ux-atlas@0.1.4
UX skills for agents. Each skill is skills/<name>/SKILL.md.
AI Security Review
scanned 6d ago · by lpm-firewall-aiThe postinstall hook automatically writes package-controlled instructions into the consumer project's AGENTS.md and CLAUDE.md files. This is an unconsented mutation of a broad AI-agent control surface, supporting a publish block.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
Installing the package locally with its postinstall hook enabled.
Impact
Agents that load those project files may adopt behavior directed by the package's instructions and skills.
Mechanism
The hook writes package instruction pointers into the consumer project's agent instruction files and creates .atlas configuration files, without an affirmative consent check.
Policy narrative
During local installation, the postinstall script modifies project-level AI agent instruction files and points them to instructions shipped by the package. This changes the behavior of agents that load those files without an affirmative setup choice, so the hook is an unconsented mutation of the consumer's agent control surface.
AI rationale
The postinstall hook automatically writes package-controlled instructions into the consumer project's AGENTS.md and CLAUDE.md files. This is an unconsented mutation of a broad AI-agent control surface, supporting a publish block.
Evidence
package.jsonscripts/init.mjsAGENTS.mdCLAUDE.md.atlas/config.json.atlas/README.md
Decision evidence
public snapshotAI identified this as a dangerous AI-agent capability at 94.0% confidence with low false-positive risk.
Evidence for policy risk
- The postinstall hook automatically runs scripts/init.mjs.
- The script adds package instruction pointers to the consumer project's AGENTS.md and CLAUDE.md files.
- The hook writes those pointers and creates .atlas configuration files without an affirmative consent check.
- The injected pointers direct agents to load this package's instructions and skills.
Evidence against
- The script skips global installations.
Behavioral surface
EnvironmentVarsFilesystem
HighEntropyStrings
Source & flagged code
7 flagged · loading sourcepackage.jsonView file
•scripts.postinstall = node scripts/init.mjs
High
Install Time Lifecycle Scripts
Package defines install-time lifecycle scripts.
package.jsonView on unpkg23"scripts": {
"lint": "eslint .",
"postinstall": "node scripts/init.mjs",
"index": "node scripts/sk[redacted]",
"test": "node --test scripts/*.test.mj
High
Suspicious Lifecycle Evidence
The postinstall hook automatically runs scripts/init.mjs.
package.jsonView on unpkg · L23scripts/init.mjsView file
7const END = '<!-- /ux-atlas -->';
L8: const instructions = 'node_modules/ux-atlas/instructions/AGENTS.md';
L9: const skills = 'node_modules/ux-atlas/skills/**/SKILL.md';
...
L72: if (!next.endsWith('\n')) next += '\n';
L73: fs.writeFileSync(filePath, next);
L74: }
...
L86: if (!existing.includes(legacyBriefLine)) return;
L87: fs.writeFileSync(filePath, existing.split(legacyBriefLine).join('').replace(/^\n+/, ''));
L88: }
...
L90: export function applyPointers(projectRoot) {
L91: const agents = path.join(projectRoot, 'AGENTS.md');
L92: const claude = path.join(projectRoot, 'CLAUDE.md');
Findings
1 Critical6 High2 Medium3 Low
CriticalAi Agent Control Hijackscripts/init.mjs
HighInstall Time Lifecycle Scriptspackage.json
HighExternal Ai Agent Control Surface Mutationscripts/init.mjs
HighSuspicious Lifecycle Evidencepackage.json
HighAi Review Evidencescripts/init.mjs
HighAi Review Evidencescripts/init.mjs
HighAi Review Evidencescripts/init.mjs
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
LowScripts Present
LowFilesystem
LowHigh Entropy Strings
Affected versions and remediation
This report applies to ux-atlas@0.1.4.
See all versions of ux-atlas for other recorded verdicts.
- Avoid installing ux-atlas@0.1.4. 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: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.