When a Required Tool Is Missing, Stop Guessing

A goal can require current evidence even when the environment cannot yet gather it. That is the moment an agent system is most tempted to turn a missing capability into a polished guess.

WTK now treats that gap as a decision to resolve, not permission to keep drafting.

Current state

Evidence requirements belong to the goal. They do not disappear because a useful capability has not been selected, configured, or made available.

A generic missing-tool message is not enough. It leaves the operator without a clear connection between the original obligation, the available choices, and the consequences of changing the plan.

Changes

When a required evidence capability is missing, WTK can identify compatible registered options and distinguish a usable option from one that still needs setup or explicit approval.

The choice remains with the operator where authority, cost, or configuration matters. Selecting a capability creates a revised governed goal and restarts the affected research and build work. It does not quietly patch a generated draft or erase the original evidence obligation.

If no suitable option is available, WTK keeps the limitation visible. It does not invent a free capability, silently substitute a different kind of evidence, or present a tool-free answer as if the requirement never existed.

Failures observed

This change contains several easy shortcuts:

• treating an evidence requirement as optional when the needed capability is absent;
• using a vague tool label as if it were an executable evidence path;
• silently selecting a capability with different authority or cost;
• changing a generated draft instead of revising the governed goal;
• returning confident prose because the required evidence was inconvenient to obtain.

Assumptions removed

We no longer assume that a tool declaration makes a capability available.

We no longer assume that an unavailable capability is only a setup nuisance.

We no longer assume that the system may resolve a missing evidence path by changing the goal on its own.

Evidence boundary

WTK can preserve a missing evidence obligation through capability selection and a governed goal revision. It can keep unavailable options visible and prevent a silent substitution from becoming source truth.

That does not prove that a selected capability will return relevant, fresh, or trustworthy evidence. It also does not establish that every runtime or team path will enforce the resulting obligation consistently.

Research question advanced

This cycle informs the question of when review can be reliable by protecting the evidence path before construction begins. A review cannot add confidence to an answer whose required evidence capability was silently removed from the plan.

Next hypothesis

The boundary should distinguish unavailable, incompatible, unconfigured, costly, failed, partial, and valid evidence capabilities. The next test will vary those conditions while checking that the original obligation remains visible and that no unsupported answer path appears.