# Automation Workflow

> Use when building batch automation, agent pipelines, scheduled jobs, or multi-step unattended scripts. Prefer durable plans, checkpoints, and authorized scopes only.

- Skill: `tomjiu/automation-workflow` (Agent Skill)
- Install (CLI): `npx skillmds@latest add tomjiu/automation-workflow`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tomjiu/automation-workflow/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: tomjiu (https://skillmd.com/u/tomjiu)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/tomjiu/automation-workflow

---


# Automation Workflow

## When

- Multi-step jobs without constant human steering
- Batch RE/analysis, scraping, report generation, CI-like local loops
- Hermes cron / Codex long jobs / shell pipelines

## Rules

1. **Scope + authorization** written first (paths, targets, rate limits).
2. **Idempotent steps** + checkpoint files (resume after crash).
3. Log inputs/outputs; never invent tool results.
4. Prefer existing tools (Hikari search, keenable fetch, Ghidra headless, project CLIs) over new frameworks.
5. Offensive automation only on authorized systems.

## Pattern

```text
plan.md          # goals, steps, success criteria
state.json       # current step, retries, artifacts
artifacts/       # outputs
```

1. Write plan (or load `writing-plans`)
2. Execute step N; write artifact; update state
3. On failure: record error, backoff, do not silently skip verification
4. Finish with `verification-before-completion`

## Related local assets

- Search: Keenable / Tavily Hikari MCP
- RE batch: `ghidra-headless-mcp` mirror (enable when Ghidra ready)
- Skills: `writing-plans`, `executing-plans`, `reverse-engineering-workflow`

## Avoid

- Unbounded browser MCP loops for simple HTTP
- Loading every reverse skill at once