registry / loopwithai-runner / 0.1.0

loopwithai-runner@0.1.0

Local runner for Loop with AI — executes your project's queued agent jobs (report/implement/drift) on your machine using your own AI CLI (Claude, Antigravity, or a custom agent). No API credits.

Static Scan Results

scanned 2h ago · by rust-scanner

Static analysis flagged 12 finding(s) at 72.0% confidence. This version is warn-only unless an AI or security-team review confirms malicious behavior.

Static reason

One or more suspicious static signals were detected.

Decision evidence

public snapshot

Behavioral surface

Source

ChildProcessEnvironmentVarsFilesystemNetworkShellWebSocket

Supply chain

HighEntropyStringsUrlStrings

ManifestNo manifest risk signals triggered.

scanned 2 file(s), 31.1 KB of source, external domains: fpsiszputyiacbvjzytx.supabase.co

Source & flagged code

4 flagged

pdf.mjsView file

Lines 1-23javascript

1// Minimal HTML → PDF via headless Chrome/Edge (CDP over the built-in WebSocket). Ported from Dimitra's

2// browser.cjs — the runner's only job for documents is to PRINT the already-finalized HTML the app sent.

3// It renders nothing and writes no code. Auto-detects a browser; set DOC_BROWSER to override.

4import { spawn } from 'node:child_process'

5import http from 'node:http'

High

Child Process

Source & flagged code

4 flagged · loading source

pdf.mjsView file

3// It renders nothing and writes no code. Auto-detects a browser; set DOC_BROWSER to override. L4: import { spawn } from 'node:child_process' L5: import http from 'node:http'

High

Child Process

Package source references child process execution.

pdf.mjsView on unpkg · L3

1// Minimal HTML → PDF via headless Chrome/Edge (CDP over the built-in WebSocket). Ported from Dimitra's L2: // browser.cjs — the runner's only job for documents is to PRINT the already-finalized HTML the app sent. L3: // It renders nothing and writes no code. Auto-detects a browser; set DOC_BROWSER to override. L4: import { spawn } from 'node:child_process' L5: import http from 'node:http' ... L10: const CANDIDATES = [ L11: process.env.DOC_BROWSER, L12: 'C:\\Program Files\\Google\\Chrome\\Application\\chrome.exe',

High

Same File Env Network Execution

A single source file combines environment access, network access, and code or shell execution; review context before blocking.

pdf.mjsView on unpkg · L1

runner.mjsView file

91async function api(method, path, body) { L92: const res = await fetch(API + path, { method, headers: { 'x-api-key': KEY, 'content-type': 'application/json' }, body: body ? JSON.stringify(body) : undefined }) L93: const json = await res.json().catch(() => ({})) ... L96: L97: // Run a command, capture stdout; resolve '' on error (git may not be present / not a repo). L98: function sh(cmd, args, cwd) { L99: return new Promise((resolve) => { L100: const cp = spawn(cmd, args, { cwd, shell: WIN }) L101: let out = ''

High

Command Output Exfiltration

Source combines command execution, command-output handling, and outbound requests; review data flow before blocking.

runner.mjsView on unpkg · L91

29// SYNCEFFORTS_DRY (optional) "1" = don't call Claude; post a stub (for testing the loop) L30: import { spawn } from 'node:child_process' L31: import fs from 'node:fs' ... L35: L36: const API = process.env.SYNCEFFORTS_API_URL || 'https://fpsiszputyiacbvjzytx.supabase.co/functions/v1/se-api' L37: const KEY = process.env.SYNCEFFORTS_API_KEY ... L39: const CLAUDE = process.env.SYNCEFFORTS_CLAUDE_CMD || 'claude' L40: const REPO = process.env.SYNCEFFORTS_REPO || process.cwd() L41: const PERMISSION = process.env.SYNCEFFORTS_PERMISSION || 'bypassPermissions' ... L44: const DRY = process.env.SYNCEFFORTS_DRY === '1' L45: const WIN = process.platform === 'win32' L46: if (!KEY || !PID) { console.error('Set SYNCEFFORTS_API_KEY and SYNCEFFORTS_PROJECT_ID.'); process.exit(1) }

High

Sandbox Evasion Gated Capability

Source gates dangerous network, credential, or execution behavior behind CI, host, platform, time, or geo fingerprint checks.

runner.mjsView on unpkg · L29

Findings

5 High3 Medium4 Low

HighChild Processpdf.mjs

HighShell

HighSame File Env Network Executionpdf.mjs

HighCommand Output Exfiltrationrunner.mjs

HighSandbox Evasion Gated Capabilityrunner.mjs

MediumNetwork

MediumEnvironment Vars

MediumStructural Risk Force Deep Review

LowScripts Present

LowFilesystem

LowHigh Entropy Strings

LowUrl Strings

runner.mjsView file

Lines 9-49javascript

9// then the runner reports what changed (diffstat + files + summary).

