Windmill

Build automation workflows and internal tools with Windmill's code-first platform.

Security Scan

The skill is an instruction-only guidance document for Windmill and is internally consistent with its stated purpose; it doesn't request credentials, install software, or instruct the agent to access unrelated system data.

Purpose & Capability

Name/description (Windmill automation) match the SKILL.md content, which contains platform-specific guidance (scripts, flows, secrets, webhooks, self-hosting). Nothing requested or required is unrelated to building/operating Windmill workflows.

Instruction Scope

SKILL.md is advisory only and does not tell the agent to read arbitrary files, environment variables, or external endpoints. It warns about not storing secrets in variables and about webhook auth—these are prudent operational notes, not scope creep.

Install Mechanism

No install spec and no code files are present (instruction-only), so nothing is written to disk or fetched at install time.

Credentials

The skill declares no required environment variables, credentials, or config paths. The guidance discusses secrets handling but does not attempt to access or require any secrets itself.

Persistence & Privilege

always is false and model invocation is allowed (default). The skill does not request persistent presence or system-wide config changes.

Assessment

This skill is a documentation/instruction-only helper for Windmill workflows and appears coherent and low-risk. Before installing, confirm you trust the publisher/source (metadata shows no homepage and 'Source: unknown'), because instruction-only skills can still influence agent behavior. If you expect the agent to interact with a Windmill instance, ensure any required credentials are provided only through official, secure connectors and keep secrets out of plaintext variables as the guide advises.

Script Traps

• Main function signature determines input schema — Windmill infers from type hints, wrong types break the UI form
• Return value is the script output — forgetting to return means downstream steps get nothing
• Python dependencies go in inline requirements.txt comment — not a global file, each script is isolated
• TypeScript runs on Bun — Node.js-specific APIs may not work

Flow Execution

• results.step_name fails if step hasn't run yet — conditional branches cause undefined access errors
• Parallel branches need explicit configuration — default is sequential, not concurrent
• Suspend steps wait forever without timeout — set explicit timeout or flow hangs indefinitely
• Error handlers only catch step failures — script syntax errors bypass handlers

Scheduling Pitfalls

• Timezone defaults to server timezone — set explicitly or jobs fire at unexpected times
• Concurrent execution allowed by default — add mutex lock if jobs shouldn't overlap
• Schedules attach to scripts/flows — no standalone schedule entities, delete script = delete schedule

Secrets and Variables

• Secrets cannot be read back from UI after creation — store originals securely elsewhere
• Variables are plaintext and visible — never put sensitive data in variables, only secrets
• Path format matters — u/username/secret for user, f/folder/secret for shared

Self-Hosting

• PostgreSQL is the only state — workers are stateless, back up only the database
• Single container includes workers — fine for small loads, separate workers for scale
• Worker count determines parallelism — one worker = one concurrent script execution

Webhook Triggers

• Each script/flow gets unique webhook URL — changes if you rename the script
• Webhook payload becomes script input — schema must match expected arguments
• No built-in auth on webhooks — validate tokens in script logic or use reverse proxy

Common Mistakes

• Testing flows without testing scripts first — debug scripts individually
• Expecting state between runs — use variables or external storage for persistence
• Hardcoding paths — use wmill.get_resource() for portability between workspaces