🔥 iscooked — Am I Cooked?

iscooked is a local security scanner for AI services, containers, and agent settings. It runs as one Bash script and keeps the report on your machine.

Start a scan

The scanner needs Bash 4 or later. Some macOS installations provide Bash 3.2. Check your Bash version first:

bash --version

To download, read, and run the scanner:

curl -fsSL https://iscooked.com/iscooked.com -o iscooked.com
less iscooked.com
bash iscooked.com

If you want one command after you trust the source:

curl -fsSL https://iscooked.com/iscooked.com | bash

You can also run a local checkout:

git clone https://github.com/johnpippett/iscooked.git
cd iscooked
bash iscooked --help
bash iscooked

Scoring

The terminal shows a cooked percentage, a 40-character meter, a verdict, and result counts. This example is synthetic. It does not scan your machine.

YOUR COOKED SCORE

20% cooked  [▒▒▒▒▒▒▒▒                                ]

SLIGHTLY WARM

0 critical  1 warnings  4 passed
4 unknown  9 skipped  (18 results)

Turn down the heat. Check the warnings above.

In this example, one warning and four unknown results add 20 points. Nine skipped results add no points. Start with critical findings. Then review warnings, unknown results, and skipped areas.

The score is a capped heuristic. It is not a probability of compromise.

Result	Terminal label	Points
Critical	🔥 COOKED	+10
Warning	⚠ WARMING UP	+4
Passed	✅ SAFE	+0
Unknown	❓ UNKNOWN	+4
Skipped	⏭ SKIP	+0

Score	Verdict
0–14	LOOKING FRESH
15–39	SLIGHTLY WARM
40–69	MEDIUM RARE
70–100	FULLY COOKED

A critical finding or warning can change a low-score verdict to SLIGHTLY WARM. A report with only unknown or skipped results can show STILL DEFROSTING. The scanner keeps unknown and skipped counts visible.

What it checks

The scanner has 16 check areas:

ID	Check area	Looks for
01	Network Exposure	Common AI service ports and listener bind addresses.
02	API Authentication	Authentication on selected model-list routes for identified Ollama, LM Studio, and vLLM services.
03	Model File Permissions	World-readable or world-writable files in common model directories.
04	Docker / Container Risks	Root users, privileged mode, host networking, daemon sockets, and host-root mounts.
05	GPU Driver Exposure	GPU device permissions and known management listeners.
06	Telemetry / Phoning Home	Scanner opt-out values and exact known entries in /etc/hosts.
07	Firewall Status	Recognized Linux and macOS firewall backends. A missing firewall is critical only when an AI service listens beyond localhost; otherwise it is a warning.
08	SSL/TLS Configuration	Plain HTTP on non-loopback AI service listeners.
09	AI Process Enumeration	Candidate AI processes and the account that runs each process.
10	Sensitive File Exposure	World-readable .env files that contain common API key names.
11	History & Logs Leakage	API key patterns in shell history and permissions on common AI log directories.
12	Ollama-Specific Checks	OLLAMA_HOST, OLLAMA_ORIGINS, and selected service settings.
13	Browser Remote Debugging	Chromium-family debugging flags, listeners, and /json/version metadata.
14	MCP Configuration	Selected Model Context Protocol client files, filesystem grants, and permissions.
15	Agent Gateway Configuration	Selected OpenClaw settings when the installed CLI has a reviewed version.
16	Remote Model Code	Remote-code flags in recognized vLLM and Text Generation Inference launches.

Unknown and skipped results

UNKNOWN means that the scanner could not establish a state. SKIP means that it did not examine an area. A missing optional tool or unsupported setup can cause a skip. A failed observation can cause an unknown result.

Unknown results add four points. Skipped results add no points. Both results limit coverage. Use the unknown-results guide for diagnostic commands and manual checks for firewall, Docker, network, browser, file, agent, model, and environment results.

Requirements

Type	Requirement
System	Linux or macOS.
Shell	Bash 4 or later. Some macOS installations provide Bash 3.2.
Required commands	awk, basename, cat, find, grep, ps, stat, tr, uname, wc, and whoami.
Home directory	HOME must name an existing absolute directory.
Socket evidence	ss or netstat. Linux falls back to /proc/net/tcp when both are missing. Failed inspection produces unknown results.
Optional checks	Python 3 enables JSON and structured API, browser, MCP, agent-gateway, and model-code checks. curl, docker, and nvidia-smi enable related probes.
Docker timeout	Docker metadata calls use timeout, gtimeout, or Python 3 for a five-second limit.

Run bash --version to examine the shell version. Elevated privileges can improve some firewall and port checks.

