Compare¶
AgentCraft vs. the alternatives — when to use which.
AgentCraft vs. raw Hermes Agent (240+ skills)¶
| AgentCraft | Raw Hermes | |
|---|---|---|
| Curation | 20-40 O&G-relevant artifacts, vetted | 240+ skills, mixed quality |
| Discovery | Public site with search + filter | hermes skills list in terminal |
| Trigger prompts | 3 copy-paste sentences per artifact | You write your own |
| Source attribution | Every artifact links to its Drive canon PDF | None |
| Maintenance | CI runs every flow weekly | Curator (idle skill detection) |
| Cost transparency | Per-artifact cost estimate on the page | You figure it out |
| Offline scripts | First-class citizens (stdlib Python) | Mixed |
| Private | Public (read by anyone) | Private to you |
Use AgentCraft when: you want to find a specific O&G or API-auditing artifact fast, see a sample output, copy a trigger prompt, run it.
Use raw Hermes when: you're building a one-off skill for a personal workflow that doesn't need to be public.
AgentCraft vs. Claude Code plugin ecosystem¶
| AgentCraft | Claude Code plugin | |
|---|---|---|
| Unit of value | The trigger prompt (recipe) | The install command (module) |
| Distribution | GitHub repo + static site | Plugin marketplace |
| Format | SKILL.md + scripts + flows |
Plugin with slash commands, agents, hooks |
| Use case | "I want to do X — show me how" | "I want to extend Claude Code with capability Y" |
| Lock-in | None — pure files, no plugin manager | Plugin manager + marketplace |
| Versioning | semver in the filename, CHANGELOG.md | Marketplace versioning |
Use AgentCraft when: you're a domain expert with a real workflow to share, and you want a low-overhead way to publish it.
Use Claude Code plugins when: you're building a complex multi-agent system with hooks, statuslines, and slash commands.
AgentCraft vs. n8n / Make.com / Zapier¶
| AgentCraft | n8n / Make / Zapier | |
|---|---|---|
| Interface | Trigger prompt (text) | Visual workflow editor |
| Execution | In your agent (LLM in the loop) | Serverless functions |
| Best for | Tasks that need LLM judgment + multi-step + multi-tool | Tasks that are pure data plumbing, no judgment |
| Cost | Per-token + per-API-call | Per-operation (often flat fee) |
| Local | Yes | No (cloud) |
| Custom code | Python scripts in canon/ |
Code nodes (sandboxed) |
| Debugging | Read the agent's transcript | Visual node inspector |
Use AgentCraft when: the task needs the agent to interpret, decide, or write — e.g., "review this NCR against API Q1 11th ed." or "produce a 7-doc audit deliverable."
Use n8n when: the task is pure data plumbing — e.g., "when a new row hits Airtable, send a Slack message." No judgment, no LLM needed.
AgentCraft vs. Ahrefs / Semrush (for SEO audits)¶
Not comparable directly. AgentCraft is a library of artifacts, not a SaaS. But for the small overlap (the "build a website" flow includes SEO basics like meta tags, OG tags, sitemap, robots.txt), AgentCraft is free and the artifact is copy-paste, not a $99/mo subscription.
AgentCraft vs. a Notion / Confluence wiki¶
| AgentCraft | Wiki | |
|---|---|---|
| Format | Executable code + agent-loaded skills | Static documents |
| Runnable | Yes | No |
| Trigger prompts | Yes | No |
| Version control | git, semver, CHANGELOG | Page history (no semver) |
| Search | MkDocs Material search + Drive | Notion / Confluence search |
| Update friction | PR + CI | Click "Edit" |
Use AgentCraft when: the artifact is a procedure the agent can run.
Use a wiki when: the artifact is reference material the human reads (a glossary, a policy doc, a knowledge base).
The decision in one sentence¶
AgentCraft is for "I want to do this O&G or API-auditing task and I want a recipe I can run in 30 seconds" — not for "I want to extend my agent" (Claude Code plugin), "I want to plumb data between SaaS apps" (n8n/Make), or "I want to look something up" (wiki).