agent-skills

Score and grade OpenClaw skills against the AgentSkills spec, apply fixes, and explain why each change is better. Use this skill whenever someone asks to score, grade, rate, or benchmark a skill's quality — even if they just say 'is this good enough to ship' or 'how does this skill look'. Use when: (1) scoring a skill before shipping, (2) grading skill quality against the spec, (3) checking if a skill is production-ready, (4) evaluating trigger description accuracy, (5) benchmarking a CLI wrapper skill for completeness, (6) running the evaluate→fix→explain loop on any skill. Triggers on: 'skill eval', 'score this skill', 'grade this skill', 'rate this skill', 'is this skill good enough', 'is this skill production quality', 'evaluate this skill', 'how good is this skill', 'benchmark skill quality'. NOT for: creating or structurally editing skills (use skill-creator), running behavioral test cases in Docker/subagents (use Claude Code's eval framework or Skill Eval), editing agent workspace files directly.

Build and use a user-owned communication style profile with the VoiceForge CLI. Use when importing voice transcripts or writing samples, analyzing how someone communicates, refreshing or exporting a portable voice profile, writing with confirmed style guidance, managing cloned voices, or generating speech audio.

Design or implement polished ASCII and ANSI animations for command-line and terminal applications. Use when adding spinners, loading states, splash screens, animated banners, terminal effects, or CLI/TUI motion in Go, JavaScript, TypeScript, Rust, or Python.

Interview a user and generate SOUL.md, USER.md, AGENTS.md, TOOLS.md, IDENTITY.md, and MEMORY.md files for OpenClaw, Hermes, or other agent harnesses. Use when onboarding a person or fleet, designing a distinct agent persona, creating a SoulForge profile, or auditing generated identity and context files.

Work with the CLWatch changelog-tracking CLI when the user asks to inspect AI coding-tool releases, compare changelog versions, or check what changed. CLWatch and changelogs.info are currently being rebuilt, so discover the installed command surface before use and do not assume legacy commands or APIs still exist.

Apply mental models and cognitive frameworks to problems, decisions, and content analysis. Use when the user asks to 'think through' a problem, says 'apply mental models', mentions specific models like inversion/second-order thinking/bottlenecks/first principles, needs structured decision analysis, wants to understand cognitive biases at play, or asks 'what framework should I use'.

Turn a conversation, voice note, transcript, or guided interview into one source-grounded canonical article and reviewed channel-specific drafts. Use when someone wants to be interviewed about an idea, create long-form content from speech, repurpose one piece across channels, build a content package, or run the Talk Once workflow with an optional VoiceForge profile.

Forge and manage fleets of OpenClaw agents. Use when: (1) creating/listing/inspecting agents (create, list, inspect), (2) managing agent lifecycle (bind, activate, deactivate, destroy, deploy), (3) sharing agents (export, import), (4) fleet observability (status, cost, logs, health), (5) fleet sync and drift detection, (6) comparing agents (diff). NOT for: editing individual agent files (use read/edit tools directly).

Recursive breadth/depth research using parallel sub-agents. Spawns workers to search, scrape, and extract findings iteratively, then synthesizes a final report. OpenClaw-native (web_search/web_fetch only, no external APIs). Use when the user says 'research [topic]', 'deep research', 'deep dive', 'investigate', 'look into', 'find out about', or wants thorough multi-source research on any topic. NOT for quick single-query lookups.

Control Instagram via instagram-cli — browse feed, send/read DMs, view and post stories, upload content, manage multiple accounts. Use when the user mentions Instagram, asks to check/post DMs, view stories, upload a photo/reel, or manage Instagram accounts.

Run multi-model parallel planning workflows using the multiplan CLI. Sends a task simultaneously to Claude (correctness lens), Gemini (scale lens), Codex (speed lens), and GLM-5 (critic lens), cross-examines the results, then converges on a weighted final plan. Use when the user asks to: plan something with multiple models, run a multiplan, get a stress-tested technical plan, use multiplan on a task, plan with all models, or run parallel planning. NOT for simple one-model plans — use this when the user explicitly wants multi-model or multiplan.

Mine conversations for research-worthy topics using LCM search, deep research them via deep-research skill, store findings, and report digests. Closes the loop between conversations and knowledge. Use when the user says 'research flywheel', 'mine conversations', 'find research topics', 'what should we research', or wants to surface insights from past sessions.

Validate, publish, and distribute OpenClaw skills to the agent-skills mono-repo and ClawHub. Checks SKILL.md structure, frontmatter compliance, and line counts. Use when the user says 'publish a skill', 'ship this skill', 'validate a skill', 'distribute skill', 'add skill to repo', or mentions skill-publisher.

Break down a YouTube video, article, or local file into structured findings, lens-ranked ideas, and output notes using the breakdown CLI. Use when the user asks to analyze, break down, summarize into actionable notes, or turn content into a vault note, PRD, or task list.

Write, rewrite, or edit copy in CyperX's exact voice for websites, bios, project pages, notes, changelogs, and product copy. Use when the user asks to "write like me", match their talking style, sound more like CyperX, mine transcripts for voice, or turn stiff/generic writing into a direct, casual, internet-native builder voice.

Research spin-off product ideas from an existing project or concept. Use when: the user says 'research spinoff', 'spin off ideas', 'what else could we build from X', 'what other apps/tools from this', or wants to explore new products derived from an existing project. Runs a multi-agent evaluation chain: research → evaluate → scout competition → rank → present.