Custom Claude Skills
Most of my work is building things with AI. This is the layer underneath that: a set of skills I wrote to make working with AI itself more reliable. They handle memory, continuity, and chaining work together, so the assistant gets better at the job instead of starting from scratch every time.
The idea
When you work with an AI assistant every day, the same friction shows up over and over. It forgets what happened last session. Patterns across hundreds of conversations stay invisible because no one can read all of them. Useful workflows get rebuilt by hand each time instead of being saved and reused. None of that is a model problem. It is a tooling problem.
So I built skills to fix it. A skill is a reusable instruction set the assistant can load on demand, with a clear job and a defined output. These three are deliberately meta. They are not about a single task like writing copy or cleaning data. They are about making the assistant a better collaborator across all of that work.
Conversation Archaeologist
This one mines all past conversations to build a comprehensive profile of how someone actually works. Where a single session only shows you today, this looks across hundreds of sessions for the patterns you cannot see yourself: what works, what does not, where time gets sunk, and what is worth automating.
It reads the conversation history and analyzes it through a structured framework covering professional context, recurring projects, success and failure patterns, lessons learned, decision-making, communication style, tools, and goals. The headline output is a ranked list of automation opportunities, each scored by frequency, time cost, and rough return, so the highest-leverage things to fix float to the top. It writes the whole thing out as a user manual you can keep and share.
Why it matters: it turns a pile of scattered history into a real self-audit. Instead of guessing what to automate next, you get an evidence-based answer grounded in your own work.
Handover
This is a shift-change report for the next session. At the end of a long working session it reviews the entire conversation and writes a HANDOVER.md that captures what got done, what worked, what did not, the key decisions and the reasoning behind them, the current state (what is working, what is incomplete, what is broken), and a clear, prioritized list of next steps.
It also maps the important files touched and flags repetitive manual work it noticed along the way, with specific automation suggestions. The point is continuity. When you come back tomorrow, or a different session picks up the thread, nothing gets lost and no one has to reconstruct context from memory.
Why it matters: the gap between sessions is where context dies. A good handover is the difference between picking up exactly where you left off and starting half-blind. The honest record of failed attempts and dead ends is often the most valuable part, because it stops the next session from repeating them.
Skill Composer Studio
The first two skills capture and remember. This one builds. It chains multiple existing skills into a single multi-step workflow, with the output of each step handed off automatically as the input to the next. It supports conditional logic, so a workflow can branch based on what an earlier step produced, and it defines the handoff points, data transformations, and error cases as one composite skill you can run again.
So if a job is really research, then analysis, then a written report, you do not run three things by hand and copy results between them. You describe the workflow once and it stitches the pieces together into a repeatable unit.
Why it matters: it treats skills as building blocks instead of one-offs. That is how you go from doing the same multi-step process manually every time to having a programmable pipeline you trust.
Why this set, together
On their own each skill is useful. Together they form a loop. Archaeologist looks across everything and tells you what is worth automating. Handover keeps each session's context from leaking away. Composer turns the patterns you find into repeatable workflows. Capture, remember, build.
For an AI Implementation Engineer that is the real signal. The job is not just prompting a model well. It is designing the scaffolding around the model so the work compounds: memory, continuity, and reusable workflows. I build that scaffolding for myself, which is exactly the muscle I would bring to building it for a team.