apiblaze@0.21.47
APIblaze CLI — Chat with your APIs, Manage your API keys, users and groups with the APIblaze serverless proxy
AI Security Review
scanned 21d ago · by lpm-firewall-aiSource inspection found a guarded Codex MCP setup that can refresh after a user has installed it, which merits a warning under the agent integration policy. The manifest has no install hook, and the inspected credential transfers are tied to user-supplied chat destinations and actions.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
A user selects the Codex integration or runs apichat with --install-mcp codex; a saved installation can be refreshed on later CLI runs.
Impact
Codex can connect to the configured APIblaze proxy. This is a guarded first-party integration; no unconsented broad control change was established.
Mechanism
The CLI writes the proxy URL and optional API key into a named MCP server entry in CODEX_HOME/config.toml or ~/.codex/config.toml.
AI rationale
Source inspection found a guarded Codex MCP setup that can refresh after a user has installed it, which merits a warning under the agent integration policy. The manifest has no install hook, and the inspected credential transfers are tied to user-supplied chat destinations and actions.
Evidence
dist/index.jsCODEX_HOME/config.toml~/.codex/config.toml
Decision evidence
public snapshotAI called this Suspicious at 84.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The package has no install lifecycle hook; its prepublishOnly script builds and typechecks.
- The Codex setup writes an APIblaze MCP entry with its URL and optional API key, and refuses to replace a conflicting non-APIblaze entry.
- Setup requires an explicit --install-mcp command or a user choice; after installation is saved, later CLI runs can refresh the entry.
- The React widget obtains its token from the embedding application's getToken callback and sends it with a chat request to the supplied endpoint or project host when the user sends a message.
- The CLI presents its LLM key as optional for bring-your-own-model use and says it is sent to APIblaze only per chat turn.
Evidence against
- The manifest points to the CLI bundle and has no preinstall, install, or postinstall hook.
- The agent configuration change is scoped to an APIblaze MCP entry and is guarded by explicit command or user selection.
- The inspected credential flows are tied to user-supplied destinations and chat actions; no unrelated credential recipient was confirmed.
Behavioral surface
Source & flagged code
9 flaggeddist/index.jsView file
32L33: // package.json
L34: var version;
...
L93: function cacheFile() {
L94: return import_path.default.join(import_os.default.homedir(), ".apiblaze", "cli-policy.json");
L95: }
...
L104: for (const name of CONTROL_PLANE_ENV) {
L105: const v = process.env[name];
L106: try {
...
L154: const base2 = globalThis.fetch.bind(globalThis);
L155: globalThis.fetch = (async (input, init) => {
L156: const url = typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
Critical
Credential Exfiltration
Source appears to send environment or credential material to an external endpoint.
dist/index.jsView on unpkg · L3232Trigger-reachable chain: manifest.main -> dist/index.js
L32:
L33: // package.json
L34: var version;
...
L93: function cacheFile() {
L94: return import_path.default.join(import_os.default.homedir(), ".apiblaze", "cli-policy.json");
L95: }
...
L104: for (const name of CONTROL_PLANE_ENV) {
L105: const v = process.env[name];
L106: try {
...
L154: const base2 = globalThis.fetch.bind(globalThis);
L155: globalThis.fetch = (async (input, init) => {
L156: const url = typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
Critical
Trigger Reachable Dangerous Capability
A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.
dist/index.jsView on unpkg · L32Findings
3 Critical7 High3 Medium5 Low
CriticalCredential Exfiltrationdist/index.js
CriticalTrigger Reachable Dangerous Capabilitydist/index.js
CriticalPrevious Version Dangerous Deltadist/index.js
HighChild Processdist/index.js
HighShell
HighSame File Env Network Executiondist/index.js
HighCredential Redirect Persistencedist/index.js
HighSandbox Evasion Gated Capabilitydist/index.js
HighCross File Remote Execution Contextdist/index.js
HighTrigger Reachable Credential Exfiltrationdist/index.js
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
Affected versions and remediation
This report applies to apiblaze@0.21.47.
See all versions of apiblaze for other recorded verdicts.
- Review the evidence and your use of apiblaze@0.21.47 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.