Tool descriptions are part of the control surface
The wording that tells a model what a tool does is an instruction the model will follow.
Change the description and you change the behaviour, including when to use it and what to pass. Anybody who can edit those descriptions can influence what the agent does without touching the prompt or the code, which makes them configuration that deserves review and version control.
More on AI agents
- Agent memory can preserve poisoned contextIt stays in the water
- Agent identity should be separate from user identityTwo necks, one badge
- Delegated agents create authority chainsThe thread stays attached
- Agent loops need budgets and stopping conditionsSomething has to say enough
- High-impact tools need stronger confirmationMatch the catch to the consequence
- Tool results are untrusted inputs tooNo sieve on the back route