> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentweb.pro/llms.txt
> Use this file to discover all available pages before exploring further.
> ## Agent Instructions
> ---
> About this documentation: Emma is the AI marketing agent built by AgentWeb (the product is also marketed as HireEmma; AgentWeb, Emma, and HireEmma refer to the same product). Emma learns a company's brand, ICP, and positioning, then executes marketing work inside that company's own accounts: research, ad campaigns on Google Ads, Meta, and LinkedIn Ads, LinkedIn and email outbound, organic content, CRM lead enrichment and ICP scoring, and landing pages deployed to the customer's domain. Actions that spend money or post publicly require the user's confirmation.
> For what Emma can read, draft, publish, and spend on per channel, see https://docs.agentweb.pro/capabilities. For an overview of the product, see https://docs.agentweb.pro/what-is-emma. For terminology (skill, work plan, scheduled chat, template), see https://docs.agentweb.pro/concepts.
# Skills
> Teach Emma reusable rules and judgment — what to enforce, what tools to use, and what never to do — so she applies them consistently every time you install the skill.
A **skill** is a bundle of tool grants plus standing rules that change how Emma behaves while it's active. Some skills are mostly tools — Email Management grants `email` and `list_connected_accounts`. Others are almost entirely methodology — Meta Hook Engineering grants only three read tools and is 90% standing rules about how to write ad hooks. Most are somewhere in between.
The standing rules are what make skills more than a tool list. A skill can enforce hard constraints — never fabricate metrics, cost is in micros not dollars, draft-only never auto-send, recommendation-only never deploy — that hold for the entire conversation. Without the skill installed, those constraints don't exist.
**Skills are per-conversation, not a permanent setting.** Installing a skill loads it for the current chat. A new conversation starts with foundation tools only — your skills don't carry over. If you need a skill every time, install it at the start of each conversation, or use a [work plan](/user-guide/work-plans) that installs it automatically via its step configuration.
## Skills vs. work plans
These solve different problems and they compose.
| | Skill | Work plan |
| --------------------- | ---------------------------------------- | --------------------------------------- |
| What it is | A capability — tools plus standing rules | A sequence — ordered steps with gates |
| Question it answers | *Can* Emma do this? | *In what order*, with what checkpoints? |
| Grants tools | Yes | No — it installs skills that do |
| Has order and state | No | Yes |
| Persists across chats | No | No (but the plan definition persists) |
Reach for a **skill** when the reusable thing is judgment and standing rules — "here's how we evaluate quality, always apply this lens." Reach for a **work plan** when the reusable thing is a sequence with checkpoints — "do these five steps in this order, stop here for my approval."
They connect at the step level: a work plan step can declare the skills it needs, and launching the plan auto-installs them. Mature capabilities often ship as both — Meta Ads has skills *and* plans that sequence them.
## Built-in skills
AgentWeb ships a catalog of skills built and vetted by professional marketers and founders — the same people who built the [work plans](/user-guide/work-plans). These cover paid ads (Google, Meta, LinkedIn), organic content (LinkedIn posts, X/Twitter), outbound (email, LinkedIn outreach, cross-channel sequences), CRM and lead management, SEO, meeting intelligence, and more.
You don't need to install these manually for most work. When you ask Emma to do something, she installs the relevant skills automatically. But knowing they exist helps when you want to be specific — ask Emma *"what skills do you have for Meta ads?"* and she'll list them with what each one does.
## Create a skill
There are two paths.
### Conversational: ask Emma to build it
The best skills come from work you've already refined. When you and Emma have developed a process that works — you've iterated on the rules, corrected the edge cases, and the output is consistently what you want — ask her to save it.
*"Make this a skill"* or *"Save this as a skill called \[name]"* works. Emma composes the skill from what you've built together: the rules become the system prompt, and the tools you used become the tool grants.
**Example — a blog review SOP:**
You've been working with Emma to review blog articles before publication. Over several conversations, you've dialed in the checklist: every post must reference at least one ICP persona from Market Strategy, the CTA must match the campaign goal, no unsupported claims, tone must match founder voice in Brand Guidelines, and the post must be under 1,200 words.
Once that checklist is stable, say:
*"Make this a skill called Blog Review SOP. Every time I ask you to review a blog post, apply these rules."*
Emma saves it. Next time you start a new conversation and say *"review this blog post,"* install the skill first — or ask Emma to install it — and she applies the full checklist without you repeating it.
