api-anthropic@1.0.1
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThe source implements persistent agent configuration changes through an explicit CLI command, which falls under the warning policy for user-command agent mutation. No concrete malicious chain or unconsented install-time behavior was identified.
Static reason
One or more suspicious static signals were detected.
Trigger
Invoke api-anthropic with a key argument or a valid ANTHROPIC_AUTH_TOKEN environment variable.
Impact
Subsequent Claude Code sessions use the changed gateway and authentication settings; existing authentication and model preferences can be overwritten.
Mechanism
Read and rewrite .claude/settings.json with the selected gateway, token, model settings, and onboarding state, then launch Claude Code unless --no-launch is supplied.
AI rationale
The source implements persistent agent configuration changes through an explicit CLI command, which falls under the warning policy for user-command agent mutation. No concrete malicious chain or unconsented install-time behavior was identified.
Evidence
bin.js.claude/settings.json
Network endpoints1
api-anthropic.help
Decision evidence
public snapshotAI called this Suspicious at 96.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- package.json exposes bin.js as an explicit CLI and defines no install lifecycle hooks.
- bin.js selects a key from command arguments or ANTHROPIC_AUTH_TOKEN and a gateway from ANTHROPIC_GATEWAY_URL or https://api-anthropic.help.
- bin.js reads the user's .claude/settings.json and replaces its authentication and gateway environment settings.
- bin.js saves the configuration persistently and normally launches Claude Code using it.
Evidence against
- Configuration requires invoking the CLI with a key; npm installation alone does not activate it.
- Platform checks select Windows command variants; they do not conceal execution from analysis.
- No direct network request, secret-file harvesting, obfuscated payload, or runtime self-dependency appears in the inspected entrypoint.
Behavioral surface
ChildProcessEnvironmentVarsFilesystem
UrlStrings
Source & flagged code
9 flaggedbin.jsView file
15const path = require('path')
L16: const { spawn, spawnSync } = require('child_process')
L17:
L18: const GATEWAY = process.env.ANTHROPIC_GATEWAY_URL || 'https://api-anthropic.help'
L19: const isWin = process.platform === 'win32'
L20: const C = { cyan: '\x1b[36m', green: '\x1b[32m', red: '\x1b[31m', yellow: '\x1b[33m', dim: '\x1b[2m', reset: '\x1b[0m' }
...
L61: // ---- 2) configure ~/.claude/settings.json (merge, non-destructive) ----
L62: const dir = path.join(os.homedir(), '.claude')
L63: const file = path.join(dir, 'settings.json')
L64: let cfg = {}
L65: try { cfg = JSON.parse(fs.readFileSync(file, 'utf8')) } catch { /* fresh */ }
L66:
High
Sandbox Evasion Gated Capability
Source gates dangerous network, credential, or execution behavior behind CI, host, platform, time, or geo fingerprint checks.
bin.jsView on unpkg · L15•Runtime or CLI code links a write operation to an explicit external AI-agent control path:
= (m) => console.log(`${tag} ${c.green}${m}${c.reset}`)
const warn = (m) => console.log(`${tag} ${c.yellow}${m}${c.reset}`)
const fail = (m) => console.error(`${tag} ${c.red}${m}${c.reset}`)
const args = process.argv.slice(2)
const nolaunch = args.includes('--no-launch')
const key = (args.find((a) => !a.startswith('-')) || process.env.anthropic_auth_token || '').trim()
if (!key || !/^sk-/.test(key)) {
fail('missing or invalid api key.')
console.error(`\nusage: npx api-anthropic <api_key>`)
console.error(`example: npx api-anthropic sk[redacted]\n`)
process.exit(1)
}
function commandexists(cmd) {
const finder = iswin ? 'where' : 'which'
cons
High
Findings
2 High8 Medium2 Low
HighSandbox Evasion Gated Capabilitybin.js
HighExternal Ai Agent Control Surface Mutationbin.js
MediumEnvironment Vars
MediumSuspicious Lifecycle Evidencepackage.json
MediumAi Review Evidencebin.js
MediumAi Review Evidencebin.js
MediumAi Review Evidencebin.js
MediumAi Review Evidencebin.js
MediumAi Review Evidencebin.js
MediumAi Review Evidencebin.js
LowFilesystem
LowUrl Strings
Affected versions and remediation
This report applies to api-anthropic@1.0.1.
See all versions of api-anthropic for other recorded verdicts.
- Review the evidence and your use of api-anthropic@1.0.1 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.