`uofa` CLI reference (v0.17.0)

Generated from uofa --help-all at 2026-08-30T02:53:22Z. 27 subcommands available.

Synopsis

uofa [--no-color] [--verbose] [--repo-root PATH] [--pack NAME] <command> [...]

Global flags inherited by every subcommand: --no-color, --verbose, --repo-root, --pack. See uofa --help for details.

Commands

uofa keygen

generate ed25519 keypair for signing

uofa keygen [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--force]
                   path

Flag	Description
--force	overwrite an existing keypair (every package signed with the old key stops verifying)

uofa sign

sign (or re-sign) a UofA file

uofa sign [-h] [--no-color] [--verbose] [--repo-root PATH]
                 [--pack NAME] --key KEY [--context CONTEXT] [--output OUTPUT]
                 [--as ROLES]
                 file

Flag	Description
--key, -k	ed25519 private key (PEM)
--context, -c	JSON-LD context file
--output, -o	output path (default: overwrite input)
--as ROLES	comma-separated signing roles: issuer, reviewer. One party signing both scopes uses one invocation (--as issuer,deciding-engineer); separate parties use separate invocations with their own keys.

uofa verify

verify hash + ed25519 signature (C1 integrity)

uofa verify [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--pubkey PUBKEY] [--context CONTEXT]
                   [--decision-pubkey DECISION_PUBKEY]
                   file

Flag	Description
--pubkey	ed25519 public key for the measurement/issuer scope. No default: without it, verify tries the wheel-shipped anchors and names which one matched.
--context, -c	JSON-LD context override (default: the package’s own @context)
--decision-pubkey	public key for a decision signature. REPEATABLE: stacked decisions (source acceptance + concurrence + encoder attestation) carry signatures from several parties, and each key matches by key identity.

uofa shacl

validate against SHACL profiles (C2 completeness)

uofa shacl [-h] [--no-color] [--verbose] [--repo-root PATH]
                  [--pack NAME] [--raw] [--explain]
                  [--explain-functions EXPLAIN_FUNCTIONS]
                  [--explain-format {text,json,markdown,html}]
                  [--explain-backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                  [--explain-model EXPLAIN_MODEL]
                  [--explain-base-url EXPLAIN_BASE_URL]
                  [--explain-max-items EXPLAIN_MAX_ITEMS] [--explain-no-cache]
                  file

Flag	Description
--raw	show raw pyshacl output instead of friendly messages
--explain	run the interpretation pipeline after the primary analysis
--explain-functions	comma-separated list of interpretation functions to run (values: explain, group, contextualize, cross, narrative). Default: all applicable for the command.
--explain-format {text, json, markdown, html}	output format for the interpretation block. Default: same as the command’s primary —format, falling back to text.
--explain-backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	LLM backend for explain (overrides [llm] backend in uofa.toml)
--explain-model	model name on the chosen backend (overrides [llm] model)
--explain-base-url	base URL for openai-compatible backends
--explain-max-items	limit interpretation to top N items by severity
--explain-no-cache	bypass cached interpretation results

uofa rules

detect quality gaps with Jena rule engine (C3)

uofa rules [-h] [--no-color] [--verbose] [--repo-root PATH]
                  [--pack NAME] [--rules RULES] [--context CONTEXT] [--build]
                  [--raw] [--format {summary,turtle,ntriples,jsonld,json}]
                  [--output OUTPUT] [--explain]
                  [--explain-functions EXPLAIN_FUNCTIONS]
                  [--explain-format {text,json,markdown,html}]
                  [--explain-backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                  [--explain-model EXPLAIN_MODEL]
                  [--explain-base-url EXPLAIN_BASE_URL]
                  [--explain-max-items EXPLAIN_MAX_ITEMS] [--explain-no-cache]
                  file

