@stackrail-io/aprf

Local APRF CLI — collect evidence, assess from collector statusHints, render REPORT.html.

Pinned Check catalog via @stackrail-io/aprf-engine. No StackRail backend required.

Install

npm install -g @stackrail-io/aprf
# or
npx @stackrail-io/aprf@latest version

Requires Node.js ≥ 22.

Cursor: install the plugin under plugins/aprf/ (see its README) — skill/commands call this CLI.

Commands

# Collect evidence graph + per-plugin import reports
aprf collect --target /path/to/app --out ./aprf-assessment

# Score profile mandatories (requires --system-type, or TTY prompt)
aprf assess --out ./aprf-assessment --system-type ai-application --profile core

# Framework / SDK primitive gate (infers systemType=ai-framework)
aprf assess --out ./aprf-assessment --profile framework

# Application with capability lenses (additive)
aprf assess --out ./aprf-assessment --system-type ai-application \
  --capabilities rag,agents --profile core

# Dry-run Check set / claim metadata (skill Phase 0)
aprf resolve-target --system-type ai-framework --json

# Render + verify HTML
aprf report --in ./aprf-assessment/assessment.json --out ./aprf-assessment/REPORT.html
aprf verify ./aprf-assessment/REPORT.html

# One shot: collect → assess → report → verify
aprf audit --target . --out ./aprf-assessment --system-type ai-application --profile core
aprf audit --target . --out ./aprf-assessment --profile framework

Live credentials (collect / audit)

Same flags on collect and audit. Providing --base-url or admin/limited creds auto-enables live mode (no separate --live required). Passwords/tokens are never written to reports.

Prefer environment variables set outside chat (secret manager or non-echoing prompt). Do not pass passwords/tokens on argv — process lists and CI logs can capture them:

# Prerequisites already set in the local shell:
#   APRF_ADMIN_EMAIL + APRF_ADMIN_PASSWORD  — or — APRF_ADMIN_TOKEN
# Optional AUTHZ-M1: APRF_AUTHZ_LIMITED_EMAIL + APRF_AUTHZ_LIMITED_PASSWORD
#   — or — APRF_AUTHZ_LIMITED_TOKEN
npx @stackrail-io/aprf audit --target . --out ./aprf-assessment --system-type ai-application --profile core \
  --base-url http://127.0.0.1:8080

| Flag / Env | Used by | | --- | --- | | --base-url / APRF_AUTH_PROBE_BASE_URL | AUTHN-M1, AUTHZ-M1, AUTHN-M2 | | APRF_ADMIN_TOKEN (or discouraged --admin-token) | AUTHN-M2, AUTHZ-M1 (temp user) | | APRF_ADMIN_EMAIL + APRF_ADMIN_PASSWORD | sign-in → JWT | | APRF_AUTHZ_LIMITED_* | AUTHZ-M1 denial probe |

Other collector evidence: drop measured JSON under ./aprf-assessment/imports/<pluginId>/, or set collector-specific env (e.g. GITHUB_TOKEN with live mode for github-actions).

Assess engine (deterministic)

Aligned with auditor scoring.yaml / confidence.yaml / evidence-precedence.yaml:

• Status from imports/*/…-report.json statusHint, else collector detail (CHECK status=…)
• Evidence-graph nodes with relatedCheckIds attached (precedence-ranked)
• Mandatory gate: every applicable mandatory is PASS or NOT_APPLICABLE
• NOT_APPLICABLE is excluded (passed: false) — not a vanity pass
• Default assess scores profile mandatories only (collector hints outside the gate are ignored). --full scores the non-deprecated catalog (ai-application only; forbidden with --system-type ai-framework).
• Unscored profile Checks → NOT_DEMONSTRATED (blocker if mandatory)
• recommendedScore = severity-weighted recommended Checks only (null / n/a under default profile assess; use --full to score recommended)
• audit without --plugins / --full collects only plugins that map to resolver effectiveCheckIds (profile ∪ capabilities ∪ lenses)
• Required --system-type ai-application|ai-framework|non-ai-platform for assess/audit/resolve-target (TTY prompts when omitted; non-TTY needs the flag, APRF_SYSTEM_TYPE, or --profile framework). No silent Core default.
• Optional --capabilities chatbot,rag,agents,… (ai-application; additive lenses)
• Optional --lens rag,agents,voice,coding
• aprf resolve-target --json prints resolver output for skill/dry-run
• Framework assessments set assessmentKind=aprf-framework + scope.reportBanner (not a Core production claim)
• Each control may include informative crosswalks and threatIntel from the pinned catalog (never gate inputs)
• REPORT.html executive summary includes Top threat exposure across unmet controls

Agent YES/NO/DON'T KNOW attestation fills remain in skills/aprf-auditor.

Versions

| Package | Role | | --- | --- | | @stackrail-io/aprf | This CLI | | @stackrail-io/aprf-engine | Normative Check catalog | | @stackrail-io/aprf-framework-definition | Profiles / lenses / resolveAssessmentTarget |

aprf version prints CLI + catalog SemVer.

License

Apache-2.0