registry / AI-agent control-surface risk / @akashpandey/agent-relay

@akashpandey/agent-relay

Pass the baton between Claude Code, Codex, Antigravity, and OpenCode with a single Unix command.

This report shows the evidence for @akashpandey/agent-relay@1.0.2. 1 of 13 known versions is blocked or flagged. The newest version, 1.5.1, is allowed. See all versions.

AI Security Review

scanned 14d ago · by lpm-firewall-ai

The postinstall hook invokes an installer that deletes existing skill directories across several agent configuration locations and replaces them with package-controlled links. This is concrete unconsented install-time agent control-surface mutation.

Static reason

One or more suspicious static signals were detected.

Trigger

A global npm install or a direct install into the package directory.

Impact

Existing skill contents at those locations can be deleted and replaced, changing agent behavior.

Mechanism

The hook runs the installer, which removes existing non-symlink skill directories in several agent configuration locations and creates symlinks to the package skill.

Policy narrative

On global install or direct installation into the package directory, npm runs a postinstall hook that invokes the skill installer. The installer removes existing non-symlink directories at multiple agent skill locations, then links those locations to the package-owned skill. This is an automatic install-time mutation of broad AI-agent configuration surfaces.

AI rationale

Evidence

package.jsonscripts/postinstall.mjsinstall-skill$HOME/.agents/skills/agent-relay$HOME/.claude/skills/agent-relay$HOME/.gemini/skills/agent-relay$HOME/.gemini/antigravity/skills/agent-relay$HOME/.config/opencode/skills/agent-relay$HOME/.codex/skills/agent-relay

Decision evidence

public snapshot

AI identified this as a dangerous AI-agent capability at 97.0% confidence with low false-positive risk.

Evidence for policy risk

The package runs a postinstall script.
On global installs or direct installs, postinstall invokes the skill installer.
The installer removes existing non-symlink skill directories in multiple agent configuration locations and replaces them with links to the package’s skill.
The flagged Codex configuration code runs only through an explicit MCP setup path; it does not reduce the postinstall risk.

Evidence against

The postinstall action is gated to global installs or direct installation into the package directory.

Behavioral surface

Source

ChildProcessEnvironmentVarsFilesystemNetworkShell

Supply chain

HighEntropyStringsUrlStrings

ManifestNo manifest risk signals triggered.

scanned 19 file(s), 272 KB of source, external domains: 127.0.0.1

Source & flagged code

6 flagged

package.jsonView file

Lines 54-85json

54 "docker-compose.sample.yml",

55 "README.md",

56 "LICENSE"

57 ],

58 "keywords": [

59 "ai",

60 "agents",

61 "coding-agents",

62 "mcp",

63 "orchestration",

64 "claude-code",

65 "codex",

66 "opencode",

67 "antigravity",

68 "relay",

69 "delegation",

70 "multi-agent"

71 ],

72 "scripts": {

73 "sync:skills": "node scripts/sync-plugin-skill.mjs",

74 "postinstall": "node scripts/postinstall.mjs",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L74

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg · L74

75 "prepack": "npm run sync:skills",

76 "test": "node --test tests/*.mjs",

77 "test:mcp": "node --test tests/test-mcp-server.mjs",

78 "test:wrappers": "./tests/test-wrappers.sh",

79 "test:package": "node tests/package-smoke.js",

80 "doctor": "./agent-doctor",

81 "dashboard": "./agent-dashboard start",

82 "dashboard:dev": "npm --prefix dashboard run dev"

83 }

84}

dashboard/server.jsView file

Lines 1-23javascript

1import http from 'node:http';

2import fs from 'node:fs';

3import path from 'node:path';

4import { exec } from 'node:child_process';

scripts/agent-cli.mjsView file

Lines 243-283javascript

243 printDashboardCommands();

244 return;

245 }

246

247 const dockerReady = spawnSync('docker', ['info'], { stdio: 'ignore' }).status

scripts/finalize-run.shView file

•path = scripts/finalize-run.sh kind = build_helper sizeBytes = 753 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

scripts/finalize-run.shView on unpkg

agentView file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 10

High

Semantic Analysis Limited

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

agentView on unpkg

Source & flagged code

6 flagged · loading source

package.jsonView file

•scripts.postinstall = node scripts/postinstall.mjs

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.mjs

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg

dashboard/server.jsView file

3import path from 'node:path'; L4: import { exec } from 'node:child_process'; L5: import { fileURLToPath } from 'node:url';

High

Child Process

Package source references child process execution.

dashboard/server.jsView on unpkg · L3

Findings

5 High5 Medium4 Low

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processdashboard/server.js

HighShell

HighSame File Env Network Executionscripts/agent-cli.mjs

HighSemantic Analysis Limitedagent

MediumAmbiguous Install Lifecycle Scriptpackage.json

MediumNetwork

MediumEnvironment Vars

MediumShips Build Helperscripts/finalize-run.sh

MediumStructural Risk Force Deep Review

LowScripts Present

LowFilesystem

LowHigh Entropy Strings

LowUrl Strings

Affected versions and remediation

This report applies to @akashpandey/agent-relay@1.0.2.

The versions list below shows the verdict for each recorded version.

Avoid installing @akashpandey/agent-relay@1.0.2. Remove it from direct dependencies and check your lockfile for transitive copies.
Choose an independently verified alternative or release. This report does not establish that other versions are safe.
If this version ran, investigate the affected machine and build environment. Rotate credentials it could access and rebuild from a trusted environment.

Evidence last updated: 2026-10-08 21:02:55.016Z (UTC).

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

Versions

13 known versions

1.5.1latestUnknown—AllowUnder ReviewOct 21, 2026, 8:58 AM 1.4.0Unknown—AllowUnder ReviewOct 15, 2026, 6:07 PM 1.3.1Unknown—AllowUnder ReviewSep 28, 2026, 3:05 PM 1.3.0Unknown—AllowUnder ReviewSep 26, 2026, 10:22 AM

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Versions

Related packages

Other AI-agent control-surface risk packages

Runtime surface

Behavioral surface

Related packages

Other AI-agent control-surface risk packages

Source & flagged code