**This is a skill because the reusable thing is judgment, not a sequence.** The value is the standing rules — the checklist Emma applies every time. If your review also needed ordered steps with approval gates between them ("first check structure, then let me approve before checking tone"), that would be a [work plan](/user-guide/work-plans) instead. Many SOPs could go either way; choose based on whether the *order and checkpoints* matter or just the *rules*.
### Manual: build it in Settings
Go to **Settings → Skills** and create the skill directly. This gives you control over each field:
| Field | What it does |
| ------------- | ------------------------------------------------------------------------------------------- |
| Name | How the skill is found and installed. Keep it short and distinctive. |
| Description | What the skill does — shown when listing available skills. |
| System prompt | The standing rules Emma follows while the skill is active. This is where methodology lives. |
| Tool names | The tools this skill grants access to. |
The manual path is better when you're writing rules from scratch rather than capturing something you've already done, or when you want to edit the system prompt directly.
## Skill sharing
### Default: shared with your whole organization
If you belong to an organization, a skill is visible to every member of it the moment you save it. There is no per-skill privacy step and no share action to take — creating a skill stamps it with your organization, and org-wide skill sharing is **on by default**. Teammates see it in their own **Settings → Skills** list, tagged *"Shared by \[your name]"*, and can install it in their conversations.
**Don't put anything confidential in a skill.** The name, description, system prompt, and tool list are readable by every member of your organization as soon as you save. Never paste credentials, API keys, passwords, or account logins into a system prompt, and leave out customer names and other details you wouldn't show the whole team. Write the skill assuming a teammate will read it — one can.
The **Private** badge on a skill means "not published to all AgentWeb users." It does not mean hidden from your organization.
If you aren't in an organization, your skills are yours alone — there's no one to fan them out to.
### Turn org-wide skill sharing off
Skill sharing is one org-wide switch, not a per-skill setting. An org owner or admin can turn it off at **Settings → Organization → Org-wide Sharing** with the **Skills** toggle; plain members can't change it and don't see the section. With it off, each member sees only their own skills in **Settings → Skills**.
Treat that toggle as a visibility setting, not access control. It hides other members' skills from the lists; it does not lock a skill a teammate has already found, because Agent Mode still installs any skill carrying your organization's ID. Keeping secrets out of skills is the protection that holds.
### Share with all AgentWeb users
If your skill would be valuable beyond your organization, you can submit it for review to share with the entire AgentWeb user community. Submit from **Settings → Skills**.
The AgentWeb team reviews submissions before they go live platform-wide. Not every submission is accepted — the bar is whether the skill is general enough to be useful across different businesses and industries, rather than specific to one organization's workflow.
**Prove it inside your org first.** Build the skill, run it for a week, fix what breaks, and get a teammate to try it — they can already see it. Then submit for platform review if it's genuinely general. Skipping straight to submission usually means a rejection and a rework.
## Tips for writing good skills
* **Be specific in the system prompt.** "Review content for quality" is too vague. "Check that every claim has a source, the CTA matches the campaign goal, and the word count is under 1,200" is a skill.
* **State what the skill must NOT do.** The best built-in skills include explicit constraints — never fabricate metrics, never auto-send, never deploy without approval. Negative rules prevent the failures that positive rules miss.
* **Grant only the tools the skill needs.** A blog review skill that can also send emails will eventually send an email.
* **Name it for retrieval.** Emma matches skill names when you ask her to install one. "Blog Review SOP" beats "My Content Quality Checker v2."
* **Test in a fresh conversation.** The skill that worked when you built it had the full conversation context. A fresh chat with only the skill installed shows you what's actually in the system prompt versus what was in your head.
## Troubleshooting
The skill may not be installed in this conversation. Ask *"what skills are active right now?"* Skills don't persist across chats — you need to install them each time, or use a work plan that installs them automatically.
The original conversation had context the skill doesn't capture — your corrections, examples, and clarifications lived in the chat, not in the system prompt. Open the skill in **Settings → Skills**, read the system prompt, and add what's missing.
Your skills appear in **Settings → Skills**. In conversation, ask Emma to *"list my skills"* or *"find the skill called \[name]."*
They only see it if they're in the same organization and org-wide skill sharing is on. An owner or admin can check the **Skills** toggle under **Settings → Organization → Org-wide Sharing**.
The bar for platform-wide publication is generality — the skill needs to work across different businesses. Rework it to be less specific to your org, or keep it as an org-shared skill where it's already useful.
## Next steps
* [Work plans](/user-guide/work-plans) — ordered workflows that can auto-install skills per step.
* [Agent Mode](/user-guide/agent-mode) — where you build and run skills.
* [Report a bug](/user-guide/report-a-bug) — if a skill isn't working as expected.