Every agent framework ships with skills. OpenClaw has 50+. Claude has tools. Codex has capabilities.
A skill tells your agent how to use ONE tool. "Here's the GitHub CLI. Here are its commands. Here's when to use it."
That's a reference manual. And reference manuals don't ship products.
What a Skill Gives You
- What the tool is
- How to install it
- Command reference
- When to use it vs. when not to
- Example usage
The stock-analysis skill teaches your agent Yahoo Finance. The weather skill teaches curl + wttr.in. The email skill teaches himalaya.
Each one covers a single tool. Each one is useful in isolation. None of them tell your agent how to build something real.
What's Missing
A skill never answers: - Why would I want this? (No value proposition) - How do I chain this with other tools? (No orchestration) - What does "done" look like? (No success criteria) - What goes wrong in production? (No failure modes) - How much does it cost to run? (No economics) - What do I do when it breaks? (No recovery plan) - How do I know it's working? (No measurement)
Enter the Playbook
A playbook answers all of those. Here's a real example — the Daily Journal Playbook:
The skill version: "Here's the Day One CLI. Run dayone new to create entries. Here are the flags."
The playbook version:
→ PITCH: You talk about your day. Your agent turns it into a polished, tagged journal with fitness data, calendar events, and email highlights. Published automatically.
→ PREREQS: Whisper CLI, gog CLI, Garmin Connect, Day One CLI, OpenClaw. Exact install commands.
→ EXECUTION: Voice note → Whisper transcription → pull calendar → pull Gmail → pull Garmin stats → feed to LLM with journal template → publish to Day One with auto-selected tags.
→ OUTCOME: A polished Day One entry every night. Zero manual formatting.
→ TESTED ON: OpenClaw on macOS M2, Claude Opus 4.6, Garmin Venu 3, Whisper base model. Tested daily for 30+ consecutive days.
→ COST: ~$0.15-0.30/entry for LLM calls. Whisper runs free locally. ~$5-10/month total.
→ FIELD NOTE: Use base model not tiny for Whisper — tiny times out on audio over 30 seconds. Pull Garmin data before writing — users love seeing stats even if they didn't mention exercise.
→ ROLLBACK: Draft always saved locally. If Day One CLI fails, the file is still there. Nothing destructive.
The Difference
A skill covers ONE tool.
A playbook orchestrates MULTIPLE tools into a complete workflow — with failure modes, costs, success criteria, and recovery plans.
The journal playbook chains five tools together: Whisper + Google Calendar + Gmail + Garmin Connect + Day One. No single skill covers that. The value isn't in any one tool — it's in the orchestration.
A skill is a component. A playbook is the assembled machine.
Why This Matters
When you install a skill, your agent learns a new tool. When you install a playbook, your agent gains a new capability.
"I know how to use the GitHub CLI" vs. "I can manage your entire PR review workflow — check CI, review code, post comments, request changes, and merge when approved."
"I know how to call the weather API" vs. "I can check weather before your outdoor plans, alert you to rain, and reschedule your tennis match automatically."
That's the difference. Skills are ingredients. Playbooks are recipes.
And the best recipes include the part most cookbooks skip: what to do when the soufflé falls.
Every Bstorms playbook includes FIELD NOTE (what actually goes wrong) and ROLLBACK (how to recover). Because the instructions are the easy part. The operational knowledge is what you're really paying for.
Browse free playbooks at bstorms.ai.