Everyone’s favorite task: updating their resume. My calendar reminds me every six months to get this task done, and about a year ago I started to use LLMs to help make the process less painful and more consistent. As of this writing (mid-November 2026) I’m finding myself making a few changes here and there, and I want to share my process with you.
In this post I explain in detail my 7-step process for refreshing my resume (and LinkedIn) from six months ago as well as changes I made to the process itself. Takeaways:
- Learn how to interact with LLM skills to keep your resume sharp.
- Use the interactive turns with the LLM to learn how to write a better resume.
- Understand how to leverage the LLM to self-improve the existing skills.
Note: This is a long-form article. I chose not to split it into multiple posts so the entire process could be captured in one artifact, making it easier to use as a reference.
Caveats about AI
I consider myself a pragmatist when it comes to LLMs, coding harnesses, and the like — meaning I reach for those tools when they are helpful rather than because it’s fast, requires less thought, or to be on the bandwagon. Being in the driver’s seat is important to me. Some of you seasoned LLM pros will find 47 different ways I could super-optimize the process I describe below. That’s fine. Friction is important to me. I want to feel the edits real-time and make sure I maintain editorial control. I’m sure you can find ways to “one-shot” your resume — this ain’t it.
This post was written entirely by me (yes, even the em-dashes). I’m comfortable using an LLM to write my resume for two reasons. First, I’m not a resume expert and need guidance about understanding and applying best practices. Second, AI is likely the first “reader” my resume will encounter, so I chose to make it easy for AI to consume.
Tools
VS Code helps me manage the input, intermediate, and output markdown files produced by the skills. Having the markdown preview is handy, and I run the harness in a terminal window.
Claude Code is the harness for triggering the skills and running the interactive session.
- Settings: Opus 5.5, medium effort
- I have a Pro subscription, but most of the work is lightweight enough such that you could likely get by on the free version.
My resume skills.
Step 1: define the resume objective
Each of the sections should be internally consistent, and I’ve found getting clear about what I want the reader to take away is a solid start.
The initial version of the define-resume-objective skill was a riff on Matt Pocock’s “grill-me” skill: Ask me questions one at a time until we come to an understanding. The skill tells Claude to act as a resume expert in my field, and to ask me for my current resume summary paragraph and my About Me section on LinkedIn. (These were given in markdown format in a working-files subfolder.) This helps it understand how I previously presented myself professionally so we can expand, refine, or pivot.
Here’s a summary of the turns:
- Claude recognized I wanted to stay broad: individual contributor, tech leadership, enablement, coaching. It asked me for a concrete example of how I adjusted my leadership style to fit the situation with some examples so it could understand specifics.
- I responded that my company is a consultancy that meets clients where they are. Influence comes from me being an expert, a coach, and someone that builds trust with small wins.
- Claude asked if I wanted to highlight a tech stack, practices, etc. and to pick 3-5 areas I want to be known for.
- I mentioned I’m strongest in .NET/Microsoft, but wanted to pivot my objective to focus on operational excellence — things like Lean thinking, systems thinking, theory of constraints, etc. — given most engineers I’ve seen aren’t interested in these despite them being pivotal to understanding how to effect change.
- Claude gave me three versions of an objective: systems-focused, short and punchy, and impact-first. Each of these were written as if I’m currently looking for a job, which isn’t true (and it didn’t know), so I told it why I’m updating my resume (e.g., some clients want to see it, I never know if I’ll be laid off). Claude then explained that “seeking a position as X where I can Y” is common, but inappropriate for my purpose (i.e., clients understanding what I’m about, showing how I create impact, and incremental updates that aren’t tied to landing a specific job).
- Claude reworked the three versions, and also dropped first person pronouns which I was fine with.
- I had Claude combine short and punchy with impact-first and said why: My job is to empower teams to help themselves (outlasting the engagement), and our company is seeing more clients want impact-driven work rather than T&M.
- The new version needed work: It pigeon-holed me within just Microsoft tech and claimed that my goal was to help teams work “faster and safer.” My experience section makes it obvious I work with Microsoft tech, so that won’t be missed by the reader later on. Claude said the faster/safer bit was common in DevOps/DORA wording, but admitted it clashed with my “pragmatic over prescriptive” stance. Teams have other goals and I want to help them with those first.
