Repository navigation
Releases
OniOps v0.4.9-beta
OniOps can now be installed from PyPI, web UI included.
pipx install oniops # or: pip install oniops (inside a venv)
oniops demo # one-minute practice run, no config needed
oniops init # creates ~/.oniops/config.yaml and .env with a random secret key
oniops # web UI on http://127.0.0.1:8501Added
pip install oniops/pipx install oniopsnow gives a working install, including the web UI. The package bundles the web app, templates, static files, playbooks and the example config.oniops initcreatesconfig.yaml,.env(randomONIOPS_SECRET_KEY, mode 600) anddata/in the OniOps home folder. It never overwrites existing files and never prints the key.ONIOPS_HOMEsets where an installed copy keeps its config and data (default~/.oniops). Source checkouts, Docker and the systemd setup are unchanged.- CI now builds the package, installs it outside the repository and runs
init,demoand the web UI. - Releases publish to PyPI through trusted publishing; no API token is stored.
Changed
- README rewritten in plainer terms, with a "Where OniOps fits" section that points to more mature tools and a "Known limitations" section.
- All links now use github.com/ares-369/oniops.
Known limitations
- No scored benchmark results are published yet.
oniops demoruns against a small bundled practice target; it shows the workflow, not real-world detection.- The pip install is new and so far tested on Linux only.
This is beta software for authorized testing only. See DISCLAIMER.md.
ONI-Ops v0.4.8-beta
ONI-Ops v0.4.8-beta — pre-release (beta)
Security fixes, authenticated target sessions, and an easier install/deploy.
Beta software for authorized security testing only; see DISCLAIMER.md and NOTICE.
Security
- Reverse-proxy identity header could be forged. The web UI runs behind
Werkzeug's ProxyFix, which rewrites REMOTE_ADDR from X-Forwarded-For. In
auth.mode: proxy the trusted-proxy check compared that rewritten value, so a
client able to reach the port directly could send X-Forwarded-For set to the
trusted proxy plus the identity header and authenticate as any user. The check
now validates the real socket peer, not the forgeable header. - Running
python webui.pydirectly started the unhardened Flask dev server and
skipped the startup security check. The entry point now routes through
oni.server.main() (Waitress plus the exposure/auth checks), same as
python -m oni.
Added
- Authenticated target sessions with refresh. A target_auth config block logs in
to a target per role (primary/low_priv/peer): json_login (token APIs such as
VAmPI and Juice Shop) or form_login (form logins such as DVWA). A role's
session is re-minted and the request retried once on a 401/403, so short-lived
tokens (for example VAmPI's ~60s JWTs) no longer collapse the access-control
differential into anonymous requests. Credentials are referenced only as
${ENV_VAR} and never reach findings, logs or reports. See docs/TARGET_AUTH.md. - One-command install (scripts/install.sh): creates a virtualenv, installs
ONI-Ops, seeds config.yaml/.env from the examples without overwriting them, and
generates a random ONI_SECRET_KEY when missing. Idempotent. - Public-app benchmark harness:
oni benchmark --suite <crapi|juice_shop|vampi> --target <ip>scores an assessment against labeled known bugs. Only verified
findings count; precision is a range until untriaged findings are classified.
Fixed
- The web UI now honours ONI_DB_PATH, so it uses the same ledger as the CLI.
- Browser discovery redacts bare jwt/bearer/credential query values and pins the
start host to its scope-validated IP at launch, closing the DNS re-resolution
window between the scope check and navigation. - The labeled-benchmark scorer only lets a VERIFIED finding satisfy a label, so
an unverified match stays visible for triage and false-positive accounting. - AWS region validation accepts GovCloud, China and ISO partition names.
Changed
- Easier, safer install and deploy: the Docker image ships playbooks/ and curl;
the worker image ships katana and a wordlist; Docker Compose requires a login;
the systemd unit runs as a dedicated non-root user from /opt/oni; the Proxmox
LXC script points at the real repository; known_hosts defaults to the writable
data directory. CLI:oni assessrejects an unknown profile instead of
silently starting the autonomous loop, commands print clean errors instead of
tracebacks, andoni doctor --check-runnerfails when the runner is
unreachable.
Upgrade note
- No config change is required. To use authenticated target sessions, add a
target_auth block (see docs/TARGET_AUTH.md) and set the referenced ${ENV_VAR}
credentials in your environment.
ONI-Ops v0.4.7-beta
ONI-Ops v0.4.7-beta — pre-release (beta)
Security, assessment quality and deployment improvements. Beta software for
authorized security testing only; see DISCLAIMER.md and NOTICE.
