Assumption Hunter
Points out what your agent assumed without checking.
What it does
It reads a plan, a diff or an answer and picks out the premises that were never stated or verified, like how a webhook behaves or whether a delete should be permanent. For each one it weighs how likely it is to be wrong and what happens if it is, then says whether to check it, ask you, or carry on with a reversible default.
Why it's useful
When a request leaves something open, agents fill the gap with a guess and keep going as if it were a requirement. Most guesses are harmless. A few, like assuming a payment webhook arrives exactly once or that a deletion can be permanent, are expensive when they turn out wrong.
It tries not to be annoying. Trivial conventions and easy-to-change defaults are left alone, it aims for three to five high-impact findings instead of listing everything unspecified, and it asks one question at most, only when the answer blocks safe progress. A clean audit is a valid result.
It keeps what it checked separate from what it inferred, and says what it couldn't inspect. A model reviewing its own work can repeat its own mistakes, so treat the result as a second look, not independent verification.
How it classifies each premise
- Confirmed
- You stated it, or it was checked against files, tests or docs.
- Reasonable default
- Not specified, but low-risk and easy to change.
- Unverified
- Plausible, but nothing establishes it yet.
- Contradicted
- Conflicts with your requirements or the evidence.
- Blocking ambiguity
- Different readings lead to very different or unsafe results, so it asks.
Audit mode (default)
It only reads and reports. It doesn't edit or revert anything, so what happens next is up to you.
Strict mode (ask for it)
The agent stops and asks you before doing anything the skill would flag. This is an instruction to the agent rather than something enforced, so it depends on the agent following it.
What a report looks like
illustrative example
Assumption audit: payment webhook handler Finding: each webhook event arrives exactly once Status: Unverified Evidence: none available. The handler credits the account on every event. Risk: a retried event would credit the account twice. Next step: store processed event IDs and skip repeats. Finding: deleting a customer removes their rows permanently Status: Blocking ambiguity Evidence: the request says "delete customer" and nothing about retention. Risk: if records must be kept, this can't be undone. Blocking question: should deleted customers be removed for good, or kept in a recoverable state? Safe path: soft delete behind a flag until retention rules are confirmed.
Install
The skill is a folder with a SKILL.md in it, following the Agent Skills format. There's nothing to build or run, and it doesn't need an API key.
01Claude Code
Clone it into your personal skills folder to use it in every project.
$ git clone https://github.com/ritenv/assumption-hunter ~/.claude/skills/assumption-hunterTo use it in a single repo, clone it into
.claude/skills/assumption-hunterat the project root instead. If you commit that folder, everyone working on the repo gets the skill.02Claude apps
Zip the folder and upload it as a custom skill from Settings in claude.ai or the desktop app.
$ git clone https://github.com/ritenv/assumption-hunter && zip -r assumption-hunter.zip assumption-hunter -x "*.git*"03Codex, Cursor and other agents
If your agent supports Agent Skills, drop the folder into its skills directory. If it doesn't, paste the contents of
SKILL.mdinto your project instructions (AGENTS.md, a rules file, or a system prompt) and the agent can follow it from there.
What's in the folder
- assumption-hunter/
- SKILL.md
- README.md
- references/classification-guide.md
- templates/assumption-audit.md
Example prompts
Agents can pick the skill up on their own when a task matches its description, but naming it in your prompt is more reliable.
- Use Assumption Hunter to audit this implementation plan.
- Review this Git diff for unsupported assumptions.
- Use Assumption Hunter in strict mode while planning this migration.