- Finally, Claude gave me four options for a lead sentence and a recommended choice. I agreed with its recommendation.
- Claude finished by summarizing what we accomplished.
With the session complete and resume-objective.md in hand, I read it out loud to myself. I’m a huge fan of this multi-modal approach where you see and hear the words to see if it’s accurate and feels natural. I asked myself, “Is this something I believe is a genuine overview I’d say aloud as I’m introducing myself to another professional?”
Being a process person, I completed the PDCA loop. (PDCA = Plan, Do, Check, Act. I gave Claude inputs and a plan via the skill definition, interacted with the harness, and checked intermediate and final outputs. Act is the continuous improvement step.) I asked Claude to inspect our session and suggest improvements to refine the skill.
- It listed three times I found flaws in the process, why they happened (root causes), and what would prevent them next time.
- Without being asked (LLMs love to work ahead!) it realized from the README that outputs from this first skill feed into all the other ones. Some minor tweaks saved me from having to specify where those subsequent skills could find what they needed.
- It also called out several aspects that were working well and should be left alone.
What ultimately changed was specifying more things upfront:
- This objective is more of a value statement, and some other aspects should be assumed (i.e., not looking for a job now, no first person, stack-neutral, no assumed outcomes, pragmatic first, must-haves for the statement, requiring approval if AI adds something I didn’t ask for).
- It made the workflow sequence explicit: look at prior work, clarify audience early, gather specifics when the input is too general, give 2-3 options with rationales.
- It specified the output structure (this is the technically accurate term for one-shot prompting): purpose, roles I’m suited for, audience, key differentiators, leadership examples, domains/industries, tech stack, any decisions that could impact other resume sections.
Step 2: collect and integrate new accomplishments
This skill starts with my objective from the previous step and integrates information about what I’ve been up to for the past six months. Before we get into the turns, I’ll explain how the input (other than the objective from Step 1) works.
- The experience section on my resume is chunked by employer. Although I only need the bits about my current employer, I’ll need other parts of the resume later, so I run the entire Word doc through word2md. I copy/paste my current employer section into its own markdown file
current-role-experience.md. - I keep a brag sheet in Google Docs for each company I’ve worked for; my current one is called Geoff’s Accomplishments at Lean TECHniques. This doc is a comprehensive list of anything professionally relevant I’ve ever done at this job, broken down chronologically by month. I update this weekly. Example things that make it onto the list:
- Books, training courses, certificates completed
- Internal, client, or community events attended
- Ways I helped my company (e.g., running book clubs, refining interview processes, mentoring other engineers)
- Ways I helped my clients (e.g., running workshops, pairing, learning their domain, shipping tickets, building relationships)
- Feedback and praise from peers, clients
The brag sheet is a superset of all things that prove I’m a decent professional. Only a fraction of these end up on the resume, but the whole picture gives Claude something to consider and weave into a cohesive story based on my objective. (I highly recommend having a brag sheet. It makes self-reviews much easier and it gives you a wealth of things to assemble into a promotion argument when the time comes.)
My case is bit different because I work at a consultancy and have been placed at multiple clients. My brag sheet for last month has one parent bullet Client “Financial Services Company” ({real name here}) and a second parent bullet Lean TECHniques internal projects.
(Note: In the last six months, I finished with one client and started at another. My May resume didn’t include a section for internal work, and the last six months had several things in that bucket I wanted to include.)
Here’s a summary of the turns:
- Claude read the objective, role experience, and peer feedback markdown files. It checked my full resume to see where I put certs and continuous learning so it wouldn’t add those in my experience section. Claude pulled out the new bits from my previous and current clients and figured out what was already covered by existing bullets. It highlighted things it wanted to dig into given my focus on operational excellence. Lastly it successfully found stuff that was too insignificant to include on a resume (e.g., compliance training, one-off local events).
- It asked how to handle internal firm contributions: treat it like another client sub-block, work it into existing clients by role title, or put them somewhere else entirely. I chose a new sub-block.
