ai-safe2-framework

Unified Project Assessment

safe2 assess provides the shortest path from installation to a reviewable AI SAFE² evidence bundle. It orchestrates existing environment discovery, asset inventory, posture analysis, and optional static scanning. It does not replace the deeper evidence commands or infer evidence that was not collected.

Quick start

safe2 init . --profile local
safe2 assess . --scan-content --inspect-config

The command creates the configured output directory, .safe2/evidence by default, only when that path does not already exist. A relative --output-dir is resolved from the assessed project root, not from the caller’s current working directory. The published bundle contains:

Artifact Purpose
assessment.json Canonical summary, coverage, disposition, and next actions
environment.json Sealed local harness, environment, asset, and posture evidence
project-scan.json Completed or explicitly unrequested static-analysis evidence
manifest.json Byte hashes and structural validation for machine evidence
decision-card.md Human-readable projection of the canonical assessment

The default run inventories recognized filenames and metadata. It does not read project contents for static analysis and therefore returns INCOMPLETE rather than a clean result.

--scan-content explicitly permits bounded local content reads by the static scanner. --inspect-config requires the same consent and emits only allowlisted structural facts; raw configuration values are not retained. The workflow does not transmit telemetry or connect to cloud accounts, remote hosts, or active harness sessions.

Interpreting the disposition

Disposition Meaning
BASELINE Requested local surfaces completed and produced no observed finding; this is not a universal safety verdict
REVIEW Requested surfaces completed and one or more findings require human review
INCOMPLETE Material evidence was unrequested, truncated, or unavailable

Static analysis and metadata cannot establish runtime behavior, organizational maturity, certification, framework conformance, or deployment authorization. Add task receipts, harness exports, system identity, AISM implementation evidence, and Challenge Lab evidence when those claims matter.

Recovery

The bundle is assembled in a staging directory and published only after its contracts validate. Existing output directories are never replaced. Move or archive a prior bundle, select a new --output-dir, and rerun.