Flag	Description
--rules, -r	path to .rules file
--context, -c	JSON-LD context file
--build	auto-build the Jena JAR if missing
--raw	show raw output without coloring
--format, -f {summary, turtle, ntriples, jsonld, json}	output format (default: summary). ‘json’ is the parsed-firings shape suitable for snapshot tests; ‘jsonld’ is the raw RDF.
--output, -o	write reasoned output to a file (default: stdout)
--explain	run the interpretation pipeline after the primary analysis
--explain-functions	comma-separated list of interpretation functions to run (values: explain, group, contextualize, cross, narrative). Default: all applicable for the command.
--explain-format {text, json, markdown, html}	output format for the interpretation block. Default: same as the command’s primary —format, falling back to text.
--explain-backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	LLM backend for explain (overrides [llm] backend in uofa.toml)
--explain-model	model name on the chosen backend (overrides [llm] model)
--explain-base-url	base URL for openai-compatible backends
--explain-max-items	limit interpretation to top N items by severity
--explain-no-cache	bypass cached interpretation results

uofa check

full pipeline: SHACL + integrity + rules (C1+C2+C3)

uofa check [-h] [--no-color] [--verbose] [--repo-root PATH]
                  [--pack NAME] [--pubkey PUBKEY] [--context CONTEXT]
                  [--rules RULES] [--skip-rules] [--build] [--oos | --no-oos]
                  [--derivations | --no-derivations] [--explain]
                  [--explain-functions EXPLAIN_FUNCTIONS]
                  [--explain-format {text,json,markdown,html}]
                  [--explain-backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                  [--explain-model EXPLAIN_MODEL]
                  [--explain-base-url EXPLAIN_BASE_URL]
                  [--explain-max-items EXPLAIN_MAX_ITEMS] [--explain-no-cache]
                  file

Flag	Description
--pubkey	ed25519 public key
--context, -c	JSON-LD context file
--rules, -r	path to .rules file
--skip-rules	skip the Jena rule engine (no Java required)
--build	auto-build the Jena JAR if missing
--oos	force OOS rules on for this run, overriding pack config
--no-oos	force OOS rules off for this run, overriding pack config
--derivations	force derivation pre-pass on for this run, overriding pack config
--no-derivations	force derivation pre-pass off for this run, overriding pack config
--explain	run the interpretation pipeline after the primary analysis
--explain-functions	comma-separated list of interpretation functions to run (values: explain, group, contextualize, cross, narrative). Default: all applicable for the command.
--explain-format {text, json, markdown, html}	output format for the interpretation block. Default: same as the command’s primary —format, falling back to text.
--explain-backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	LLM backend for explain (overrides [llm] backend in uofa.toml)
--explain-model	model name on the chosen backend (overrides [llm] model)
--explain-base-url	base URL for openai-compatible backends
--explain-max-items	limit interpretation to top N items by severity
--explain-no-cache	bypass cached interpretation results

uofa report

render a credibility report for a bundle, or for an HF model id/URL (fetch + extract)

uofa report [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--format {text,markdown,json}]
                   [--output OUTPUT] [--deterministic] [--revision REVISION]
                   [--save-bundle PATH] [--no-save-bundle]
                   [--extract-backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                   [--extract-model EXTRACT_MODEL]
                   [--extract-base-url EXTRACT_BASE_URL] [--cou TEXT]
                   [--mrl N] [--raidex PATH] [--raidex-hub] [--raidex-run]
                   [--raidex-args ARGS] [--raidex-yes]
                   source

examples:
  uofa report package.jsonld --pack model-credibility
      a saved bundle: deterministic, no extraction

  uofa report allenai/OLMo-2-1124-13B-Instruct --pack model-credibility
      a live HuggingFace card: fetch, extract, assess. Section [1] is
      documentation completeness, section [3] evaluation sufficiency.

  uofa report owner/model --pack model-credibility --cou 'clinical triage' --mrl 3
      scope the assessment to a decision. --cou raises W-EV-COU-05 to
      Critical; --mrl enables COMPOUND-EV-01. Omitting them is not a
      pass -- the readout states the escalation was not assessed.

  uofa report owner/model --pack model-credibility --raidex-hub
      attach independently furnished benchmark results, which is what
      makes reported-vs-furnished divergence (W-EV-DIV-07) expressible.

