registry / @aspectlylabs/mikka / 0.0.306
@aspectlylabs/mikka@0.0.306
Local Mikka daemon: runs agents through the user's own CLIs
AI Security Review
scanned 2d ago · by lpm-firewall-aiThe scanner’s shell, process, environment, and network signals correspond to the daemon’s agent execution and model credential broker. The reviewed broker sends keys to the configured provider or official provider endpoint, with HTTPS validation and redirects disabled.
Static reason
One or more suspicious static signals were detected.
Trigger
A user starts an approved agent run using a model credential.
Impact
Model prompts and the supplied API key reach the selected model provider as part of the run; no unrelated credential exfiltration or hidden collection was identified.
Mechanism
The daemon brokers model requests over a local socket to the configured HTTPS provider or the official Anthropic/OpenAI API endpoint.
AI rationale
The scanner’s shell, process, environment, and network signals correspond to the daemon’s agent execution and model credential broker. The reviewed broker sends keys to the configured provider or official provider endpoint, with HTTPS validation and redirects disabled.
Evidence
main.jsagents/native-container.mjspackage.json
Network endpoints2
api.anthropic.comapi.openai.com
Decision evidence
public snapshotAI called this Clean at 94.0% confidence as Benign with low false-positive risk.
Evidence for AI clean decision
- The manifest exposes a CLI and module entry points, with no install lifecycle script listed.
- Custom model provider URLs must use HTTPS and cannot include embedded credentials, query strings, or fragments.
- The API key broker uses the configured provider URL or the official Anthropic/OpenAI API base and sends requests there with redirects disabled.
- A model credential is used when starting a run that passes a tool approval gate.
- The native runner rejects real provider keys in its environment and accepts only broker placeholders.
Evidence against
- The package runs agents through external CLIs and includes expected process execution, credential brokering, and network requests for that purpose.
- The credential broker and native runner use local sockets to pass model requests; the reviewed flow does not show credentials being sent to an unrelated collector.
Behavioral surface
ChildProcessCryptoEnvironmentVarsFilesystemNetworkShell
Source & flagged code
6 flaggedgithub-credentials.mjsView file
4import { randomUUID } from 'node:crypto'
L5: import { spawn } from 'node:child_process'
L6: import { pathToFileURL } from 'node:url'
High
Child Process
Package source references child process execution.
github-credentials.mjsView on unpkg · L4mikka.jsView file
4562ai: "7.0.142",
L4563: execa: "10.1.1",
L4564: pino: "10.4.0",
High
6339assertSafeId("run id", runId);
L6340: const common = resolve4(await worktreeGitDir(repoPath, worktree), "..", "..");
L6341: if (basename(common) !== ".git") throw new Error(`${repoPath} is not a standard clone`);
...
L6350: }
L6351: async function pinnedGit(repoPath, worktree, env = process.env) {
L6352: const gitDir = await worktreeGitDir(repoPath, worktree);
L6353: const workTree = realpathSync4(worktree);
L6354: return (args, extra, options) => execa3("git", [...SAFE_GIT, ...args], {
L6355: cwd: workTree,
Findings
5 High2 Medium5 Low
HighChild Processgithub-credentials.mjs
HighShellmikka.js
HighSame File Env Network Executionmikka.js
HighCredential Redirect Persistencemain.js
HighCommand Output Exfiltrationagents/native-container.mjs
MediumNetwork
MediumEnvironment Vars
LowFilesystem
LowHigh Entropy Strings
LowUrl Strings
LowCopyleft License
LowSuspicious Lifecycle Evidencepackage.json
Affected versions and remediation
This report applies to @aspectlylabs/mikka@0.0.306.
See all versions of @aspectlylabs/mikka for other recorded verdicts.
- Review the evidence and your use of @aspectlylabs/mikka@0.0.306 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.