Blog

AI Skills and Agents

hazel@head2bioconsulting.com (Dr Hazel Pinheiro) — Sun, 04 Oct 2026 16:14:26 GMT

Skills and agents: turning a repeated task into a tool

Every marketing team has tasks that get done the same way again and again. Proofread this copy. Check this claim against its source. Turn this brief into an outline. If you explain the same steps to an AI every time, you're doing work a skill can do for you.

A skill is one workflow. You describe the steps once, in a file, and call it with /skills whenever you need it. It runs inside your current session, so it has your project context. My rule: one skill, one workflow. A skill that does five things is hard to test and hard to fix.

A practical note for anyone who builds skills in Claude Cowork: when you download one, the associated files come with it, even though it doesn't look like they will.

An agent groups related skills. You call agents with /agents, and you describe the process the agent should follow in an .md file. Take a scientific content writer as an example. It might draft the piece, check each claim against the source, proofread, and format the result. Those are separate skills. The agent is the process that ties them together and decides the order.

Try it in planning mode first. Before an agent acts, planning mode shows you what it intends to do. That's where you catch a misread instruction, before it has written 3,000 words.

My rule of thumb for choosing between a skill and an agent:

A single, repeatable task: make it a skill.
A multi-step job that combines several skills and needs its own process: make it an agent.

Where the files live matters too, and I'll cover that in the configuration issue. In short: skills and agents you want in every project sit in your global folders, and ones that only make sense for one project sit inside that project.

Three tips for building AI skills and agents:

1Start from a task you already do well by hand. You can't describe a process you haven't got.
2Write the steps the way you'd brief a new colleague.
Test on real work and fix the skill and agent files, not the output.

Once the skills and agents are defined, they are consistently applied every time.

If you'd like help building skills and agents that fit your team's workflows, get in touch.

Next issue: how to brief an agent like a user story, with a clear definition of done.

Which repeated task in your team would you turn into a skill first?

Previously in this series:

→ Issue 2: Managing the context budget

→ Issue 1: Context is king

Managing the context budget

hazel@head2bioconsulting.com (Dr Hazel Pinheiro) — Sun, 04 Oct 2026 13:16:11 GMT

Context is a budget: five habits that keep AI output sharp

Claude uses the word "context" for two different things, and that can be confusing.

Project context is what you give Claude: the project files and background information covered in my last blog.
Session context is the working memory Claude actually has in one particular session.

When a session's working memory fills up, the quality of the output drops. The two are related. The more files Claude has to read to action a task, the faster the session fills. Everything counts against it: your messages, the files it reads, its own answers. Claude Code can compact a session automatically (autocompact), but I don't rely on it to decide which parts of my project matter. I decide.

If you use Claude Code, you can keep an eye on how much working memory you've used, and make sure key learnings from each session are committed to memory so they aren't lost for future work.

These are the five habits I use.

Check the meter. Type /context to see how much you've used. Look before a big task, not after things go wrong. There's also a way to have a visual circular red dial show up when you pass 50%, so you don't have to remember to look.
Summarise before you compact. Whenever the context gets above about 50%, I ask Claude to update Progress.md so any key learnings are committed to that file. The important parts then live in a file that survives, not only in a chat that gets compressed.
Compact at 50%. That's my rule of thumb. It isn't a law, so watch how your own sessions behave and adjust.
Start a new session with a handover prompt. Before I start the new session, I ask the existing one to write a structured handover prompt to continue the work in a new session. I paste that whole prompt into the new session and start again with a fresh window of context and fresh working memory. That way, the new session is fully up to speed with what has already been done, what has been learned, and what needs to be done next.
Send deep research to a subagent. A subagent has its own context (working memory). Using subagents to conduct literature searches, draft web pages or research particular topics is a good idea whenever you're primarily interested in the outcome of that research - because the heavy reading doesn't fill up the context of your main session. You just get the result.

Two commands are worth knowing alongside these:

/clear wipes the current session's context, so a new task starts clean.
/resume takes you back to an earlier session with its context, when you need it again.

Why does this matter for life science marketing work?

Because the tasks are long. A messaging framework, a set of web pages or a claims review can run across many steps. If the AI has quietly lost the rules you set at the start, you get an inconsistent result, or a lower-quality one, and the inconsistency is hard to spot.

The same goes for what it learned along the way. If that information drops out of the session's working memory, the quality of the output drops with it. And if the learnings were never written to a file, future sessions don't benefit from what this one worked out.

The habits above take a few minutes to learn. After that they cost almost nothing.

If you'd like help building this into your team's workflow, get in touch.

Next issue: skills and agents, and how to package a repeatable workflow.

Have you noticed AI output getting worse late in a long session?

Previously in this series:

Context is king

Context is king

hazel@head2bioconsulting.com (Dr Hazel Pinheiro) — Sun, 04 Oct 2026 12:55:16 GMT

Context is king - and how to keep your projects separate.

Over the last six months I've been learning how to apply AI tools to websites, with life science marketing in mind. I'm starting this blog to share some of my learnings.

I'm starting with context, because everything else in this series depends on it.

Claude Code knows nothing about your technology until you tell it, and it won't say so. It will write a fluent paragraph anyway. In life science marketing, the details that make a paragraph usable are the ones it can't guess: the approved claim, the intended use, the audience, what the regulator allows.

Everything you give it before it starts working is its context. With too little, it fills the gaps with guesses. With too much, the important parts get buried and you pay for the extra in tokens. I aim for just enough, chosen on purpose, with pointers to specific files instead of everything pasted in.

If you work on several projects at once, there's a second problem: keeping them apart. Information about one product can be irrelevant for another. Context from one project should not leak into the next.

Here's the structure I use. Each project gets its own folder and its own workspace, with four files:

CLAUDE.md: operating guidance for the AI and rules for this project. I keep it under 200 lines, and most of it points to the other files.
Instructions.md: the project scope, constraints, people involved, what needs to be accomplished, and a folder map of all the relevant project files so that the AI can easily locate the files it needs.
Project context.md: background on the project and relevant technologies.
Progress.md: what's been done and what comes next. I ask Claude to update it regularly.

CLAUDE.md points to the other three, so every session starts from the same background. Because these are plain text files, I can read them, correct them and keep them tidy myself. When the AI gets something wrong, I fix the file, not the chat.

Getting started is quick. Claude Code has an /init command that creates a first version of CLAUDE.md. If you already use Claude Chat or Cowork, you can export its memory and start from that. Either way, expect to edit it.

Two tips:

1. Start smaller than you think. Three accurate paragraphs beat a 40-page briefing. Add a line each time you catch yourself repeating an instruction.

2. Keep one workspace per project. The AI can only see the files in that workspace, so it can't quote from a project it can't see. That's the point.

If you'd like help building this into your team's workflow, get in touch.

Next issue: what happens when the context fills up, and how to manage it.

Which detail about your technology, product or company do you find yourself re-explaining to AI most often?