Flag	Description
--format {text, markdown, json}	output format (default: text)
--output	write the report to FILE instead of stdout
--deterministic	id mode: skip the LLM and map the card with the README section/keyword scan (approximate; needs no model)
--revision	id mode: model-card git revision (default: latest)
--save-bundle PATH	id mode: write the generated bundle to PATH (default: ./__..jsonld)
--no-save-bundle	id mode: do not keep the generated bundle
--extract-backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	id mode: LLM backend for extraction (else README scan)
--extract-model	id mode: model name on the chosen backend
--extract-base-url	id mode: base URL for openai-compatible backends
--cou TEXT	state the context of use this assessment is scoped to (run-context; raises W-EV-COU-05 to Critical)
--mrl N	state the model risk level of the decision this evidence informs, 1-5 (run-context; enables COMPOUND-EV-01)
--raidex PATH	ingest a local raidex results.json as Group-B evidence
--raidex-hub	fetch this model’s published raidex record from the cloudronin/raidex-results dataset
--raidex-run	run raidex eval now and assess its output (needs: pip install uofa[raidex]; UNVERIFIED against a live run, see docs/live-run-verification.md)
--raidex-args ARGS	passthrough arguments for raidex eval, uninterpreted
--raidex-yes	skip the —raidex-run confirmation (for CI)

uofa validate

validate all examples against SHACL profiles

uofa validate [-h] [--no-color] [--verbose] [--repo-root PATH]
                     [--pack NAME] [--dir DIR] [--verify] [--pubkey PUBKEY]

Flag	Description
--dir	directory to scan (default: all pack examples)
--verify	also verify hash + signature integrity on each file
--pubkey	ed25519 public key (default: keys/research.pub)

uofa init

scaffold a new UofA project (template + keys + uofa.toml)

uofa init [-h] [--no-color] [--verbose] [--repo-root PATH]
                 [--pack NAME] [--profile {minimal,complete}] [--dir DIR]
                 [--provider PROVIDER] [--model MODEL]
                 name

Flag	Description
--profile {minimal, complete}	starting profile (default: complete)
--dir	parent directory (default: cwd)
--provider	LLM provider for uofa extract (default: ollama)
--model	LLM model for uofa extract (default: llama3.2)

uofa diff

compare weakener profiles between two UofA files (COU divergence)

uofa diff [-h] [--no-color] [--verbose] [--repo-root PATH]
                 [--pack NAME] [--build] [--skip-rules] [--explain]
                 [--explain-functions EXPLAIN_FUNCTIONS]
                 [--explain-format {text,json,markdown,html}]
                 [--explain-backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                 [--explain-model EXPLAIN_MODEL]
                 [--explain-base-url EXPLAIN_BASE_URL]
                 [--explain-max-items EXPLAIN_MAX_ITEMS] [--explain-no-cache]
                 file_a file_b

Flag	Description
--build	auto-build the Jena JAR if missing
--skip-rules	compare static hasWeakener arrays instead of running rules
--explain	run the interpretation pipeline after the primary analysis
--explain-functions	comma-separated list of interpretation functions to run (values: explain, group, contextualize, cross, narrative). Default: all applicable for the command.
--explain-format {text, json, markdown, html}	output format for the interpretation block. Default: same as the command’s primary —format, falling back to text.
--explain-backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	LLM backend for explain (overrides [llm] backend in uofa.toml)
--explain-model	model name on the chosen backend (overrides [llm] model)
--explain-base-url	base URL for openai-compatible backends
--explain-max-items	limit interpretation to top N items by severity
--explain-no-cache	bypass cached interpretation results

uofa explain

interpret pre-existing structured output from rules/check/diff/shacl

uofa explain [-h] [--no-color] [--verbose] [--repo-root PATH]
                    [--pack NAME] (--from-file FROM_FILE | --from-stdin)
                    [--input-type {rules,check,diff,shacl}]
                    [--functions FUNCTIONS] [--format {text,json,markdown}]
                    [--max-items MAX_ITEMS] [--no-cache]
                    [--backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                    [--model MODEL] [--base-url BASE_URL]

