative. review-request is invoked when the change warrants independent eyes, and review-resolution validates findings before anything is changed in response. A review verdict stands separate from completion: verification-gate owns the authoritative pass or fail against the current tree.
• The diagnostic path has three distinct owners. debugging owns causal investigation, web-debug supplies browser and runtime evidence when the symptom lives in a page, and tdd drives the regression behavior once the cause is known. The fix then returns to the workflow that discovered the failure.
• The workflow is composable rather than mandatory-linear. Every step after scope-triage is reached on a condition. Route A terminates in implementation with proportional verification, and branch-finish runs only when a branch or workspace lifecycle exists.

Conditional capabilities, invoked when the work calls for them rather than in sequence: skill-crafting, tdd, debugging, web-debug, git-worktree-isolation, parallel-agents, frontend-crafting, vitest, typescript, echarts.

Have fun ;)