CLI Reference (oss)
The `oss` command-line client for Aegis Sovereign — AI-safety enforcement, red-team certification, model clearance, approvals, SLOs, regulatory reports, and audit-chain anchoring, all from your terminal or CI pipeline.
Overview
The oss CLI ships with the Aegis Sovereign Python SDK and talks to the platform API. It is the fastest way to wire AI-safety gates into CI/CD: most commands exit non-zero on a blocking condition, so they double as pipeline gates. Two command groups are covered here — oss safety (the AI Safety Platform) and oss audit (audit log + immutable-ledger anchoring).
1# The CLI is installed with the Python SDK
2pip install aegissovereign
3
4# Configure endpoint + token (or use OSS_API_URL / OSS_API_TOKEN env vars)
5oss login --api-url https://sovereign.yourcompany.com --token $OSS_PAT
6
7oss --help
8oss safety --helposs safety — AI Safety Platform
Every safety capability is reachable from the CLI. Validation, clearance, and agentic checks exit 1 when blocked so they can gate a build.
1# Block the build if a prompt template trips the safety engine
2oss safety validate "$(cat prompt-template.txt)" --target input
3
4# Gate deployment on model clearance (exit 2 = not deployable)
5oss safety clearance gate my-fine-tune --workspace ws-prod
6
7# Run red-team certification with a domain library, fail on any miss
8oss safety redteam run --domain financial --strict --output json --output-file cert.json| Command | What it does |
|---|---|
| `oss safety validate "<text>"` | Run text through the Safety Policy Engine (regex + ML ensemble). Exits 1 if blocked. `--target output`, `--dry-run`, `--output json` |
| `oss safety rules` | List active safety rules. Filter with `--category` / `--severity` |
| `oss safety classifier status` | Show the active ML classifier ensemble; warns if NullClassifier (regex-only) is active |
| `oss safety classifier test "<text>"` | Run text through the classifier locally |
| `oss safety mask "<text>"` | Redact PII in place (email, SSN, card, secrets). `--engine presidio` |
| `oss safety image <file>` | Scan an image for unsafe content (multimodal). Exits 1 if blocked |
| `oss safety conversation <file.json>` | Detect crescendo / persistence / refusal-bypass across a multi-turn conversation |
| `oss safety redteam run` | Run the red-team certification suite. `--domain medical|financial|legal`, `--strict`, `--output json` |
| `oss safety redteam multi-turn` | Run the built-in multi-turn attack chains |
| `oss safety clearance gate <model>` | Pre-deployment clearance gate (bias + robustness + safety). Exits 2 if NOT deployable |
| `oss safety clearance evaluate|list` | Recompute/persist or list ModelSafetyRegistry clearance verdicts |
| `oss safety agentic tools` | List registered tool-call safety policies (risk tier, approval, allow-list) |
| `oss safety agentic validate <tool> --args '{…}'` | Validate a tool call through the agentic guard. Exits 1 if blocked |
| `oss safety agentic delegation <from> <to>` | Validate an agent→agent delegation against the trust boundary |
| `oss safety slo list|status <id>` | List safety SLOs or show burn-rate + error budget for one |
| `oss safety review list|decide` | List pending approvals or record an approve/reject decision (audit-logged) |
| `oss safety integrations list|test <provider>` | List configured integrations or send a test event |
| `oss safety report generate <framework>` | Generate a safety report (eu_ai_act, nist_ai_rmf, fedramp_poam, sr_11_7) |
oss audit — Audit log & ledger anchoring
The oss audit group reads the tamper-evident per-workspace audit chain and manages external anchoring (see the Audit-Chain Anchoring doc). verify-anchor exits 2 if the live chain head no longer matches the latest external anchor — i.e. tampering.
oss audit anchor ws-prod # publish the current chain head
oss audit anchors ws-prod --limit 20 # list past anchors + receipts
oss audit verify-anchor ws-prod # exits 2 if the chain was altered after anchoring| Command | What it does |
|---|---|
| `oss audit tail` | Show recent audit entries (`--follow` to stream) |
| `oss audit export` | Export audit entries to CSV |
| `oss audit anchor <workspace>` | Publish the current chain head to the external immutable ledger |
| `oss audit anchors <workspace>` | List past anchors and their ledger receipts |
| `oss audit verify-anchor <workspace>` | Recompute the live head and compare to the latest anchor. Exits 2 on mismatch |
Exit Codes
Commands are designed to gate pipelines. 0 always means pass/allowed.
| Code | Meaning |
|---|---|
| `0` | Pass / allowed / consistent |
| `1` | Blocked — validation, agentic, image, or red-team failure |
| `2` | Gate not met — clearance not deployable, or anchor mismatch (tampering) |