Flag	Description
--from-file	read structured output from FILE
--from-stdin	read structured output from stdin
--input-type {rules, check, diff, shacl}	override auto-detection of input type
--functions	comma-separated list of interpretation functions to run
--format {text, json, markdown}	output format (default: text)
--max-items	limit interpretation to top N items by severity
--no-cache	bypass cached interpretation results
--backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	LLM backend (overrides [llm] backend in uofa.toml)
--model	model name on the chosen backend
--base-url	base URL for openai-compatible backends

uofa schema

generate JSON Schema or import constants from SHACL shapes

uofa schema [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--output OUTPUT] [--emit {json,python}]
                   [--freeze VERSION]

Flag	Description
--output, -o	output path (default depends on —emit format)
--emit {json, python}	output format: json (JSON Schema, default) or python (import constants)
--freeze VERSION	freeze the generated schema to spec/schemas/.json with a versioned $id, e.g. —freeze v0.5. Frozen versions are published at https://uofa.net/schemas/.json for adopters to pin, and are never regenerated. Refuses to overwrite one.

uofa packs

list and inspect installed domain packs

uofa packs [-h] [--no-color] [--verbose] [--repo-root PATH]
                  [--pack NAME] [--detail]
                  [name]

Flag	Description
--detail	show detailed pack information

uofa catalog

list all weakener patterns across active packs

uofa catalog [-h] [--no-color] [--verbose] [--repo-root PATH]
                    [--pack NAME] [--format {table,json,md}] [--all-packs]

Flag	Description
--format, -f {table, json, md}	output format (default: table)
--all-packs	enumerate patterns across every installed pack (overrides —pack). Use this for a complete catalog view; the default —pack=vv40 only surfaces core + vv40 patterns.

uofa migrate

migrate UofA files from v0.3 to v0.4

uofa migrate [-h] [--no-color] [--verbose] [--repo-root PATH]
                    [--pack NAME] [--from FROM_VER] [--to TO_VER] [--dry-run]
                    file

Flag	Description
--from	source version (default: v0.3)
--to	target version (default: v0.4)
--dry-run	print changes without modifying the file

uofa import

import an Excel workbook into a UofA JSON-LD file

uofa import [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--output OUTPUT] [--sign] [--key KEY]
                   [--check] [--protocol-check] [--profile {minimal,complete}]
                   [--base-uri BASE_URI] [--evidence EVIDENCE]
                   [--sip-pubkey SIP_PUBKEY]
                   [--decision-pubkey DECISION_PUBKEY]
                   [file]

Flag	Description
--output, -o	output path (default: same directory, .jsonld extension)
--sign	sign the output after generation
--key, -k	path to ed25519 private key (required with —sign, or auto-detected from project)
--check	run all quality gates on the output
--protocol-check	gate the import on the reference-encoding conformance checks. Runs the workbook-side checks against the input and the package-side checks against the artifacts beside the output. Any failure exits non-zero.
--profile {minimal, complete}	override profile auto-detection
--base-uri	namespace to mint identifiers under, e.g. https://acme.example/uofa. Overrides [project] base_uri in uofa.toml. Defaults to a reserved example.org placeholder; uofa.net is refused.
--evidence	evidence sidecar from uofa evidence seal; its manifest, source pins and corroboration are folded in BEFORE hashing, so the seal sits inside the signature scope
--sip-pubkey	SIP measurement public key for verifying a SIP-bundle input (default: keys/research.pub)
--decision-pubkey	engineer public key for verifying a SIP-bundle hasDecisionRecord on import

