live · scanning repos
Scanrepo
github.com
github.com / JustVugg

JustVugg/colibri

C·1697 files·commit 0a5504a·scanned 1d ago·cached ✓
15/100
LOW RISK
Minor findings consistent with the project type. Nothing reachable from install hooks.

score capped at 15 — 49,172+ stars — findings likely false positives

verdict accurate?
49,172 stars
40% file coverage
Partial architecture graph
Desktop app detected
threat-state: lowlive

FINDINGS ░▒▓

warningSuspicious Windows batch commandBatch file contains commands commonly used to download and execute remote payloads (powershell, certutil, bitsadmin, curl to pipe, etc.).c/coli.cmd
rem thing to type: `coli chat --model D:\models\glm52_i4` works from cmd and
rem PowerShell exactly as `./coli chat` does elsewhere.
setlocal enabledelayedexpansion
set "COLI_HERE=%~dp0"
+5
warningPython subprocess / os.system usageos.system() or subprocess with shell=True / suspicious commands can execute arbitrary shell commands. Plain pip/package installs are usually benign.c/tests/test_bench_ux_failure.py
            env = dict(os.environ, REPS="1", TMPDIR=temporary)
            result = subprocess.run(["bash", str(SCRIPT), str(engine)], env=env,
                                    text=True, capture_o...
+4
warningPython subprocess / os.system usageos.system() or subprocess with shell=True / suspicious commands can execute arbitrary shell commands. Plain pip/package installs are usually benign.c/tests/test_bench_ux_greedy.py
            env = dict(os.environ, REPS="2", COLI_TEMP="0.8", TEMP="0.7", DRAFT="3", RECORD=str(output))
            result = subprocess.run(["bash", str(SCRIPT), str(engine)], env=env,
              ...
+4
criticalPython dynamic code executioneval(), exec() or compile() on untrusted input can execute arbitrary code. Common in Python malware for running obfuscated payloads.c/tests/test_coli_markdown.py
    namespace = {"sys": sys}
    exec(compile(ast.Module(body=nodes, type_ignores=[]), str(CLI), "exec"), namespace)
    return namespace
+7
criticalPython dynamic code executioneval(), exec() or compile() on untrusted input can execute arbitrary code. Common in Python malware for running obfuscated payloads.c/tests/test_convert_positioned_write.py
    namespace = {"os": os_module, "_positioned_write_lock": threading.Lock()}
    exec(compile(ast.Module(body=[helper], type_ignores=[]), str(CONVERTER), "exec"), namespace)
    return namespace["_po...
+7
warningPython subprocess / os.system usageos.system() or subprocess with shell=True / suspicious commands can execute arbitrary shell commands. Plain pip/package installs are usually benign.c/tests/test_fp8_repack.py
        outdir = os.path.join(self.tmp.name, "out_dry")
        rc = os.system(
            f'{sys.executable} "{os.path.join(os.path.dirname(__file__), "..", "tools", "repack_fp8_passthrough.py")}" '...
+4
criticalPython dynamic code executioneval(), exec() or compile() on untrusted input can execute arbitrary code. Common in Python malware for running obfuscated payloads.c/tests/test_inkling_converter_lock.py
    namespace = {"os": os, "sys": sys}
    exec(compile(ast.Module(body=[helper], type_ignores=[]), str(CONVERTER), "exec"), namespace)
    return namespace["acquire_output_lock"]
+7
infoSuspicious files are not reachable from entry pointsFlagged files exist but are not imported by any entry point. They may be dead code, tests, or attack payloads triggered by another mechanism.+2
warningHigh-entropy string literalsFound 4 long strings with high Shannon entropy. This is common in obfuscated payloads that hide URLs, keys, or bytecode.c/resource_plan.py+3
infoSuspicious file in repoExecutable file (START-HERE.bat) in repositorySTART-HERE.bat+2
infoSuspicious file in repoExecutable file (build_cuda.bat) in repositoryc/build_cuda.bat+2
infoSuspicious file in repoExecutable file (coli.cmd) in repositoryc/coli.cmd+2

Scores are heuristics. A “safe” verdict means no known-malicious patterns were found — clever malware can look boring. Wrong verdict? Flag it above; confirmed false positives become regression tests.

682 files scanned @ 0a5504a | 11/9/2026 | heuristic scan — always review manually

risk by category
code execution25
network & exfiltration0
file system access0
obfuscation3
supply chain12
owasp / injection0
telemetry
files 682/1697rules hit 12engine v10commit 0a5504a

github

JustVugg/colibri

A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold ones streamed from NVMe. Written in C, runs on the hardware you already own.

C
49172
5766
132d
1697 files
682 scanned(40%)
0a5504a

network indicators░▒▓

Hosts this code names or would connect to. 41 found, decoded and call-site destinations first. Informational, not scored.

architecture░▒▓

entry (1) flagged (13) pkg (207)
607 nodes · 2058 edgesscroll to zoom · click node to jump to finding
LOW (15/100) — JustVugg/colibri — ScanRepo