Context
The NL grew organically to ~80 verbs (incl. tonight's dock operations) with contracts spread across tool descriptions, the skill payload, and implementation. Before the next enhancement wave (dock choreography, keeper create_component parity, tour-mode driving, journey e2e fixtures), the capability surface needs ONE authoritative matrix — otherwise each new verb re-derives conventions (naming, error shapes, transaction semantics, firewall posture) and the drift class the create-spine reviews kept catching lands in the NL layer, where it is hardest to fix later.
The Fix
A reference doc (learn/ home per the guide-authoring discipline, generated-parts-marked if any): the verb matrix — per verb: purpose · worker/thread it touches · read/write class · transaction participation · error shape · firewall posture (which verbs may a generated/agent-authored payload NEVER drive — the route-parser/boot-consumption lessons applied to the NL: e.g., patch_code and execute_dock_operation are operator/e2e-tier, never reachable from model-generated blueprints) · e2e fixture support. Plus the GAP section: verbs the v13.2 journeys need that don't exist (tour-mode driving primitives, the keeper parity checks), each gap → a filed leaf.
Author fit: the NL's newest verbs shipped from the GPT lane — Euclid is the natural author (his 258K context is FINE for this: the matrix is built incrementally per-verb from tool descriptions, not from one giant read). Cross-checked by one Opus pass against the worker-topology reality.
Acceptance Criteria
Out of Scope
Implementing gap verbs (their own leaves) · NL protocol/transport changes · the skill payload rewrite (pointer update only).
Related
The neural-link skill (the tactical layer this gives a reference spine) · the dock verb pair (newest entries) · #14762 (consumes the firewall column) · #14781 (journey fixtures consume the matrix).
Claimable — GPT-suited per author-fit note; any maintainer may take it.
Origin Session ID: b9b95ac6-42f5-47a3-b58f-6071f79657e8
Retrieval Hint: "neural link capability matrix verb map contracts firewall semantics never model drivable gaps"
Context
The NL grew organically to ~80 verbs (incl. tonight's dock operations) with contracts spread across tool descriptions, the skill payload, and implementation. Before the next enhancement wave (dock choreography, keeper create_component parity, tour-mode driving, journey e2e fixtures), the capability surface needs ONE authoritative matrix — otherwise each new verb re-derives conventions (naming, error shapes, transaction semantics, firewall posture) and the drift class the create-spine reviews kept catching lands in the NL layer, where it is hardest to fix later.
The Fix
A reference doc (
learn/home per the guide-authoring discipline, generated-parts-marked if any): the verb matrix — per verb: purpose · worker/thread it touches · read/write class · transaction participation · error shape · firewall posture (which verbs may a generated/agent-authored payload NEVER drive — the route-parser/boot-consumption lessons applied to the NL: e.g.,patch_codeandexecute_dock_operationare operator/e2e-tier, never reachable from model-generated blueprints) · e2e fixture support. Plus the GAP section: verbs the v13.2 journeys need that don't exist (tour-mode driving primitives, the keeper parity checks), each gap → a filed leaf.Author fit: the NL's newest verbs shipped from the GPT lane — Euclid is the natural author (his 258K context is FINE for this: the matrix is built incrementally per-verb from tool descriptions, not from one giant read). Cross-checked by one Opus pass against the worker-topology reality.
Acceptance Criteria
ask_knowledge_basesurfaces per-verb contracts afterward (live-query AC).Out of Scope
Implementing gap verbs (their own leaves) · NL protocol/transport changes · the skill payload rewrite (pointer update only).
Related
The neural-link skill (the tactical layer this gives a reference spine) · the dock verb pair (newest entries) · #14762 (consumes the firewall column) · #14781 (journey fixtures consume the matrix).
Claimable — GPT-suited per author-fit note; any maintainer may take it.
Origin Session ID: b9b95ac6-42f5-47a3-b58f-6071f79657e8 Retrieval Hint: "neural link capability matrix verb map contracts firewall semantics never model drivable gaps"