uofa extract

extract assessment data from evidence documents into an Excel template

uofa extract [-h] [--no-color] [--verbose] [--repo-root PATH]
                    [--pack NAME] [--model MODEL] [--output OUTPUT]
                    [--glob GLOB] [--thinking]
                    [--prompt-version PROMPT_VERSION]
                    [--extract-backend {ollama,anthropic,openai,openai-compatible,bundled,mock}]
                    [--extract-model EXTRACT_MODEL]
                    [--extract-base-url EXTRACT_BASE_URL] [--protocol-check]
                    [--keyless]
                    [source ...]

Flag	Description
--model	legacy litellm model string (e.g. ‘ollama/qwen3.5:4b’). For new configs, prefer —extract-backend + —extract-model.
--output, -o	output Excel path (default: {source}-extracted.xlsx)
--glob	file filter pattern (e.g. ‘.pdf’ or ‘.pdf,*.docx’)
--thinking	enable thinking/reasoning mode (slower, may improve accuracy)
--prompt-version	tag for scoring log tracking (e.g. ‘v2-detailed’)
--extract-backend {ollama, anthropic, openai, openai-compatible, bundled, mock}	LLM backend for extract (overrides [llm] backend)
--extract-model	model name on the chosen backend (overrides [llm] model)
--extract-base-url	base URL for openai-compatible backends (e.g. Together AI, vLLM)
--protocol-check	report the reference-encoding conformance checks against the written workbook. Informational here: a fresh extract has no citation anchors, because anchors are what the review pass produces. The gate is on uofa import.
--keyless	extract without a language model: no API key, no network, no spend. Fills only the fields with a route measured to beat a null model, and leaves the rest blank rather than guessing

uofa adversarial

generate / orchestrate / classify synthetic credibility evidence packages

uofa adversarial [-h] [--no-color] [--verbose] [--repo-root PATH]
                        [--pack NAME]
                        {generate,run,analyze,prep-review,bundle,judge,triage,adjudicate}
                        ...

uofa setup

install or verify the LLM extract runtime (Ollama + qwen3.5:4b)

uofa setup [-h] [--no-color] [--verbose] [--repo-root PATH]
                  [--pack NAME] [--bundle BUNDLE]
                  [--create-bundle CREATE_BUNDLE] [--model MODEL]
                  [--port PORT] [--no-byo] [--yes] [--no-verify]
                  {verify,uninstall} ...

Flag	Description
--bundle	install from an offline bundle (REQ-DIST-004 air-gapped path)
--create-bundle	package the current install into a tar.gz for an offline machine
--model	model tag to pull (default: qwen3.5:4b)
--port	port for the managed Ollama daemon (default: 11434)
--no-byo	skip BYO Ollama detection; install a UofA-managed copy
--yes, -y	skip confirmation prompts (downloads, uninstall)
--no-verify	skip the post-install verify step

uofa demo

run the C1+C2+C3 pipeline against a bundled fixture (no setup needed)

uofa demo [-h] [--no-color] [--verbose] [--repo-root PATH]
                 [--pack NAME] [--no-passage] [--no-jsonld]

Flag	Description
--no-passage	skip printing the input passage (useful for piping)
--no-jsonld	skip printing the pre-computed JSON-LD

uofa interrogate

measure a surrogate and emit a signed evidence bundle (or init to set up)

uofa interrogate [-h] [--no-color] [--verbose] [--repo-root PATH]
                        [--pack NAME] [--adapter ADAPTER]
                        [--benchmark BENCHMARK] [--reference REFERENCE]
                        [--scope SCOPE] [--output OUTPUT] [--key KEY]
                        [--seed SEED]
                        {init} ...