- One of the reasons I initially chose to use LLMs to help with my resume is that I feel compelled to include everything and need help trimming things out. The skill said to limit subsections to 6-8 bullet points. Claude gave me three choices: use more bullets for recent work and fewer as we go into the past, limit my entire employer experiences to 6-8 total bullets, or use 6-8 for every client. The first one has a good balance.
- Claude wrote
current-role-experience-nov2026.mdwith seven bullets for my current gig, five for my previous gig, three for internal contributions, and two to my first-ever gig. It also explained why it kept what it did (e.g., meets teams where they are, measured time savings, built capability that outlasts, taught testing skills to support AI-enabled development). It also explained why it cut what it did, and was on the fence about one item I was praised highly for. - Next it asked me what came of the process mapping exercise I did (e.g., found bottlenecks, surfaced expensive handoffs, removed steps, got the team to commit to an improvement). I like this aspect of having an LLM partner; it finds gaps and asks you to fill them.
- The next question was about a deployment fix I put in place. I ended up cutting it because although it demonstrated technical competency, it was a one-off experience with little impact. I told Claude to dig into a more lasting change where I helped a data engineering team stand up a .NET REST API in Azure.
- On that topic, Claude started pulling the thread. Did I build it myself? From scratch or from a template? Did I help the team assume ownership? Did it ever ship?
- Claude got confused about why the type of work shifted last month and what team names meant in my brag sheet. We ended up wordsmithing this into showcasing how I’m an embedded engineer on 10-12 week rotations.
- At this point I forgot that I hadn’t instructed Claude to anonymize client and team names. (Not a big risk because it’s more of a faux pas, plus the resume content thus far only existed on my local machine.) To this end, it pointed out something else that could be anonymized: the type of firm our interns worked with. After some further questions, I dropped the type of firm because it didn’t help me sell my resume any more effectively.
I copied/pasted the updated experience for my current company into my resume Word doc and read each bullet aloud, making adjustments where statements felt wordy or awkward.
As before, I prompted Claude to consider the current session and look for improvements to the skill as written.
- State how to handle multiple clients and firm work (i.e., order to list them, where internal work goes, how many bullets each client gets).
- Add an anonymization rule.
- Read known files before asking for them. (The skill definition said “ask me for” various inputs, when instead Claude could find those itself.)
- Confirm engagement facts in one opening question. (Claude couldn’t determine from my notes that I left a client or switched teams within my current client.)
- Send non-experience items elsewhere and list them. (I have certs, skills, and other learning items blended with accomplishments, so I can tell Claude where those go on the resume.)
- Put the riskiest decisions first using the following order: structure, bullet budget, engagement facts, impact for bullets that matter most, anonymization review, final approval.
- Ask for what’s significant, not only what can be measured. (Claude doesn’t know the significance of items on my brag sheet.)
- Reuse the “wording discipline” rules from the
define-resume-objectiveskill. (For example, no first person, don’t infer or assume benefits without checking with me first.) - Pin down the output (i.e., don’t edit the input file, list what bullets were dropped and why, only write files locally).
- Say what peer feedback is for (e.g., supporting or ranking bullets, what goes on a resume vs LinkedIn).
Step 3: update experience sections
The update-experience-sections skill is one of my favorites because it helps me kill my darlings (i.e., keeps my resume from getting too long). The remaining bullets under employers in the Experience section are analyzed to see how they can be rewritten more concisely while limiting the number of bullets based on recent, mid-career, and older roles. Again, bullets for the chopping block are the ones that don’t align with my objective.
Here’s a summary of the turns after giving Claude my objective and full resume in markdown format:
- First, Claude showed a bullet tapering plan by employer.
- Next it showed me a draft experience section for my 2021-2023 job and why it changed things. Similar bullets were merged, one was promoted because it was one of my differentiators, another was reworded to use Lean terminology, and another was dropped because my current job showcased it better. I disagreed with how aggressively it trimmed a leadership role down to six bullets, so I asked Claude to increase that role to eight instead.
