Nesta página

Automation Skills for AI Agents: Recurring Work, Browsers and CI

Last updated: 2026-11-10

An automation workflow connecting a trigger, a task and a reviewed result
SkillHey illustration: define the trigger, the work and the review point.

The right automation depends on what repeats. Marketing Loops helps specify a recurring marketing process, Nature Literature Pipeline focuses on literature monitoring, Browser Harness connects an agent to a browser, and CI/CD and Automation addresses software delivery pipelines.

These Skills guide different tasks. Their published instructions describe workflows and dependencies; they do not establish that a scheduler, browser session, account integration or deployment already exists in your environment. This guide draws on source documentation and catalog records, not a comparative execution test.

Pick the repeated job

Install this Skill.
npx skills add coreyhaines31/marketingskills --skill marketing-loops
Installation options
Install with shared product context for Claude Code.
npx skills add coreyhaines31/marketingskills --skill product-marketing marketing-loops -a claude-code
List the collection’s available Skills.
npx skills add coreyhaines31/marketingskills --list
Install for Codex in the current project.
npx skills add Yuan1z0825/nature-skills --agent codex --skill nature-literature-pipeline --yes --copy
Installation options
Install for Claude Code in the current project.
npx skills add Yuan1z0825/nature-skills --agent claude-code --skill nature-literature-pipeline --yes --copy
Instalar conforme o autor
npx skills add addyosmani/agent-skills --skill ci-cd-and-automation

Agent support

Declared by the skill author

SkillAgent support
Marketing LoopsClaude Code, Codex CLI, Cursor, GitHub Copilot, DeepSeek Harness, Gemini CLI, Grok Build, Hermes Agent, OpenCode, Pi, Kiro, Rovo Dev, Qoder, Mistral Vibe, Google Antigravity
Nature Literature PipelineClaude Code, Codex CLI, Hermes Agent, OpenCode
Browser HarnessClaude Code, Codex CLI, Cursor
CI/CD and AutomationClaude Code, Codex CLI, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Kiro, Google Antigravity

A full list is a declaration, not a test.

Need Starting point What to define
Repeat a marketing review Marketing Loops Cadence, input, output and reviewer
Monitor new literature Nature Literature Pipeline Search sources, scoring rules and delivery
Operate a browser Browser Harness Authorized session, target page and permitted actions
Check and deliver code CI/CD and Automation Quality gates, deployment target and recovery

Specify the loop before scheduling it

Marketing Loops describes recurring agent workflows with cadence, inputs, outputs and review boundaries. The upstream task is useful when a team repeats a process such as gathering content ideas or reviewing a campaign. Write the trigger and expected artifact first: for example, a weekly draft containing source links and unresolved questions, reviewed by a person before distribution.

Illustration connecting inputs, cadence and review in a recurring marketing loop
SkillHey illustration from its Marketing Loops listing; it illustrates a process rather than a measured outcome.

Decide what happens if an input is missing or the same item appears twice. Record the run time, sources, output location and error, so a reviewer can distinguish a quiet week from a failed run. Specify who can alter the schedule and what action requires approval. A Skill can help design the process; execution and scheduling depend on the actual tools you connect.

Monitor a research area

Nature Literature Pipeline describes a scheduled literature workflow covering discovery, configurable scoring, reading, delivery and local archival. Read its upstream instructions and keep the complete required directory and dependencies. Configure the sources, access and delivery mechanism for your environment.

Start with a narrow search, a documented time window and a destination for results. Check the search log and a sample of included and excluded papers before relying on a recurring report. Scoring can help prioritize reading; it does not establish that a paper is correct or relevant to your decision. The research guide covers question framing and evidence checks in more detail.

Work through an authorized browser session

Browser Harness connects an agent to a real browser through the Chrome DevTools Protocol. The project describes using a browser for interactive and JavaScript-rendered tasks. The connected session and any existing login determine what the agent can access, so use an account and environment appropriate to the task.

Give the agent a bounded objective: identify the page, the data to collect and whether it may submit forms. For an initial run, request a draft or extracted preview and inspect it. Websites change, and a selector or navigation step can fail. Keep a record of failed pages and review a sample of extracted fields before using the result downstream.

Use explicit gates for code delivery

CI/CD and Automation helps plan pipelines with quality gates, failure feedback and deployment or rollback considerations. Its upstream entry is aimed at software delivery rather than a general business workflow. Supply the repository's current commands, target environment and deployment constraints. Check proposed changes against the actual CI provider and project documentation.

An example sequence is a test run, build, review of the resulting artifact and deployment to a chosen environment. Decide what happens when a gate fails and how the last working release can be restored. A generated pipeline configuration should be reviewed and exercised in an appropriate environment before it is trusted for production.

Brief an automation task

Trigger and frequency: [event or schedule]
Input and access: [sources, authorized accounts and time range]
Steps: [what to read, transform and produce]
Output: [destination, format and owner]
Review boundary: [what stays a draft and who approves it]
Failure: [how to report missing input, duplicates and errors]

Start with one representative run and inspect its result. Once the workflow and failure handling are understood, choose the scheduler or integration that fits your environment. Browse the Automation topic for related published Skills.

Browse all Automation AI Skills →