automation
Skill1 starsChanged 4 months agoMITView on GitHub
Workflow architecture and design across automation tools. Use for cross-tool decisions, multi-system integrations, and automation strategy. For deep, single-platform builds, use that tool's own dedicated skill or documentation.
14 other projects use a version of this.
Adopted by
Most recently taken up by
- gkurt/ai-soulstheir own version, first seen Nov 8, 2026
- yaqob3653/vlypse-skillstheir own version, first seen Nov 8, 2026
- linxule/carreltheir own version, first seen Nov 8, 2026
- jstilb/ai-assistanttheir own version, first seen Oct 30, 2026
- harshsinghmp/muse-skillstheir own version, first seen Oct 15, 2026
- cutec-chris/PawLiatheir own version, first seen Oct 15, 2026
Used with
Projects with it also use refactor.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
GitHubSkillrefactor
Surgical code refactoring to improve maintainability without changing behavior. Covers extracting functions, renaming variables, breaking down god functions, improving type safety, eliminating code smells, and applying design patterns. Less drastic than repo-rebuilder; use for gradual improvements.
Also in 3 projects that use this
Other versions14 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 14
- vercel-labs/native★ 7.8ksame name
- vercel-labs/zero-native★ 7.8ksame name
- proophboard/skills★ 6same name
- ffsshhttiikk/opencode-agents-skills★ 2same name
- linxule/carrel★ 2same name
- AHX47/claw-skills-any★ 1same name
- alizafarbati/opencode-agents-mcp★ 1same name
- cutec-chris/PawLia★ 1same name
- harshsinghmp/muse-skills★ 1same name
- yaqob3653/vlypse-skills★ 1same name
4 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository60 other files it gives its agents.
- abmSkill
- agent-teamSkill
- analyticsSkill
- blog-auditSkill
- blog-postSkill
- brandingSkill
- briefSkill
- business-acquisitionSkill
- business-hacksSkill
- chrome-extensionSkill
- claude-code-auditSkill
- closingSkill
- cold-outreachSkill
- communitySkill
- content-ideaSkill
- content-planSkill
- contentSkill
- croSkill
- customer-successSkill
- emailSkill
- entrepreneurSkill
- fact-checkSkill
- financeSkill
- fundraisingSkill
- ghlSkill
- growthSkill
- gtmSkill
- handoffSkill
- hooksSkill
- humanize-strip-aiSkill
- leadsSkill
- local-seoSkill
- manualSkill
- marketing-hacksSkill
- n8nSkill
- nurtureSkill
- offerSkill
- optimize-vs-skillSkill
- paid-mediaSkill
- partnershipsSkill
- personal-brandSkill
- portable-promptSkill
- product-marketingSkill
- productSkill
- prSkill
- replit-qualitySkill
- researchSkill
- sales-managementSkill
- save-sessionSkill
- scalingSkill
- seo-hacksSkill
- seoSkill
- social-postSkill
- technical-auditSkill
- tone-of-voiceSkill
- ui-ux-auditSkill
- uxSkill
- vibe-codingSkill
- web-designSkill
- youtube-seoSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
18 rules. 0 are common, 17 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 11 edited versions we read keep these.
17
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
If not specified, ask one question to clarify before proceeding.
- Distinctive: line 27
Deterministic inputs over AI guessing: If the data needed is structured and predictable, do not use an AI node.
- Distinctive: line 28
Warn about downstream breakage: Every workflow has failure modes.
- Distinctive: line 29
Simpler is better: If a 3-step workflow solves the problem, do not build a 10-step workflow.
- Distinctive: line 31
Error handling is not optional: Every workflow that touches external APIs needs error paths.
- Distinctive: line 38
Before building anything, run this check.
- Distinctive: line 66
Prefer Code nodes over complex node chains when logic is conditional.
- Distinctive: line 67
Use Set node for data normalization before passing to external services.
- Distinctive: line 70
Always test with real data, not mock data.
- Distinctive: line 76
Use data stores for state that needs to persist between runs.