- For my 2019-2021 job, it merged elements from ten bullets into six without dropping leadership and other elements from my objective. One of the bullets for flow mapping got a promotion, and some duplicated terms were replaced. It also noticed this was the only job where my bullets didn’t state what kind of industry this was; recall that my summary statement mentions specific regulated industries, and this company is not a household name. A small wording update was worked into the first bullet to reflect that.
- My 2015-2019 job had seven bullets merged into five such that leadership, mentoring, and process maturity topics were grouped together rather than separate line items. It also wordsmithed “breaking major features into smaller pieces” as “breaking down major features” which sounds more like how I’d say it today.
- My oldest job was distilled from eight bullets to four. Claude reworded some of the bullets to be consistent with my objective. Because this job had the most bullets eliminated, it told me why certain ones didn’t make the cut and also asked me if any should be put back in. It dropped one important bullet about how I did onsite support and interacted with end-users, which was not super common for software engineers.
The bulk of the update work is done at this point, leaving my resume at four full pages. I’ve read resumes for people with longer careers than me that fit into two pages, but I’m satisfied with my 18-year career distilled into four, especially given I haven’t followed a typical career progression.
Claude’s suggestions for next time:
- Read the resume as one file rather than having separate files for each section. The markdown is easy enough for Claude to figure out what bullets go with what job.
- Base bullet counts on how broad the role was, not just how long ago it was. Allow more bullets for leadership roles or for roles with evidence that appears nowhere else.
- Make sure each domain (e.g., healthcare, finance, government) in the objective appears somewhere in a bullet.
- List what gets dropped from each role before asking for approval. Flag unusual differentiators (e.g., field work, customer-facing tasks) as worth keeping rather than cutting for length.
- Define the output format more precisely and save after reviewing each job.
- Sharpen the description to state that the skill is for rewriting prior-role experience to align with the resume objective.
- Add a check for client name anonymization.
Step 4: update skills section
Step 2 accumulated any new skills I’ve utilized over the last six months, and this step is about combining what I already have from my resume with the new ones, then distilling them down into something that fits my narrative.
Here’s a summary of the turns:
- Claude picked up my new focus of “operational excellence” and how that might change the presentation. It recognized the new skills from the last step, and highlighted questions where there are new ambiguities.
- The first question was about what to do about the “Governance & Compliance” section. My summary mentions one of my differentiators (working in regulated industries), so maybe that’s worth keeping. I also have some older compliance frameworks from 2019 that I haven’t kept up with. Options: leave as is, broaden it, drop it. After asking for guidance, it suggested eliminating the section and moving the relevant frameworks into Engineering Practices skills. The older frameworks are already on my resume in bullets under Experience.
- The second suggestion was akin to leaving off table stakes business tools such as operating systems and Microsoft Office from the resume. This time it’s Miro, Confluence, and Jira.
- The third question asked about my level of comfort with non-Microsoft frameworks given my bullet points show I’ve used them; Claude used the phrase “they show range without suggesting depth,” which is accurate. Likewise, AWS dropped out of the Cloud & DevOps section given I never did much with it, and I’m doing far more Azure at my current client.
- The next choice was about how to order the skill categories. Claude provided a suggestion and rationalized its decision. It chose AI-Assisted Development, Testing & Quality, and Operational Excellence as the top spots because my objective statement lists those as my specialties. The next two sections — Engineering Practices and Coaching & Leadership — are applicable at pretty much any company, and the tech-specific sections — Languages & Frameworks and Cloud & DevOps finished out the section.
- The last few turns were about how to strengthen the testing skills, because all I had listed was TDD and a few frameworks. Since my last resume update, I’ve done more integration testing and forgot that I did approvals/characterization testing at my previous client. Although I have conceptual knowledge of (and opinions about) ATDD or spec-driven design, they’re not strong enough to list on the resume yet.
With the new skills section written to markdown, I reviewed it and copy/pasted it into the Word doc.
Here are suggestions Claude made for improving the update-skills-section skill:
- Try to find the inputs in the
working-filesfolder rather than asking the user for them. This also takes advantage of a suggestion it made in Step 2 where the output includes a section about skills to consider for this step. - Check skill depth to see what makes the front page. Single bullet point? Not used in the last 5 years? Something I’d be uncomfortable pairing on?