CLI

For a downloaded scanner, use bash iscooked.com [options]. For a local checkout, use bash iscooked [options].

Option	Action
-h, --help	Show help without a scan.
--version	Show the scanner version without a scan.
--json	Write one JSON report. Requires Python 3.
--no-color	Disable terminal colors.
--fail-on LEVEL	Select the findings that produce exit status 1. Use critical, warning, or unknown.

The scanner also disables colors when output is redirected, when NO_COLOR is set, or when TERM=dumb.

JSON reports

--json writes one report to standard output. It uses schema_version: 1 and sets completed: true after the report is complete. A report contains scanner metadata, the platform, summary counts, score data, coverage counts, and a findings array.

Each finding contains a check ID and title, a status, a message, and points. Check IDs identify areas. They do not identify stable rules. Messages can change between releases. The score includes a capped value and an uncapped raw_value; its kind is heuristic and it includes unknown results.

Coverage counts distinct areas with observations, unknown results, and skipped results. These groups can overlap. An area can start without complete coverage. Reports can contain local paths and account names. Examine a report before you share it.

Example commands:

bash iscooked.com --help
bash iscooked.com --version
bash iscooked.com --no-color
bash iscooked.com --json > report.json
bash iscooked.com --json --fail-on critical > report.json

Exit status

Without --fail-on, a completed scan returns 0 regardless of its findings. With a failure policy, the scanner returns 1 when the selected results occur:

Policy	Exit status 1 for
critical	Critical findings.
warning	Critical findings or warnings.
unknown	Critical findings, warnings, or unknown results.

Skipped results do not trigger these policies. Invalid options, missing required tools, unsupported systems, invalid HOME, and unsupported requirements return 2 before the scan. A runtime error can stop a scan before a complete JSON report. JSON consumers must require completed: true.

Coverage limits

The scanner reads local files, process information, listener data, and selected service metadata. It does not change settings, start an MCP server or agent, or load a model. It does not execute model code, retrieve browser tabs or cookies, or connect to a returned WebSocket URL.

Common ports, process names, container names, and paths are identification hints. They do not prove internet reachability or confirm every application identity. A non-loopback listener shows local network exposure. It does not prove internet exposure.

API checks cover /api/tags on port 11434 and /v1/models on ports 1234 and 8000. The scanner confirms service identity before it reports API access. It tests one route at a time, does not follow redirects or configured proxies, and uses a five-second total timeout.

Most file checks use Unix mode bits. They do not establish access through every parent directory, access control list, or service boundary. The sensitive-file search stops after 20 files in each bounded search.

The telemetry check reads values in the scanner environment and known host mappings. It does not establish the environment of a running service or analyze outbound traffic. OLLAMA_NO_CLOUD false values do not establish an opt-out. The Ollama check separates scanner variables from active service settings.

MCP and OpenClaw checks cover selected files and supported JSON settings. For OpenClaw, the scanner runs openclaw --version with a 1.5-second limit. It accepts at most 256 output bytes. Versions 2026.9.3 through 2026.9.6 can use the selected config checks. A command outside the scanner's fixed search path, a failed command, or another version gives UNKNOWN. The CLI version does not prove the version of a running gateway. JSON5, includes, interpolation, overrides, and policy enforcement remain incomplete.

Remote model-code checks cover recognized vLLM and Text Generation Inference command lines. They do not inspect configuration files, environment variables, or other runtimes. A model weights revision does not prove an executable code pin.

Docker checks inspect known socket paths, the active Unix daemon endpoint, and host-root mounts. A read-only Docker socket mount can still permit daemon API calls. Rootless, proxy, and uncertain access can remain warning or unknown results.

Privacy

After download, the scan sends no scan report or telemetry to iscooked.com or another remote service. It probes selected local AI and browser endpoints. Reports can contain local paths and account names, so examine them before sharing.

Develop and test

Run the test suite and the release checks from the repository root:

python3 -m pytest -q
bash -n iscooked
bash -n site/iscooked.com
cmp iscooked site/iscooked.com
python3 scripts/sync_version.py --check
git diff --check

The tests use synthetic fixtures and isolated state. They do not prove behavior on every Linux or macOS installation.

To preview the local website, run:

python3 scripts/preview.py

Then open http://127.0.0.1:8794. The preview serves site/ and does not run the scanner.

VERSION is the release version source. Run python3 scripts/sync_version.py without an argument after editing VERSION. Pass a stable MAJOR.MINOR.PATCH value to update VERSION, both scanner copies, website labels, and this README badge. Run python3 scripts/sync_version.py --check to detect drift without writing files.

License

MIT License

iscooked.com · GitHub repository