Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Turn Your Agent Into Your Own Nemesis

Turn Your Agent Into Your Own Nemesis

Workshop held at https://agentic.hamburg
Hamburg 2026

Most agents are built to be generally helpful. That's not the same as being useful for you.

I've found that I work best when collaborating with people who are complementary in their strengths.

In this workshop, we apply that idea to agents.

We configure an existing agent (Claude Code or Gemini CLI) into a personal work companion: one that reflects how you actually work, strengthens the areas you don't naturally prioritize, and knows when to slow down and ask the right questions instead of just producing output.

We start with a short, focused self-assessment to map your working patterns. Then we translate that into a working configuration, step by step: default behavior, task-specific modes, and clear rules for how the agent should behave when things are unclear or misaligned.

You'll leave with a functioning Nemesis and a calibration process for your agent.

Laptop and access to a live agent required. No coding experience needed.

Prerequisites:

​1. Cluade Pro / Gemini
2. IDE (e.g. VS Code or Zed) or Claude Cowork

Repository: https://github.com/terezaif/nemesis

Avatar for Tereza Iofciu

Tereza Iofciu

March 22, 2026
Tweet

More Decks by Tereza Iofciu

Other Decks in Technology

Transcript

  1. TURN YOUR AGENT INTO YOUR OWN NEMESIS Workshop · Agentic

    Conf Hamburg · March 2025 Tereza Iofciu
  2. WHY NEMESIS? Your nemesis in mythology The force that keeps

    you honest. The counter to your excess. In work The person who sees what you miss. Not an opponent — a complement. Your agent can be that Configured to cover your gaps, not just execute your asks.
  3. I WORK BEST WITH PEOPLE WHO COMPLEMENT ME. Low on

    structure → partner who plans ahead Fast decision-maker → partner who questions assumptions Broad thinker → partner who narrows scope Same logic. Different partner.
  4. MY NEMESIS IS... ...my complement. High scope → narrows focus

    Quick decisions → questions first Strong starter → helps you finish
  5. WHAT IF YOUR AGENT COULD DO THE SAME? Not a

    generic assistant. Your complement. Configured for how you actually work & your needs. This is your Nemesis.
  6. TODAY 01 The 4 building blocks 02 Get the starter

    config 5 min 03 Fill the questionnaire 5 min 04 Load your profile 10 min 05 Add a watchout + pick trigger 10 min 06 Generate your writing style 5 min 07 Share out
  7. THE 4 BUILDING BLOCKS YOUR AGENT CLAUDE.md Set once, rarely

    changed operator_profile.ya ml Updated as you evolve modes.md Add modes over time drift_log.md After every bad session You write the rules. The agent reads them on every session.
  8. 01 CLAUDE.md The rules. What your agent always does, no

    matter what. W H Y I T M A T T E R S Without it, the agent resets to defaults every session. This is what makes it yours.
  9. CLAUDE.md ## Alignment Gate Before any detailed output, you MUST:

    1. Restate the goal in one sentence. 2. Ask 2–4 clarifying questions. 3. Propose a direction. Then STOP. Wait for exactly: Engage
  10. 02 operator_profile.yaml Who you are. How you work. Where your

    gaps are. W H Y I T M A T T E R S The agent can't complement you if it doesn't know how you work.
  11. operator_profile.yaml complement_axes: structure_bias: "low" decision_style: "fast" starter_block: true impatient_reflector: true

    watchouts: - name: "scope_creep" signal: "plan grows, goal stays fuzzy" nemesis_response: "tighten scope"
  12. 03 modes.md Task-specific behaviour you switch on when you need

    it. W H Y I T M A T T E R S Different work needs different behaviour. Modes let you switch without re-explaining every time.
  13. modes.md ## mode:PLAN Purpose: break a project into steps. After

    Engage: - Tasks per milestone - Dependencies and risks - Next 3 actions Trigger: "mode:PLAN [your task]"
  14. 04 drift_log.md Where you record when it gets it wrong.