- Ask about practices, not just tools.
- Base categories and their order on the resume objective. Put tech stack items last to avoid typecasting myself.
- Add rules for small or redundant categories (e.g., fewer than two skills, table stakes tools, items that go in other resume sections).
- Define the output format sections (another one-shot example).
Step 5: update tagline and summary
Here’s where Claude can take advantage of all the intermediate markdown files it’s been consuming and creating. It’s able to compare what my previous wording was and apply the new focus from the updated resume.
Here’s a summary of the turns:
- Claude picked up my old and new summaries as well as the tagline for my resume (key phrases separated by pipe characters). Claude drafted a tagline and gave evidence for its choices. Then it realized that I probably wanted a short label rather than a sentence fragment.
- I asked for something that’s “easy to scan” and it provided four versions and what stood out about each one. It also recognized the headline on LinkedIn is often truncated to 60 characters.
- The first listed my three differentiators and my role. The second led with “hands-on” and used the broad term “operational excellence.” The third focused on engineering enablement. The fourth was the shortest and interestingly dropped AI so that “it won’t age as quickly”; plus my summary mentions my AI skills.
- It suggested the first one because recruiters would more easily search for and recognize Lean thinking rather than operational excellence.
- It also realized “responsible AI adoption” might lead readers to think about ethics or governance, not AI-ready engineering practices.
- At this point none of those four options looked great, so I told it more about what my client work is typically like especially around engineering maturity and agentic approaches. This yielded four more options. It was keen on keeping “hands-on” for each headline because I write code alongside teams rather than just coaching/advising. Apparently “AI-assisted” is on the way out in favor of “agentic development.”
- I chose the headline that included engineering enablement, Lean and systems thinking, and building foundations for agentic development. (I did double-check how Claude defined “engineering enablement” to ensure both of us were using the term correctly.)
- The summary option it showed me was “enablement first, consultative.” I asked for one more option in a different tone, so Claude generated one that was “warm and people-focused.” Taking advantage of my peer feedback, it said the warmer version is closer to how people describe me, and the other option is more authoritative which could land better with Staff/Principal hiring managers.
- The last option sounded best but seemed overly wordy, so I asked Claude to tighten it up while keeping the warm tone.
The tagline and summary are complete, which wraps up most of the resume.
Here are suggestions Claude made for improving the update-tagline-and-summary skill:
- Try to find the inputs in the
working-filesfolder rather than asking the user for them. - Specify the tagline format. Claude gave me a sentence fragment when what I wanted was pipe-delimited keywords.
- Ask how you see your role before writing any options. This came from me having to correct expectations about what my work entails.
- Do the tagline and summary in separate steps, each with its own way of presenting options. What worked was nailing down the tagline first so the summary could be driven from that.
- Set a target length for the summary.
- Point out terms that readers might take differently. This came from “responsible AI adoption” not being chosen and me asking Claude what it thought engineering enablement meant.
- Make the output file follow the same conventions as other skills.
- (Optional) Use your peer feedback to set the tone. I chose to skip this change because I prefer seeing 2-3 options with different tones.
Step 6: verify resume format
I generated a markdown version of the updated resume and then ran the verify-resume-format skill, which looks for content and format issues that would prevent it from being accepted by an applicant tracking system (ATS) or make it difficult to read. This skill is also a means for the LLM to check its own work, and it’s important to have a fresh context window so the LLM can focus on this task without prior writing/editing knowledge.
Claude generated a 194-line markdown review, which I will not explore in detail here; the majority of the content was essentially a to-do list for me. It is worth mentioning that my skill definition was very open-ended on the analysis format, so it was interesting to see what Claude came up with. Here are the sections:
- Overall assessment
- Summary of changes compared to the previous version
- Structure and layout (and what needs review/editing)
- Style and formatting (and what’s inconsistent)
- Content and strategy (with 7 “opportunities” to consider)
- Experience section: roles, durations, companies
- PDF publication considerations (length, font, margins, etc.)
