diagram-to-image

A marketplace of agent skills by VerySmallWoods — content creation, publishing, analytics, and security scanning. Works with Claude Code, Codex, Cursor, Windsurf, and any other coding agent that supports the skills.sh format.

Install

Universal (Claude Code · Codex · Cursor · Windsurf · ...)

Uses the skills.sh installer by Matt Pocock:

npx skills@latest add sugarforever/01coder-agent-skills

Pick the skills you want and the agents to install them on. Done.

Claude Code (native marketplace)

/plugin marketplace add sugarforever/01coder-agent-skills
/plugin install 01coder-skills@01coder-agent-skills

New skills are picked up automatically on marketplace update — no reinstall needed.

Available Skills

Content & Writing

• •personal-chinese-writing-style — Personal writing-style preferences (punctuation, structure, voice). Reference when writing, translating, or editing content.
• •video-planner — Plan videos and prepare publishing materials: read-aloud scripts (口播稿) with per-section screen-share cues, plus title, description, tags, and YouTube chapter timestamps for YouTube and Bilibili.
• •cover-image — Hand-drawn style article cover image generator with 17 style options and 10 composition layouts.
• •tweet-insight — Read a tweet plus its linked papers / blogs / system cards, then write an original Chinese share-post in your own words.
• •share-reading — Draft social-media posts to recommend an article, paper, or resource across X, Substack, and 知识星球.
• •promote-post — Write a teaser tweet for a published article that opens the story instead of summarizing it — the tweet IS the first bite, not a label on the packaging.
• •slides-video — Orchestrate slide generation + script writing to produce slides-driven narration videos (口播视频), with each slide mapping 1:1 to one voiceover section.
• •producing-video — Turn a user-provided voiceover audio + SRT into a finished, narration-synced MP4 with HyperFrames (HTML-to-video). Audio + SRT are the source of truth: scenes are timed to the SRT cues and the audio is muxed in automatically.
• •subtitle-correction — Correct speech-recognition errors in .srt subtitle files (Chinese and English) while preserving timestamps.

Publishing

• •publish-x-article — Publish Markdown articles to the X (Twitter) Articles editor with proper formatting. Inspired by wshuyi/x-article-publisher-skill.
• •publish-substack-article — Publish Markdown articles to Substack as drafts, with Markdown-to-HTML conversion.
• •publish-zsxq-article — Publish Markdown articles to Zsxq (知识星球) as drafts.

Domain-Specific

• •fpl-copilot — Fantasy Premier League copilot. Syncs live FPL data into local SQLite, analyzes players/teams/fixtures, generates self-contained HTML reports for gameweek strategy, captain picks, fixture matrices, and transfer comparisons.

Security Scanning

• •nextjs-security-scan — Security vulnerability scanner for Next.js and TypeScript/JavaScript projects. OWASP Top 10, XSS, injection, secret scanning, dependency CVEs.
• •python-security-scan — Security vulnerability scanner for Python projects. Framework-aware checks for Flask, Django, and FastAPI.

Utilities

• •codex-cli — Delegate one-off tasks to OpenAI Codex CLI, including coding reviews, implementation passes, and image generation with proactive image path discovery.
• •claude-session-manager — List and export Claude Code JSONL session transcripts from ~/.claude/projects into organized, readable Markdown with linked tool-call details.
• •codex-session-manager — List and export Codex JSONL session transcripts from ~/.codex/sessions and optional archived sessions into organized, readable Markdown.
• •mining-session-skills — Review one completed Claude Code session (located by natural-language description, via its exported Markdown) and propose a skill to create, update, or reuse — so similar work goes faster next time. Builds on claude-session-manager.
• •diagram-to-image — Convert Mermaid diagrams and Markdown tables to PNG / SVG images for platforms that don't support rich formatting.
• •interactive-input — Embed interactive UI components (multiple choice, forms, surveys) in chat responses on compatible clients.

Integrations

• •add-feishu — Add Feishu (飞书 / Lark) as an agent channel via WebSocket long connection. No public URL required.

Skill Notes

personal-chinese-writing-style

Applies the author's Chinese writing style to writing, translation, editing, subtitles, tweets, newsletters, and social posts. It enforces Chinese punctuation preferences, natural Chinese tech prose, and structure carried by the writing rather than fixed templates.