    The config is never done. W H Y I T M A T T E R S Your working style changes. The log keeps the config honest over time.
  15. drift_log.md ### 2026-03-01 — too much output What happened: Gave

    a full plan when I just needed a direction. Correction rule: Ask for goal first. Always wait for Engage.
  16. LET'S GET STARTED. T W O W A Y S

    T O E D I T Y O U R C O N F I G Manually Open the file. Change the values. Save. Tell your agent While in your editor, ask Claude Code or Gemini to update the files. github.com/terezaif/nemesis
  17. 5 MIN GET THE STARTER CONFIG Claude Code / Gemini

    CLI git clone github.com/terezaif/nemesis or download Claude Cowork open the workshop-starter folder Manual copy CLAUDE.md into your system prompt github.com/terezaif/nemesis
  18. 5 MIN E X E R C I S E

    1 THE QUESTIONNAIRE 9 questions. First instinct. No right answers. You'll get a YAML block to paste into your config. nemesis-compass.lovable.app
  19. 5 MIN E X E R C I S E

    2 LOAD MY PROFILE P R O M P T I've just completed the Nemesis questionnaire. Here are my results: [paste the YAML block from the survey here] Update my operator_profile.yaml with these values. After this: your agent knows how you work. Check it updated correctly. guides/02-edit-config.md → Prompt 1
  20. 10 MIN E X E R C I S E

    3 ADD A WATCHOUT P R O M P T Based on my profile, suggest one watchout to add to operator_profile.yaml. Make it specific to how I work — not generic. Show me the YAML block and explain why. If it doesn't land: "not quite — the pattern I actually want to catch is [describe it]." You can add more watchouts any time — log a drift entry, then ask Nemesis to formalise it. guides/02-edit-config.md → Prompt 2
  21. 5 MIN YOUR TRIGGER WORD The Alignment Gate needs an

    exact word. Mine is Engage. Pick yours. Engage ← mine U P D A T E I T I N T W O P L A C E S CLAUDE.md Do NOT proceed unless I write EXACTLY: [YOUR WORD] operator_profile.yaml alignment_gate_token: "[YOUR WORD]" Ask your agent: "My trigger word is [WORD]. Update CLAUDE.md and operator_profile.yaml everywhere."
  22. 5 MIN WRITING STYLE Don't fill this in manually. Let

    the agent observe. After a real conversation: "Based on our conversation, draft a writing_style section. What patterns did you notice? Output as YAML." Or — add this to your CLAUDE.md: "When you reach context limit and compact this conversation, first ask: Would you like me to update your writing_style section based on patterns I noticed in this session?" guides/03-writing-style.md
  23. DRIFT LOG Not automatic. Intentional. When After a session where

    something went wrong. What What happened. What the correction should be. How You write it — or ask Nemesis to draft an entry. "Nemesis got it wrong just now — [describe it]. Propose a drift_log.md entry." guides/04-drift-log.md
  24. SHARE OUT → What trigger word did you pick? →

    What watchout did your agent suggest? → What surprised you about the profile?
  25. WHAT'S NEXT Two ways to keep using your Nemesis Standalone

    repo · Keep nemesis-starter as a git project folder. · Open it in Claude Code (claude), Gemini CLI (gemini), or Cowork. · Config is scoped to that folder — Nemesis only activates there. · Best for: version-controlled, project-specific companion. System-level (always on) Copy CLAUDE.md to ~/.claude/CLAUDE.md — loads in every Claude Code session.
  26. B O N U S NOTES.MD The session memory your

    agent keeps for you. What it is A running log of decisions and open items — across sessions. Who writes it Your agent. Automatically, at the end of each session. What's in it Decisions (with rationale) · Open items (unresolved, dated). How to use it "What were the key decisions from our last session?" Check guides/05-notes.md
  27. B O N U S Keep Growing your agent The

    more you use your agent the more you will have things to change Example Even the modes.md interaction doesn’t have to be static. Should the agent ask what mode are you in? Should it detect a new mode? Should it notice what your expectations are? Check guides/06-modes.md
  28. Keep reflecting. The best sparring partners — human or agent

    — are the ones who know exactly how you work. Keep updating the config. Keep logging the drift. Keep asking better questions. Tereza Iofciu terezaiofciu.com Connect on LinkedIn → github.com/terezaif/nemesis nemesis-compass.lovable.app