norm

Use when working with the Bland API from a coding workspace, finding product documentation, editing agent JSON instead of clicking the dashboard, reconciling unsaved page context with saved versions, or debugging a failed Bland operation. Covers API discovery, workspace evidence, and page synchronization; not v1 pathway authoring.

Use when finding, reviewing, categorizing, or analyzing Bland calls in bulk, including hundreds or thousands of recent calls, reproducing a specific call failure, or producing a reproducible workspace report with evidence. Covers paginated API retrieval and code-based analysis without putting full datasets in chat.

Use when creating or calibrating Bland evaluation judges, configuring test cases and attached judges, running or interpreting evaluations and simulations, or investigating failed, partial, or inconclusive scores. Distinguishes voice agents, evaluation judges, test scenarios, saved call cohorts, and asynchronous run results.

Designing and building a NEW Bland v2 agent from scratch — scenario architecture (what scenarios represent and where boundaries belong), hub + entry authoring, step/edge/tool design, the standard conduct-rule set, and how v2 building differs from v1 pathway building. Use when creating a new v2 agent, designing its scenarios, or when someone asks how building in v2 differs from v1. If the user has not said v1 or v2, ASK before applying this doctrine.

The Bland v2 agent lifecycle — environments (dev/staging/production), publish/promote/rollback, which version answers each channel at call time, branches and merging, A/B experiments, environment variables ({{env.KEY}}), pre-deploy checks, inbound number binding and the twin-pathway fallback. Use when shipping an agent version to real traffic, wiring numbers, gating a promotion, or answering "which version is this call running".

Use when updating, repairing, debugging, or managing an existing Bland v2 agent, including prompt edits, routing, tools, settings, versions, environment configuration, and regression testing.

The hand-migration doctrine for converting a Bland v1 pathway or persona (raw JSON export) into a v2 agent snapshot — architecture rules, verbatim-carriage discipline, the full trap catalog, and the verification bar. Use whenever migrating or converting v1 pathways/personas to v2 agents, or when a migrated agent misbehaves. The /norm:migrate command drives this skill end to end.

Use when predicting or debugging Bland v2 runtime behavior, explaining routing or interruptions, or reconciling edited snapshot JSON with the builder canvas, hidden routes, and call traces.

The Bland v2 agent snapshot dialect — the exact JSON shape of an agent version (behavior graph, hub, complex scenarios, flow steps, edges, tools, settings) and the rules for authoring one by hand. Use whenever reading, writing, or debugging an agent_versions snapshot, pushing a version via POST /v2/agents/:id/versions, or carrying content into a snapshot during migration.

Use when verifying Bland v2 agents with simulations, test chat, or speech-to-speech tests; measuring interruptions or response latency; grading engine traces; or deciding whether a test is safe. Covers fixtures, the agent-testing API, voice evidence, and live-integration safety.