SKILL.md loads reference files on demand:

• •references/punctuation.md
• •references/article-structure.md
• •references/voice-and-phrasing.md
• •references/social-media-style.md

codex-cli

The codex-cli skill is intended to let other agents, such as Claude Code or OpenClaw, deliberately consult OpenAI Codex CLI for scoped one-off work. It should behave like a bridge workflow, not a replacement for the host agent: first verify that codex is installed, delegate a clearly framed task through codex exec, then inspect and report the concrete result.

When improving this skill, keep it focused on practical Codex CLI operation:

• •Prefer non-interactive codex exec workflows over interactive sessions.
• •Shape delegated prompts with Codex best-practices sections: Goal, Context, Constraints, and Done when.
• •Use read-only for planning, review, critique, and image generation unless the user explicitly wants Codex to edit files.
• •For implementation tasks, encourage plan-first delegation before write-mode execution.
• •For image generation, always locate the generated file under $CODEX_HOME/generated_images/<session-id>/ or ~/.codex/generated_images/<session-id>/, verify dimensions when an aspect ratio was requested, and report the path proactively.
• •Do not claim a specific image backend model unless Codex output, logs, or metadata prove it.
• •Do not trust Codex's final message alone when file changes matter; inspect Git status, diffs, or output files.

Future additions should be small and evidence-oriented: better session-id extraction, safer output parsing, known CLI flag changes, and stronger image/file discovery are aligned with the skill. Broad Codex product documentation, general prompt engineering, or host-agent-specific behavior should stay out unless it directly improves this delegation workflow.

claude-session-manager

Manages Claude Code JSONL transcripts from ~/.claude/projects and exports Markdown under ~/.claude/session-markdown by default. It supports full export and specific-session export. In specific mode it first lists numbered candidates with modified time, project key, short session id, cwd, and first prompt excerpt, then exports the chosen session.

The exporter is dependency-free Python. It keeps main transcript Markdown readable and writes full tool payloads to linked tool-details/*.tools.md sidecar files by default.

codex-session-manager

Manages Codex JSONL transcripts from ~/.codex/sessions/YYYY/MM/DD/*.jsonl and optionally ~/.codex/archived_sessions/*.jsonl. It uses the same full/specific mode pattern as claude-session-manager, with ~/.codex/session-markdown as the default output folder.

The exporter is dependency-free Python and understands Codex timestamp / type / payload events, including session metadata, messages, tool calls, and tool outputs.

mining-session-skills

The judgment layer on top of claude-session-manager. It locates a session by natural-language description ("the session where I fixed the connection-reset bug"), reads the exported compact transcript, segments multi-task sessions into topic arcs, mines friction signals (user corrections, repeated manual steps, supplied domain knowledge), and applies a worth-it gate — only proposing a skill when the work involves non-obvious process, taste/judgment, or a divergence from the standard approach. A clean "nothing worth making here" is a valid result.

It is conversational and gated: it confirms the session, then proposes create/update/reuse with evidence before writing anything — nothing is drafted without approval. The bundled scripts/extract_session_signals.py deterministically pulls human prompts (filtering tool results, command wrappers, and thinking blobs) with topic-arc hints, so mining stays cheap on large sessions.

Creating a New Skill

1. Create skills/<skill-name>/SKILL.md with YAML frontmatter:

``yaml --- name: skill-name description: When to trigger this skill. Be specific about trigger phrases — include Chinese and English if applicable. --- ``

1. Run ./scripts/sync-marketplace-skills.sh to regenerate the skills[] array in .claude-plugin/marketplace.json, then bump the version manually.
2. (Optional) Add references/, scripts/, assets/, or templates/ directories alongside SKILL.md for domain knowledge, automation, or report templates.

Do not add per-skill README.md files. Keep agent-facing instructions in SKILL.md and put human-facing summaries or development notes in this repository README.

See CLAUDE.md for the full convention.

Acknowledgements

• •The skills.sh installer by Matt Pocock — the cross-agent installer used in the universal install path above.
• •publish-x-article is inspired by and based on wshuyi/x-article-publisher-skill.

Contributing

Pull requests welcome — new skills or improvements to existing ones. Please follow the structure described in CLAUDE.md.

License

MIT