Added
oni benchmark --demo-evaluation: a labeled, loopback-only fixture evaluation
with precision/recall/F1, negative checks, one-to-one matching and stated
fixture-only limits. Unverified detections do not count as true positives;
duplicate alerts count as false positives.- Optional Playwright browser discovery (
oni crawl) with bounded page/time
budgets, in-scope request routing, read-only navigation and start-origin-only
credential forwarding. It reports observed XHR/fetch status, content type and
value-free form metadata. - Disposable Docker scanner workers with a non-root user, read-only filesystem,
dropped capabilities, resource limits, timeouts and a fail-closed egress
firewall. CI exercises allowed and denied egress and verifies Nmap/Nuclei
startup. SeeDockerfile.worker. - Offline Trivy scanning of images on the attack host (
oni container-scan) and
optional read-only AWS account, IAM, security group, EKS and S3 checks
(oni cloud-audit aws). - Versioned network playbooks and AI probe packs with nested discovery and source
hashes. Seedocs/SECURITY_PACKS.md. - Expanded OpenAPI BOPLA analysis for nested local references, compositions,
arrays, OpenAPI 2 body/form parameters and OpenAPI 3 vendor schemas.
Fixed and hardened
- Horizontal BOLA now requires evidence of object ownership; a user's own object
read is not reported as a second verified cross-user finding. - Agent finding replay and zero-token retests enforce supported proof rules,
reject malformed or out-of-origin paths, cap responses, disable redirects,
and report 5xx outcomes as inconclusive. - Scope checks now validate full IPv4 CIDRs and ranges, including exclusions.
- Network playbook steps are limited to supported scanner commands and must use
the assessed{target}. Intrusivesqlmapandhydrapacks require explicit
safety: intrusivescope configuration. - AWS security-group rules without TCP/UDP ports, such as ICMP, are no longer
treated as exposing all ports.
Upgrade note
- The Docker worker requires a user-defined bridge network. IPv4-only kernels
are supported; IPv6 scopes fail closed when IPv6 is unavailable.
ONI v0.4.4-beta
ONI v0.4.4-beta is a security hardening release. Recommended for everyone, and important if you use auth.mode: proxy.
Security
- Reverse-proxy login could be forged. With
auth.mode: proxyand an emptyproxy_trusted_ips— the example config's default — any client could send the identity header and log in as any user. Proxy mode now fails closed: ONI refuses to start without a trusted-proxy list, and the request handler also rejects the header when no list is set. - Secure by default. The web UI now listens on
127.0.0.1by default, and ONI refuses to start when it would be reachable from the network without a login (unless you explicitly opt in withoni.allow_unauthenticated_network: true). Docker Compose now publishes the port on127.0.0.1only. - Stronger prompt-injection filter. Tool output is now Unicode-normalised and stripped of invisible characters before scanning, base64-encoded instructions are decoded and checked, and many more phrasings and fake role tags are caught — 17 of 17 cases in a new adversarial test suite, up from 6, without altering normal output.
- The same suite verifies the layer that matters most: even a model that fully obeys an injected instruction still cannot scan outside your scope or run arbitrary or destructive commands. A pattern filter can never catch every paraphrase, so this second layer is the real protection. See SECURITY.md.
Upgrade notes
- If you open ONI from other machines, set
oni.bind_host(for example"0.0.0.0") and enable a login withauth.mode(basic, local, oidc or proxy) — otherwise ONI will not start and will tell you why. - If you use
auth.mode: proxy, setauth.proxy_trusted_ipsto your reverse proxy's IP. - Docker users: the UI is now at
http://localhost:8501on the Docker host only. See the comments indocker-compose.ymlto expose it safely.
Full details: CHANGELOG.md.
Please note
ONI is beta software. Treat findings as leads to verify, and open an issue if something looks wrong. ONI is for testing systems you own or are explicitly authorized to test; see DISCLAIMER.md.
ONI v0.4.3-beta
ONI v0.4.3-beta is a correctness and security fix release. If you run 0.4.1-beta or 0.4.2-beta, please upgrade.
Fixed
- Retests could report a still-present vulnerability as FIXED. Since 0.4.1-beta, credentials inside replay commands were masked in the database itself. Retesting an authenticated finding (for example an authenticated BOLA) therefore sent
[REDACTED-TOKEN], received a 401, and marked the issue as fixed. Replay commands are now stored intact, and secrets are masked wherever a finding leaves the server (UI, API, CLI and retest responses).- Findings saved by 0.4.1–0.4.2 whose credential was already masked will refuse to retest, with a clear message, instead of guessing. Re-run the assessment with the auth header to retest them.
- No more "FIXED" from failed retests. A 401 on an authenticated retest now reports UNCONFIRMED (the credential may have expired). If a retest cannot run at all (attack host unreachable, timeout, scope refusal), the finding is left unchanged.
- Access control. The full database backup (
/api/backup/download) and the attack-host SSH check (/api/doctor?runner=1) are now admin-only. Previously the read-onlyviewerrole could download the entire database.
Changed
- The "benchmark" added in 0.4.1-beta is renamed to validator self-test (
oni selftest,GET /api/selftest; the old names still work). It checks ONI's own validators against built-in example inputs and reports pass/fail per check. It no longer reports precision, recall or F1: those figures described the built-in examples, not how well ONI finds real vulnerabilities. A benchmark against real vulnerable applications is planned.
The 0.4.1-beta and 0.4.2-beta changelog entries have been annotated with these corrections. Full details: CHANGELOG.md.