10// Runs in SYNCEFFORTS_REPO with write permissions. You review + commit.

11//

12// Env:

13// SYNCEFFORTS_API_KEY (required) org-scoped key from the app → API & Agents

14// SYNCEFFORTS_PROJECT_ID (required) which project to serve

15// SYNCEFFORTS_API_URL (optional) se-api base (defaults to the deployed one)

16// SYNCEFFORTS_CLAUDE_CMD (optional) the Claude CLI (default: "claude"); reads prompt on stdin (-p)

17// SYNCEFFORTS_AGENT (optional) which agent CLI to run: "claude" (default) | "antigravity" | "custom".

18// Decided ONCE at startup — a running runner never switches agents.

19// SYNCEFFORTS_ANTIGRAVITY_CMD (opt) Antigravity CLI (default: "agy"); prompt via --prompt-file, --yes.

20// SYNCEFFORTS_AGENT_CMD / _ARGS / _APPROVE_ARGS / _PROMPT (stdin|arg|file) — the "custom" adapter.

21// SYNCEFFORTS_REPO (optional) repo dir for `implement` jobs (default: current dir)

22// SYNCEFFORTS_PERMISSION (optional) Claude permission mode for implement jobs

23// (default: "bypassPermissions" — required so a headless run

24// never hangs on an approval prompt; it's your machine)

25// SYNCEFFORTS_POLL_MS (optional) poll interval (default 5000)

26// SYNCEFFORTS_WATCH (optional) "0" to disable auto-Radar. When SYNCEFFORTS_REPO is a git repo,

27// the runner watches HEAD and auto-queues a `radar` drift check on

28// each new commit (credit-free — the check runs on your local Claude).

29// SYNCEFFORTS_DRY (optional) "1" = don't call Claude; post a stub (for testing the loop)

30import { spawn } from 'node:child_process'

31import fs from 'node:fs'

32import os from 'node:os'

33import path from 'node:path'

34import { printPdf, findBrowser } from './pdf.mjs'

36const API = process.env.SYNCEFFORTS_API_URL || 'https://fpsiszputyiacbvjzytx.supabase.co/functions/v1/se-api'

37const KEY = process.env.SYNCEFFORTS_API_KEY

38const PID = process.env.SYNCEFFORTS_PROJECT_ID

39const CLAUDE = process.env.SYNCEFFORTS_CLAUDE_CMD || 'claude'

40const REPO = process.env.SYNCEFFORTS_REPO || process.cwd()

41const PERMISSION = process.env.SYNCEFFORTS_PERMISSION || 'bypassPermissions'

42const POLL = Number(process.env.SYNCEFFORTS_POLL_MS || 5000)

43const WATCH = process.env.SYNCEFFORTS_WATCH !== '0'

44const DRY = process.env.SYNCEFFORTS_DRY === '1'

45const WIN = process.platform === 'win32'

46if (!KEY || !PID) { console.error('Set SYNCEFFORTS_API_KEY and SYNCEFFORTS_PROJECT_ID.'); process.exit(1) }

High

Sandbox Evasion Gated Capability

Source gates dangerous network, credential, or execution behavior behind CI, host, platform, time, or geo fingerprint checks.

runner.mjsView on unpkg · L29

48// ── Agent adapters (pluggable headless CLI). FOOLPROOF: the agent is decided ONCE here from

49// SYNCEFFORTS_AGENT (default 'claude'); nothing switches it at runtime, so existing Claude runners

Lines 71-111javascript

71 promptFileFlag: process.env.SYNCEFFORTS_AGENT_PROMPT_FILE_FLAG || '--prompt-file',

72 argFlag: process.env.SYNCEFFORTS_AGENT_PROMPT_ARG_FLAG || '-p',

73 },

74}

75const adapter = ADAPTERS