- Prioritized fix list by impact on interview/engagement conversion
There were several changes I ignored because I disagreed with Claude’s rationale. Here are the changes I did make:
- Moved Education last given how long ago I completed my degrees
- Moved Certifications to the first page under Skills
- Condensed some experience bullets and fixed several wording inconsistencies
- Adopted a new format for individual clients under my current consultancy job
- Mentioned promotions within a company separately from the experience bullets
The resume is now complete. Here are suggestions Claude made for improving the verify-resume-format skill:
- Add an Inputs section, which includes the resume to review, the previous full resume (for comparison), and the resume objective.
- Ask about constraints (e.g., overall length) before starting the analysis.
- Say when to ask questions (e.g., before the analysis, after it) rather than the vague “ask questions one at a time” prompt.
- Name the checks that found the most useful issues (e.g., skills having supporting experience, anonymization of clients, finding contradicting terms, summary of changes since the last version).
- Make the output more specific and structured.
- Add an optional recheck to prevent having to redo a full analysis.
Step 7: update LinkedIn content
I copied my LinkedIn About and Skills content into a new markdown file and ran the update-linkedin-content skill.
Here’s a summary of the turns:
- After listing anti-patterns for headlines (e.g., emojis, title + employer), Claude gave me three headline versions: same as my resume’s tag line, role noun first, domains at the end.
- I mentioned that future clients will likely look me up on LinkedIn and may not receive a resume first. This changed Claude’s focus: “What matters more is what they understand in the first few seconds,” to answer “who is this and what will they do with my team” and “does this person work with us or judge us from the sidelines.” This also changes how the About section is written. I opted for “Software Consultant & Technical Coach” as the first section of my headline.
- Next it contrasted my prior About wording with my new resume focus. Some LinkedIn specific recommendations:
- Note that readers typically only see 200-275 characters before the “see more” link.
- Write in first person.
- Use short paragraphs.
- Aim for 1,000-1,500 characters.
- Avoid pasting your resume summary as-is.
- Don’t be generic.
- Avoid AI hype and long lists of tools/keywords.
- Claude presented three options: refreshed version of my current wording, what to expect when working with me, opening with a point of view. The point of view option was a great start because it’s distinctive and doesn’t sound like I’m hedging as to not offend readers. My only issue was that it read like my main skill is helping teams adopt AI, so I asked Claude to present a few different versions with that tone.
- Lastly, I had it suggest my top five skills. Again Claude provided advice to stay consistent as well as anti-patterns to avoid based on what’s important to me and how readers will find me. The three options: mirror the headline, practitioner-first, or flow & foundations. I mixed and matched from the three to align with what I typically help clients with.
Here are suggestions Claude made for improving the update-linkedin-content skill:
- Find inputs in the
working-files/folder before asking. - Ask about the audience before drafting.
- Add Top Skills as a third section.
- Ask about tone, and check that the opening isn’t about just one topic.
- Explicitly list LinkedIn content display constraints.
- Run a terminology and conflict check.
- Define an explicit output format.
Conclusion
This instance of resume update certainly wasn’t the most expedient, but I gained far more than updated words on a page. (Updating my resume, LinkedIn, and writing/editing this blog post took about 16 hours.) The skills are much improved from their May 2026 state, I did some deep thinking about my career and how I appear to others, and I have a better grounding of what strong resumes look like. Some other observations:
- Claude knew about other skills and the order in which I suggested running them from the README. It was typical at the end of a current skill run for it to leave itself notes (i.e., persisted memory) for the next skill and ask whether I wanted to start working on it. The first part (leaving itself notes) was useful, but it was important to me that I clear the context between skills because the tasks were different.
- When there were decisions, I chose to be in the driver’s seat even though Claude had suggestions of which choice was best.
- Having Claude evaluate each session was pivotal in finding ways to make the skill get closer to achieving the goal I want sooner.
- Keeping each of the seven skills separate still seems like the right call because they build on one another sequentially (i.e., have dependencies). One course correction on Step 2 could have ripple effects.
My hope is that these skills can help you keep your resume current and learn things along the way. Feel free to check out the fruits of this exercise: resume, LinkedIn.