How Vise works - social.plus

Vise gives an AI coding agent a repeatable social.plus integration loop. You describe the outcome, answer the product decisions only your team can make, and review the result.

From request to evidence

On resumed work, vise status reads the current evidence first and routes the agent to the relevant build, Day-2, or release-review step. It does not run project sensors or change the existing compliance result.

Inspect and plan. The agent identifies the app surface, reads social.plus documentation and SDK facts, and asks instead of guessing.
Confirm decisions. Vise records the feature, placement, data target, selected capabilities, and definition of done.
Implement and check. The agent validates real SDK usage, platform lifecycle, and feature completeness while it works.
Run project checks. The app’s build, lint, typecheck, tests, and SDK smoke checks run when available.
Record evidence. The result explains what passed, what still needs work, and whether the intended surface worked at runtime.

Read the result

Ask your agent to separate the outcome into what passed, what it will fix, and what needs your decision.

Result	What happens next
`green`	Review the feature through your normal product process
`deterministic-failures` or `completeness-gap`	The agent fixes the code or finishes the agreed scope
`selected-capability-failures`	Finish the selected capability or explicitly remove it from scope
`blocked`	Answer the named product or project question
`needs-attestation`	Review evidence for correct architecture the automatic check cannot see
`contract-drift`	Confirm the changed intent or restore the implementation
`runtime-proof-waived`	Accept the explicit waiver or request a real launch
`engagement-drift`	Reopen the earlier feature or explicitly accept the change

no-platform means Vise did not find a supported app surface. Confirm that the agent is working in the correct package before concluding that there is nothing to validate.

What the evidence means

The contract, check results, attestations, and runtime receipts live in sp-vise/. Commit that folder with the application so review and CI do not depend on an agent transcript. Evidence can prove an automatic check passed, explain architecture a static rule cannot follow, or show that a specific surface mounted and loaded data at its recorded route and target. It does not replace product QA, security review, or human release approval. User-visible work still needs a real launch or an honest recorded waiver.

Claim	What Vise can establish	What remains separate
Compliance	Recorded scope, deterministic SDK/platform checks, and earlier-feature drift	Product and security judgment
Project sensors	The detected build, lint, typecheck, test, or SDK-smoke command result	Tests that were unavailable or not selected
Runtime receipt	Retained source, build, runner, route, target, artifact, and Passport lineage still match	Visual quality and unobserved SDK semantics
SDK assurance	Exact extracted symbols, models, capability anchors, provenance, freshness, and fact-depth state	Compatibility beyond the reviewed evidence
Organization policy	Accepted source, lifecycle, trust-key state, update impact, and responsible owner	Policy acceptance and release approval
Passport and advisory review	Digest-bound handoff claims and disclosed movement	The human decision to ship

Maintain an existing integration

Add, diagnose, or upgrade without losing earlier evidence.

Keep integrations verified in CI

Carry the same contract and evidence into the pipeline.

Command reference

Look up the full command surface when operating Vise directly.

Run Vise as an MCP server

Expose Vise’s local tools to an MCP-capable coding host.

​From request to evidence

​Read the result

​What the evidence means

​Related