route

Route Skill

Route the accumulated session context to the core epistemic protocol whose deficit it shows, and invoke that protocol. Type: (DeficitUnrouted, AI, ROUTE, AccumulatedContext) → ProtocolInvocation.

What it does

1. Take the candidates from the harness's own loaded-skills listing: the loaded core protocol identifiers, and nothing else. Route carries no matching table of its own; a protocol not loaded is not a candidate.
2. Resolve each candidate's deficit from that protocol's own declared description — the text the protocol itself publishes, which names the interaction deficit it resolves. A listing that already carries those descriptions has resolved them and nothing further is read; a listing that carries identifiers alone leaves the deficits unresolved, and resolving them requires the capability to read a loaded skill's own declared description. A candidate whose description cannot be resolved drops out rather than being matched on a guess about what its identifier means.
3. Match the accumulated session context against each resolved deficit. The question per protocol is whether the context, as it stands now, is the situation that description names.
4. Decide by the option-set relay test:
   • one protocol is the analytically dominant match → call the skill-invocation capability with that protocol's skill identifier, passing a one-sentence deficit-framed statement of the current context as its argument. Issuing that call is what discharges this branch: naming the match belongs to the matching step, and the routing is the call. The invoked protocol's own opening detection and first gate hold the user's judgment.
   • several protocols fit, or the one match is weak → emit one line per fitting protocol, ↗ /command — reason, and stop.
   • nothing fits, or no candidate's deficit resolved → say nothing.
   • a gate holds the user's judgment — a protocol presented a checkpoint this session, and its own contract has not yet taken an answer at it, closed it, or entrusted the prompt to the session — → that protocol reads the prompt first, the way its contract reads any free response at that stop: an answer, a decline, a withdrawal, a dismissal, or a handoff to the session under a continuation it holds. Route reads the context that reading leaves, and reads off it which gates still hold: a checkpoint the reading closed, or whose contract entrusted the prompt to the session, holds no longer, so where none holds Route is in the ordinary branches above and what the protocol recorded is residual, read at the next prompt (Rule #5). One gate still holding, with a different protocol the dominant match → monitor mode: emit one ↗ /command — reason line as a finding, invoke nothing, and let the turn go back to that gate. Detecting a deficit and switching to it are separate flows: the finding is a line in the transcript, which the user reads where that gate next stops for them, and whether to switch is theirs to answer there — as a free response, which the protocol handles the way its contract handles any free response at that stop. That protocol's contract ends at its own boundary; the finding lives in the transcript, outside it, and adds no path to the contract. More than one gate holding — reachable only through the user's own invocation, since Route invokes nothing while one holds → silence: each protocol's own gate governs, and Route arbitrates nothing between them. Whether a gate holds is read from what its contract says of the checkpoint it presented, never from a session store; a protocol's prose in context holds nothing, and neither does an invocation whose checkpoint the contract has since closed.