Please note
ONI is beta software. Treat findings as leads to verify, and open an issue if something looks wrong. ONI is for testing systems you own or are explicitly authorized to test; see DISCLAIMER.md.
v0.4.2-beta
Fixed
- Workbench Recent Jobs bar was always empty (read
paramsinstead of jobmeta); it now also refreshes when the Workbench opens. - Replay commands could expose bearer tokens/cookies in the Finding Detail Drawer; repro commands are now secret-redacted on storage and on display.
- Zero-token retest now treats 401/403/404 as FIXED for BOLA/BFLA findings.
- Clicking a findings row now opens the Finding Detail Drawer reliably; long SHA-256 hashes wrap on mobile.
Correction (added in v0.4.3-beta): masking credentials in stored replay commands, and treating every 401 on a BOLA/BFLA retest as FIXED, could make retests report still-present vulnerabilities as fixed. Both are corrected in v0.4.3-beta. Please upgrade.
ONI v0.4.1-beta
Added
- Finding Detail Drawer & Fix-First Priority Score: Clicking any finding in the ledger opens a slide-over drawer showing the full captured evidence, SHA-256 provenance hash, deterministic replay command (with one-click copy and zero-token retest), CWE / OWASP / ATT&CK tags, step-by-step engineering remediation guidance (
get_remediation_guide), and a0-100Fix-First Priority Score (compute_priority_score, inspired by Wiz / Snyk risk prioritization). AddedGET /api/findings/<id>. - Verification & False-Positive Benchmark Suite (
oni benchmark): Built-in 17-case benchmark suite (oni/benchmark.py,python -m oni benchmark, andGET /api/benchmark) measuring true-positive recall, false-positive rate (FPR), precision, and F1 score across Web/API proof validators, BOLA/BOPLA analyzers, SAST sink detection, and prompt-injection quarantine. - Authenticated Assessment Enhancements:
whatweband PoC service validation now honor the grey-boxauth_headerwhen testing behind login gates, and agent-reported findings persist secret-redacted-Hreplay commands and vulnerability categories. - Full Backend-to-Frontend Parity: Wired Assessment Baseline Drift Diff (
/api/assessments/<id>/diff), AI YAML Probe Packs & Crescendo toggles (/api/aisec/probes), White-Box SARIF v2.1.0 download & Semgrep/Gitleaks toggle (/api/whitebox/scan), Settings Self-Diagnostics (GET /api/doctor), in-browser Verification Benchmark runner, online SQLite.sqlitebackup download (GET /api/backup/download), Horizontal BOLA Peer Header (peer_auth_header), and Background Job History / Re-attach (/api/jobs) directly into the web UI.
Correction (added in v0.4.3-beta): the "benchmark suite" described above does not measure recall, false-positive rate, precision or F1 on real targets. It checks ONI's own validators against 17 built-in example inputs, and has been renamed to the validator self-test (
oni selftest). Also, masking credentials in stored replay commands (introduced here) could make retests report live issues as FIXED; this is fixed in v0.4.3-beta. Please upgrade.
Note: ONI is pre-1.0 beta software for authorized security testing only. See DISCLAIMER.md.
ONI v0.4.0-beta
ONI v0.4.0-beta focuses on making results easier to trust and the first run easier to get through.
Highlights
- Confidence on every finding. The findings ledger now shows how sure ONI is that each finding is real: Certain (verified with proof from the target), Firm (strong signal) or Tentative (a lead to verify by hand). Filter by confidence, and the most serious, most certain findings sort to the top. Inspired by Burp Suite's issue confidence.
- Step-by-step job timeline. Assessments in the Workbench now show one row per stage (port scan, discovery, CVE lookup, AI report, ...) with live status, duration, and an expandable log, instead of a wall of text. Inspired by the GitHub Actions run view.
- "Get ONI ready" checklist. New installs get a short checklist on the dashboard — accept the terms, connect an attack host and an AI model, check the scope, discover targets, run a first assessment — with a button for each step. It hides itself once everything is done. Inspired by Grafana and Tailscale onboarding.
- Helpful empty states. Empty screens now explain what belongs there and offer the next step, instead of showing a blank table. Inspired by Stripe and Notion.
Also since v0.3.2-beta
- First-run "Authorized use only" acceptance in the web UI, enforced by the server and recorded per user.
- A clear responsible-use disclaimer in
DISCLAIMER.md,NOTICEand the README. - Pages no longer wait ~6 seconds on SSH when the attack host is offline (load time 7.2s to 1.1s), and the host status is now consistent everywhere.
- Accessibility improvements: labels for icon buttons and form fields, and larger touch targets on phones.
Full details are in the changelog.
Getting started
git clone https://github.com/ares-369/oni.git
cd oni
cp config.example.yaml config.yaml # set your scope and attack host
cp .env.example .env
mkdir -p secrets && cp /path/to/attack_host_key secrets/kali_id_rsa && chmod 600 secrets/kali_id_rsa
docker compose up -d # or see docs/INSTALL.md for LXC, Windows, systemdPlease note
This is beta software. Many features are unit-tested but have had limited real-world testing, so treat findings as leads to verify, and open an issue if something looks wrong.
ONI is for learning and for testing systems you own or have explicit written permission to test. It is provided as-is; you are responsible for how you use it. See DISCLAIMER.md.