Flag	Description
--adapter	ModelAdapter ref: ‘pkg.module.ClassName’ or ‘/path/file.py:ClassName’
--benchmark	benchmark inputs (.npz/.json)
--reference	reference outputs (.npz/.json) — supplied, never generated
--scope	declared-scope config (.json/.toml)
--output, -o	output path for the signed evidence bundle (.json)
--key, -k	ed25519 SIP measurement key (auto-detected from project keys/ if omitted)
--seed	seed recorded in measurement provenance

uofa decision

review a SIP comparison (read-only) or sign an engineer decision

uofa decision [-h] [--no-color] [--verbose] [--repo-root PATH]
                     [--pack NAME]
                     {review,record} ...

uofa define

look up what a vocabulary term means

uofa define [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--site] [--search TEXT] [--list]
                   [--all-packs] [--format {text,json}]
                   [term]

Flag	Description
--site	emit the namespace-grouped payload the site build consumes
--search TEXT	search labels and definitions for TEXT
--list	list the terms in scope
--all-packs	cover every installed pack, not just the active ones
--format {text, json}	output format (default: text)

uofa vocab

look up what a vocabulary term means

uofa vocab [-h] [--no-color] [--verbose] [--repo-root PATH]
                  [--pack NAME] [--site] [--search TEXT] [--list]
                  [--all-packs] [--format {text,json}]
                  [term]

Flag	Description
--site	emit the namespace-grouped payload the site build consumes
--search TEXT	search labels and definitions for TEXT
--list	list the terms in scope
--all-packs	cover every installed pack, not just the active ones
--format {text, json}	output format (default: text)

uofa guardrail

basic guardrail: turn weakener firings into a signed engineer-commanded action (§6)

uofa guardrail [-h] [--no-color] [--verbose] [--repo-root PATH]
                      [--pack NAME] [--key KEY]
                      [--threshold {Critical,High,Medium,Low}]
                      [--action {restrict,clip,refuse}] [--min-hits MIN_HITS]
                      [--output OUTPUT]
                      file

Flag	Description
--key, -k	ed25519 key to sign the action (auto-detected from project keys/ if omitted)
--threshold {Critical, High, Medium, Low}	engineer-commanded severity threshold to trigger (default: High)
--action {restrict, clip, refuse}	engineer-commanded response on trigger (default: restrict)
--min-hits	minimum hits for a firing to count toward the trigger (default: 1)
--output, -o	output path for the package + guardrailAction (default: in place)

uofa inject

inject a known flaw and record it in a manifest; inject verify scores the result

uofa inject [-h] [--no-color] [--verbose] [--repo-root PATH]
                   [--pack NAME] [--package PACKAGE] [--out OUT]
                   [--pattern PATTERN] [--operator OPERATOR] [--site SITE]
                   [--all] [--manifest MANIFEST]
                   {verify} ...

Examples:
  uofa inject --pattern W-AL-01 --package pkg.jsonld --out /tmp/m
  uofa inject --all --package pkg.jsonld --out /tmp/m
  uofa rules /tmp/m/<mutant>.jsonld
  uofa inject verify --manifest /tmp/m/manifest.json

Flag	Description
--package	substrate package
--out	output directory
--pattern	target pattern id (e.g. W-AL-01)
--operator	operator id (e.g. MUT-DEL-01); overrides —pattern
--site	site index (default: 0)
--all	every operator at every site
--manifest	manifest path (default: /manifest.json)

uofa evidence

inventory and seal a simulation evidence folder (no model, no network)

uofa evidence [-h] [--no-color] [--verbose] [--repo-root PATH]
                     [--pack NAME]
                     {inventory,seal} ...

Examples:
  uofa evidence inventory osf-n4pjz/
  uofa evidence inventory osf-n4pjz/ --members       # every archive member
  uofa evidence seal osf-n4pjz/ -o evidence.json
  uofa evidence seal osf-n4pjz/ --source-map osf-urls.txt -o evidence.json
  uofa import extracted.xlsx --evidence evidence.json --sign --key k.key