registry / AI-agent capability abuse / osborn / 0.9.248

osborn@0.9.248

Voice AI coding assistant - local agent that connects to Osborn frontend

AI Security Review

scanned 23d ago · by lpm-firewall-ai

Static inspection establishes explicit-command agent configuration mutation, which warrants a warning under the supplied policy. No installation lifecycle trigger or concrete malicious chain supports blocking publication.

Static reason

High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source

Trigger

Explicitly launching osborn; skill writes additionally require a skill_add message through the connected room.

Impact

Changes affect Claude session retention and compaction beyond Osborn, and added skills can influence later agent behavior.

Mechanism

An imported module merges settings into the user's Claude settings file, while the frontend message handler writes supplied skill content into the global skills directory.

AI rationale

Evidence

bin/cli.jsdist/index.jsdist/claude-llm.js.claudesettings.jsonSKILL.md

Decision evidence

public snapshot

AI called this Suspicious at 94.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

The explicit CLI launches dist/index.js, which imports dist/claude-llm.js.
At module load, dist/claude-llm.js rewrites global Claude settings, setting the compaction window and retaining sessions for 3650 days without a separate consent check.
The connected frontend can send skill_add messages that write content into the user's global Claude skills directory.
package.json has no preinstall, install, or postinstall hook and no runtime self-dependency.

Evidence against

The configuration changes support the voice coding assistant's runtime functionality.
Inspected authentication fallback uses an OpenRouter key for OpenRouter and returns early when Claude authentication is available; no concrete credential theft was established.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShell

Supply chain

HighEntropyStrings

Source & flagged code

9 flagged

bin/recall.jsView file

Lines 1-25javascript

1#!/usr/bin/env node

3// osborn-recall — thin shim. Runs src/recall-cli.ts via tsx in dev, or the

4// compiled dist/recall-cli.js after `npm install`. Mirrors bin/cli.js.

6import { spawn } from 'child_process'

7import { fileURLToPath } from 'url'

High

Child Process

Package source references child process execution.

bin/recall.jsView on unpkg · L5

8import { dirname, join } from 'path'

9import { existsSync, lstatSync, symlinkSync } from 'fs'

10import os from 'os'

12// Same /workspace/.claude symlink guard as cli.js so stores resolve on Fly machines.

13try {

14 const home = os.homedir()

15 const target = '/workspace/.claude'

16 const link = join(home, '.claude')

17 if (existsSync(target) && home !== '/workspace') {

18 let needsLink = true

19 try { needsLink = !lstatSync(link).isSymbolicLink() } catch {}

20 if (needsLink) symlinkSync(target, link)

21 }

22} catch {}

24const __dirname = dirname(fileURLToPath(import.meta.url))

25const args = process.argv.slice(2)

dist/jsonl-search.jsView file

Lines 14-54javascript

14import { join } from 'path';

15import { readSessionHistory } from './session-access.js';

16// ============================================================

17// BM25 CACHE

18// ============================================================

19let bm25Cache = null

dist/index.jsView file

Lines 1-22javascript

1// Load environment variables FIRST before any other imports

2import 'dotenv/config';

3// Then hydrate volume-persisted user secrets into process.env, BEFORE any

4// module that reads its keys at load time (recall-client's RECALL_REGION const,

5// claude-auth fallback). Side-effect import — must stay right after dotenv.

6import './boot-secrets.js';

7import { voice, initializeLogger } from '@livekit/agents'

deploy.shView file

•path = deploy.sh kind = build_helper sizeBytes = 2309 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

deploy.shView on unpkg

dist/session-access.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 1

High

Semantic Analysis Limited

A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.

dist/session-access.js#virtual:normalized:round1View on unpkg

Source & flagged code

9 flagged · loading source

bin/recall.jsView file

5L6: import { spawn } from 'child_process' L7: import { fileURLToPath } from 'url'

High

Child Process

Package source references child process execution.

bin/recall.jsView on unpkg · L5

dist/jsonl-search.jsView file

34// @vscode/ripgrep bundles the rg binary — this is our own dependency L35: const { rgPath } = require('@vscode/ripgrep'); L36: if (rgPath && existsSync(rgPath))

Medium

Dynamic Require

Package source references dynamic require/import behavior.

dist/jsonl-search.jsView on unpkg · L34

dist/index.jsView file

2577const which = await new Promise((res) => { L2578: const p = spawn('which', ['code-server']); L2579: let out = ''; ... L2586: await new Promise((res, rej) => { L2587: const install = spawn('sh', ['-c', 'curl -fsSL https://code-server.dev/install.sh | sh'], { L2588: env: { ...process.env }, L2589: stdio: ['ignore', 'pipe', 'pipe'],

Findings

5 Critical3 High5 Medium5 Low

CriticalSame File Env Network Executiondist/index.js

CriticalCredential Exfiltrationdist/index.js

CriticalCommand Output Exfiltrationdist/index.js

CriticalDownload Executedist/index.js

CriticalPrevious Version Dangerous Deltadist/index.js

HighChild Processbin/recall.js

HighShell

HighSemantic Analysis Limiteddist/session-access.js#virtual:normalized:round1

MediumDynamic Requiredist/jsonl-search.js

MediumNetwork

MediumEnvironment Vars

MediumShips Build Helperdeploy.sh

MediumStructural Risk Force Deep Review

Affected versions and remediation

This report applies to osborn@0.9.248.

See all versions of osborn for other recorded verdicts.

Review the evidence and your use of osborn@0.9.248 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: 2026-09-30 16:24:59.744Z (UTC).

Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Runtime surface

Behavioral surface

Source & flagged code