- Distinctive: line 96
No native integration exists and you would otherwise write custom code
- Distinctive: line 111
When analyzing data from any platform, flag repeatable judgment calls that could become automated rules
- Distinctive: line 112
Look for patterns across data sources the user may not have connected yet (e.g.
- Distinctive: line 113
If a manual review process follows the same decision tree every time, propose turning it into an automation
- Distinctive: line 131
Test case: input example and expected output
- Distinctive: line 144
When choosing between automation approaches:
- Distinctive: line 148
Prefer the option that is easiest to debug when it breaks
- Very specific: line 149
Default to n8n/Make.com for deterministic, scheduled data movement.
“Common” is measured: kept by at least 6 in 10 of the 11 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Automation Skill
This skill handles cross-tool architecture and design decisions. For deep, single-platform work (node config, workflow recipes, funnel setup within one tool such as n8n or GoHighLevel), use that tool's own dedicated skill or documentation.
Automation work defaults to deterministic, simple architectures. AI is a last resort inside workflows, not a default. Every automation should be auditable, debuggable, and maintainable.
Project context is loaded from the active CLAUDE.md. Use the tech stack and tools defined in that context.
When invoked
$ARGUMENTS specifies the tool and the workflow goal or problem. If not specified, ask one question to clarify before proceeding.
Design principles
- Deterministic inputs over AI guessing. If the data needed is structured and predictable, do not use an AI node. Use a formula, lookup, or conditional.
- Warn about downstream breakage. Every workflow has failure modes. Call them out before they happen.
- Simpler is better. If a 3-step workflow solves the problem, do not build a 10-step workflow.
- One job per workflow. Monolithic automations break and are hard to debug. Keep scope narrow.
- Error handling is not optional. Every workflow that touches external APIs needs error paths.
- Test before shipping. Define the test case before building. What input produces what output?
Should you automate this?
Before building anything, run this check.
Automate when:
- The task takes 15+ minutes AND happens weekly or more frequently
- The process is error-prone when done manually (wrong data, missed steps, inconsistent execution)
- It requires combining data from multiple sources before making a decision
- A non-obvious automation could remove an entire category of work, not just speed up one instance
Skip automation when:
- It's a one-off or happens less than monthly
- Manual effort is under 10 minutes per occurrence
- The setup cost exceeds 3 months of manual time saved
- A CSV export/import or copy-paste gets the job done in one sitting
Think creatively when:
- You're making the same judgment call repeatedly (the pattern itself can become a rule)
- Data lives in one system but decisions happen in another (bridge them)
- You're doing manual QA or checks against known thresholds (rule-based automation)
- A human is acting as the "glue" between two systems (that glue is the automation)
Quick math: (minutes per occurrence × occurrences per month) vs. (hours to build + hours to maintain per month). If the ratio does not pay back in 3 months, skip it or simplify.
Tool-specific guidance
n8n
- Prefer Code nodes over complex node chains when logic is conditional.
- Use Set node for data normalization before passing to external services.
- HTTP Request node for APIs without native integrations.
- Webhook triggers for real-time flows; Cron for scheduled jobs.
- Always test with real data, not mock data. Edge cases appear in production, not in samples.
- For example, an SEO blog pipeline: keyword list input node → content QA checklist node → internal linking targets check → publish.
Make.com (Integromat)
- Routers over separate scenarios when flows share a trigger.
- Use data stores for state that needs to persist between runs.
- Watch the operation count. Loops inside loops kill your quota.
- Error handlers on every module that calls an external API.
- Filters at the router level, not inside individual modules.
GoHighLevel
- Pipeline stages should map 1:1 to a real buying stage, not internal team stages.
- Automation triggers: be explicit about what fires them. Ambiguous triggers cause duplicate actions.
- Workflow branching: test every branch. Default paths should handle the unexpected case.
- CRM segmentation: tag-based systems break at scale. Use contact fields for permanent attributes, tags for temporary states.
- Sub-accounts: keep automations at the sub-account level unless there is a clear reason to use agency-level triggers.
Claude (API + MCP)
Claude can act as an automation layer by calling APIs directly, using MCP connections, or receiving webhooks. Use this when the automation requires judgment, analysis, or context-aware decisions rather than simple data movement.
