Context Handoff

Context Handoff is a free browser game for software testers about the instructions file an AI coding agent reads. You sort your notes into three kinds: durable context for agents.md, session chat context for the current run only, and unsafe context that is never written down. You then write that file, choose its headings, tighten its wording, and settle conflicts with a nested AGENTS.md. An agent then reads your file line by line and breaks wherever a line was weak.

AI coding agents start every session with no memory of the last one. Whatever you want tomorrow's agent to know, you must write down tonight. The agent reads that file before it starts work. That file is called agents.md, and this game is about what goes in it. Every piece of context you hold is true right now. Some of it is still true next month. Some of it stops being true tonight. Some of it is dangerous in writing at all. Context Handoff makes you sort them. You then write the file. The game hands your file to an agent that does not remember you and never asks what you meant. ShiftSync, the Tricentis community for quality engineering, supports Context Handoff. The glossary inside the game links to more reading from ShiftSync, Testing Titbits and Ministry of Testing.

Open Context Handoff on Testing Titbits Playground · 5-9 min · AI Testing · Free, no sign-up

How to play Context Handoff

  1. A three-note warm-up teaches what an agents.md file is and what does not belong in one.
  2. A team charter states four things the team holds true. It gives the ship rate, the runner's reach, what counts as evidence, and how the team treats a hunch.
  3. Act I: nine notes arrive one at a time. The short run deals seven. Route each one as durable context for agents.md, session chat context for this run only, or unsafe context that is never written down.
  4. More than one lane is often defensible. The charter makes the call answerable. It does not make the call automatic.
  5. Act II: the notes that belong in the file become a file you author. Choose the heading each line sits under.
  6. Rewrite the rambling lines. A meter prices every line you keep, on every run, forever.
  7. A nested tests/AGENTS.md contradicts the root file two or three times. Decide which one the agent obeys.
  8. Act III: the agent reads your file line by line. Where a line was weak you watch the damage land, then patch it and run again.
  9. One last beat fires on a line you wrote well. The agent obeyed it on turn two. By turn forty it obeys the line no more. Only one of the three remedies holds. The game scores you out of 100.
  10. Optional, and outside the run: forge a base context. agents.md is the project's context and lives in the repo. A base context is yours, it lives in your AI tool, and it loads on every session before you type anything. The game hands you the interview prompt, you run it in Claude, ChatGPT or Perplexity - about ten minutes, eight questions - and nothing you say is stored. Do that and the drift beat has nothing left to drift off.

What you will learn

Frequently asked questions

What is context engineering?

Context engineering is the practice of deliberately choosing what information an AI model receives. That means the instructions, documents, examples and tools, and not the prompt wording alone. Where each piece of context lives, and how long it lives, is part of the practice.

What is an agents.md file?

agents.md is a plain-text instructions file that AI coding agents read when they start work in a repository. It carries the durable context a session needs: conventions, environments and known workarounds. Nobody has to re-explain that context on each run. An agent reads everything in the file on every future run, whether or not it is still true.

What should go in an agents.md file?

Context that is true on every run and cheap to state once. That means environments, the conventions your team always obeys, prohibitions, and known traps in the product. Keep out anything with a date, a version or a ticket number in it. Never write down a credential, a token or real customer data.

Which file wins, agents.md or a nested AGENTS.md?

Usually the file nearest to the work wins inside its own folder. The people who own the tests wrote tests/AGENTS.md, so their rules apply to their tests. The exception is a ban. A nested file can add limits to a rule from the root file, but it must never cancel a ban.

What is stale context?

Stale context is a note that was accurate when you wrote it and is not accurate any more. Last sprint's feature flags, an expired token and an old version number are all stale context. It is more dangerous than missing context, because the model uses it with confidence and nothing flags it as out of date.

How much does an agents.md file cost to keep?

An agent re-reads every line on every run, so the cost is a rate and not a one-off. A file read a hundred times a week costs a hundred times the tokens of a file read once. A line that says nothing is not free.

What is a skill file, and when should a procedure move out of agents.md?

A skill file is a separate document an agent opens when it needs it. The agent does not read it on every run. A multi-step procedure is the clearest case: a release checklist, or a data reset runbook. The procedure is durable, so it does not belong in the session. It is only needed on the runs that perform it, so all of its steps in agents.md charge every other run. Leave one line in agents.md that points at the skill file, and keep the steps in that file. That line must point at something the agent can open. A pointer to a procedure nobody wrote down is the same length as one that works, and it is worth nothing.

Why do AI agents stop following their instructions file?

The agent obeys the file only while the file is still in the context it attends to. It reads agents.md at the start of a long run. Everything that arrives after it competes for the same attention. So the agent can obey a rule on turn two and ignore it by turn forty. Nothing errors when that happens. The agent simply acts as if the line was never written. A reminder works, and then wears off again. The remedy that holds is structural. Keep the file short enough that the agent still reads the rule at the end of the run.

What is a base context, and how is it different from an agents.md file?

A base context is a written profile of how one person tests: their instincts, their tradeoffs, their risk appetite and how they want to be answered. It lives in the AI tool itself, in its custom instructions or its memory, so it loads on every session before anyone types a prompt. An agents.md file is the other half. It is the project's context, it lives in the repository, and everyone working on that repository gets the same one. You want both. The project file tells the agent what is true about this codebase; the base context tells it who it is working with. The base context is also the half that does not drift, because the tool loads it standing rather than reading it once at the top of a long run.

Who supports Context Handoff?

ShiftSync, a Tricentis community of about 80,000 testers and quality engineers, supports Context Handoff. The game is free to play and needs no account. The ShiftSync link on the result screen goes to their community. Testing Titbits builds and maintains the game.

Is this game suitable for a team session?

Yes. A full run takes about nine minutes and a short run about five. Several of the calls are deliberately arguable. A workshop seed pins everyone to the same notes, so you can compare the answers afterwards. Share the seed link from the result screen.

More free games and infographics for testers

Created by Rahul Parwal · TestingTitbits.com