sales-simular

What this skill does

# Simular (Sai) Platform Help

## Step 1 — Gather context

If `references/learnings.md` exists, read it first for accumulated platform knowledge.

Ask the user:

1. **What's your Simular challenge?**
   - A) Setting up Sai for the first time
   - B) Agent keeps failing or getting confused during tasks
   - C) Writing or debugging Simulang scripts
   - D) Browser automation with the Python API
   - E) Credit usage or pricing questions
   - F) Comparing Simular vs alternatives
   - G) Something else

2. **Which product are you using?**
   - A) Sai (cloud-based AI coworker at sai.work)
   - B) Simular Pro (local desktop agent)
   - C) Simular Browser (Python API)
   - D) Agent-S (open-source framework)

3. **Which plan are you on?**
   - A) Free trial (7 days)
   - B) Starter/Plus ($20/mo)
   - C) Pro ($500/mo)

**If the user's request already provides most of this context, skip directly to the relevant step.**

## Step 2 — Route or answer directly

| If the question is about... | Route to... |
|---|---|
| No-code AI agent workflows with app integrations | `/sales-lindy [question]` |
| AI calendar auto-scheduling with project management | `/sales-motion [question]` |
| Choosing a dedicated AI meeting note-taker | `/sales-note-taker [question]` |
| General tool integration patterns | `/sales-integration [question]` |
| Everything else | Answer directly using Step 3 |

When routing, provide the exact command.

## Step 3 — Simular platform reference

**Read `references/platform-guide.md`** for the full platform reference — capabilities, Simulang API, pricing, integrations, and workarounds for common issues.

Answer the user's question using only the relevant section. Don't dump the full reference.

## Step 4 — Actionable guidance

- For setup: recommend starting with a simple single-app task (e.g., Gmail labeling), verify it works, then chain multi-app workflows
- For agent failures: check if the target app has changed its UI, try the `stateSatisfies` perception action to debug what the agent sees, reduce `maxSteps` to isolate where it fails
- For Simulang: start from the workflow gallery examples, modify incrementally — don't write complex scripts from scratch
- For pricing: Starter ($20/mo, 10K credits) is fine for personal automation; Pro ($500/mo) only if running business-critical 24/7 workflows
- For comparisons: Simular controls the actual desktop (GUI clicks/typing) — it can automate ANY app but is slower and less reliable than API-based tools like Zapier/Make for apps that have APIs

If you discover a gotcha or tip not covered in `references/learnings.md`, append it there.

## Gotchas

> *Best-effort from research — review these, especially items about platform limitations and pricing that may be outdated.*

- **macOS-first, Windows incomplete.** Simular Pro currently runs on macOS 15.0+ only. Windows support is announced but not fully available. Don't recommend for Windows-only users.
- **GUI automation is inherently fragile.** When apps update their UI, workflows can break. CAPTCHAs, anti-bot measures, and dynamic content can confuse the agent. Always include `stateSatisfies` checks and `Respond` confirmation gates for critical actions.
- **No webhooks, no Zapier, no Make.** Simular operates by controlling the GUI — it doesn't integrate via APIs with other automation platforms. You can't trigger a Simular workflow from a Zapier zap. For API-based automation, use dedicated tools.
- **Credits are opaque.** The 10,000 credits on Starter don't map to a clear unit (not tasks, not minutes). Monitor usage closely in the first month to calibrate.
- **Pro plan is expensive for solopreneurs.** At $500/mo, Pro is positioned for businesses. Most solopreneurs should start with Starter ($20/mo) and only upgrade if they need unlimited credits or team features.
- **Agent-S open-source != Sai cloud product.** Agent-S is a research framework. Sai is the commercial product with cloud VMs and support. Don't conflate them — Agent-S requires self-hosting and your own LLM API keys.
- **Non-English text input can fail.** Users report issues entering CJK characters. If automating non-Latin-script workflows, test thoroughly.

## Related skills

- `/sales-lindy` — Lindy platform help (no-code AI agent builder with 4,000+ integrations — API-based alternative)
- `/sales-motion` — Motion platform help (AI calendar + project management)
- `/sales-saner` — Saner.AI platform help (AI personal assistant for notes/tasks)
- `/sales-do` — Not sure which skill to use? The router matches any sales objective to the right skill. Install: `npx skills add sales-skills/sales --skill sales-do -a claude-code`

## Examples

### Example 1: Solopreneur automating LinkedIn prospecting
**User says**: "I want Sai to visit LinkedIn profiles from a Google Sheet list and send personalized connection requests"
**Skill does**:
1. Reads platform-guide.md for Simulang reference
2. Explains the workflow: Open Google Sheets → GetGoogleSheetCellValue for each row → Open LinkedIn profile → Click "Connect" → Type personalized note → confirm via Respond
3. Warns about LinkedIn rate limits and anti-bot detection — recommend max 20-30 connections/day
4. Shows the approval gate pattern using `Respond({requireConfirm: true})` before each send
5. Notes this is Pro-tier territory if running 24/7 on cloud VM
**Result**: Working Simulang script outline with safety guardrails

### Example 2: Debugging a failing workflow
**User says**: "My Sai agent keeps clicking the wrong button in our CRM"
**Skill does**:
1. Explains the Click disambiguation system — `at` parameter, `mode: "vision"`, spatial relations, anchor concepts
2. Recommends using `ConceptsExist` to verify the right page loaded before clicking
3. Shows how to use `GetElements` to inspect what the agent sees vs what's actually on screen
4. Suggests adding `waitForLoadComplete: true` and `waitTime` to handle slow-loading CRM pages
5. If the CRM has an API, recommends using API calls instead of GUI automation for reliability
**Result**: Diagnostic approach and Click configuration to improve accuracy

### Example 3: Comparing Simular vs Zapier vs Lindy
**User says**: "Should I use Simular, Zapier, or Lindy to automate my workflow?"
**Skill does**:
1. Clarifies the fundamental difference: Simular controls the GUI (can automate ANY app), Zapier/Make connect APIs (faster, more reliable, but limited to supported apps), Lindy builds AI agent workflows with integrations
2. Decision framework: If the app has a Zapier integration or API → use Zapier/Make. If you need AI reasoning in the loop → use Lindy. If the app has no API and you need GUI automation → use Simular.
3. Cost comparison: Zapier from $0, Lindy from $50/mo, Simular from $20/mo ($500/mo for business use)
4. Recommends Simular only for "last mile" automation where no API exists
**Result**: Clear decision framework based on whether apps have APIs

## Troubleshooting

### Agent gets stuck on CAPTCHAs
**Cause**: Google and other services detect automated interaction and present CAPTCHAs
**Solution**: Use `Respond({requireConfirm: true})` to pause and alert you when a CAPTCHA appears. Solve it manually, then let the agent continue. For Google services, consider using the Google Sheets API actions instead of navigating the web UI.

### Agent clicks wrong element
**Cause**: Ambiguous UI elements, slow page loads, or elements hidden behind overlays
**Solution**: Use `Click({mode: "vision"})` for visual targeting. Add `WaitForConcepts` before clicking to ensure the target element is loaded. Use `anchorConcept` and `spatialRelation` to disambiguate (e.g., the "Submit" button closest to "Contact Form"). Inspect with `pageContent()` to see what the agent sees.

### Credits running out quickly
**Cause**: Complex multi-step workflows burn credits per action. Long `Act()` calls with high `maxSteps` consume more.
**Solution**: Break workflows into shorter scripts with explicit steps instead of open-ended `Act()` calls. S