When to use over n8n/Make.com:
- The "what to do" decision requires analyzing the data, not just routing it
- One-off or low-frequency tasks where building a full workflow is overkill
- No native integration exists and you would otherwise write custom code
- Chaining multiple API calls with conditional logic based on returned data
How it works:
- Direct API calls to any REST API (ad platforms, CRMs, analytics, etc.)
- MCP connections to pull and push data from connected services
- Webhook receivers to process inbound data and act on it
- The pattern: pull data → analyze → decide → push changes back
Where it fits in the stack:
- n8n/Make.com: scheduled, high-frequency, deterministic workflows (move data A to B on a timer)
- GHL: CRM-triggered automations (contact enters stage → fire sequence)
- Claude via API/MCP: analysis-heavy tasks, irregular cadences, decisions requiring context or judgment
Non-obvious opportunities:
- When analyzing data from any platform, flag repeatable judgment calls that could become automated rules
- Look for patterns across data sources the user may not have connected yet (e.g. ad spend data + CRM close rates = automated budget reallocation logic)
- If a manual review process follows the same decision tree every time, propose turning it into an automation
Constraints:
- Auth tokens and API keys must be managed securely. Do not store credentials in plain text.
- Rate limits apply to every API. Build in handling for 429 responses.
- For high-frequency scheduled jobs (every 5 min, every hour), n8n or Make.com is still the better choice. Claude is better for on-demand or weekly cadence tasks.
Output formats
Workflow design
- Goal: what problem does this solve?
- Trigger: what starts the workflow?
- Steps: numbered, one action per step
- Data: what goes in, what comes out at each step
- Error handling: what happens when a step fails?
- Test case: input example and expected output
- Edge cases to handle
Workflow review
- What it does (confirm understanding)
- Failure modes identified
- Simplification opportunities
- Missing error handling
- Recommended changes, prioritized
Architecture recommendation
When choosing between automation approaches:
- State the options clearly, including Claude via API/MCP when analysis or judgment is required
- Give a verdict with reasoning
- Flag cost implications (API calls, operations, time, token usage for Claude-based automations)
- Prefer the option that is easiest to debug when it breaks
- Default to n8n/Make.com for deterministic, scheduled data movement. Default to Claude for tasks where the action depends on interpreting the data.
Common automation patterns
Content publishing pipeline Keyword input → content brief generation → draft creation → QA checklist → internal link injection → CMS publish → tracking setup
Lead intake and routing Form/ad submit → dedup check → lead scoring → pipeline stage assignment → owner assignment → follow-up sequence trigger → notification
Reporting pipeline Data pull from source → transformation → aggregation → push to spreadsheet or dashboard → notification (Slack, email, etc.)
Outreach sequence Lead list import → dedup → sequence enrollment → touchpoint 1 → wait → reply check → touchpoint 2 or remove → conversion tracking
API-driven performance optimization Pull platform report via API (ad spend, search terms, pipeline data) → analyze against targets → flag anomalies or waste → push changes back (negative keywords, bid adjustments, status updates) → log actions taken
Cross-platform data sync with analysis Pull data from source API → compare against second source (e.g. ad platform vs CRM) → identify discrepancies → reconcile or flag → push corrections
Ad creative production line (Claude Code as orchestrator) Performance data import (CSV or API/MCP pull from Meta, Google Ads) → analyze against targets (CTR, CPA, ROAS, frequency) → flag underperformers → sub-agent 1 generates headline variants (matched to the audience's awareness level) → sub-agent 2 generates description/body copy variants → output structured creative brief (CSV, JSON, or direct push to design tool API like Figma) → optional: MCP connection pulls live platform data to close the feedback loop → log all variants and performance deltas for next iteration
Why this works: Claude Code runs sub-agents in parallel, so headline and description generation happen simultaneously. One person can produce 10x the creative volume in a fraction of the time. The loop closes when live platform data feeds back into the next analysis cycle.
Requirements: ad platform API credentials (or CSV exports as a simpler start), MCP server for live data pull (optional), design tool API or template system for final asset production (optional). Start with CSV in, CSV out. Add MCP and design tool connections as the workflow proves value.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

