psyclaw@0.27.6
A standalone, evidence-grounded social-science research agent — a conversation-driven agent that bundles the Pi runtime with psyclaw's research contracts, evidence gates, and workflows
AI Security Review
scanned 2mo ago · by lpm-firewall-aiThe package has notable user-invoked agent-runtime behavior, but source inspection found no install hook or concrete malicious chain. Its process controls limit worker tools and environment inheritance, and no exfiltration or payload staging was identified.
Static reason
No blocking static signals were detected.; source matched previously finalized malicious package; routed for review; previous stored version diff introduced dangerous source
Trigger
A user runs psyclaw or starts its chat command.
Impact
Changes the bundled Pi runtime presentation and may expose existing local skill paths to the launched Pi session; no exfiltration or unconsented install-time mutation was found.
Mechanism
User-invoked local Pi runtime branding and extension loading.
AI rationale
The package has notable user-invoked agent-runtime behavior, but source inspection found no install hook or concrete malicious chain. Its process controls limit worker tools and environment inheritance, and no exfiltration or payload staging was identified.
Evidence
package.jsondist/src/chat.jsscripts/rebrand-pi.mjsdist/src/adapters/pi/extension.jsdist/src/adapters/pi/rpc.jsdist/src/integrations/mcp.jsnode_modules/@earendil-works/pi-coding-agent/package.jsonnode_modules/@earendil-works/pi-coding-agent/dist/modes/interactive/interactive-mode.js
Decision evidence
public snapshotAI called this Clean at 91.0% confidence as Benign with low false-positive risk.
Evidence for AI clean decision
- The user-run chat entrypoint rewrites a pinned Pi dependency for branding and loads its extensions.
- The extension discovers local Claude, Codex, and agent skill directories during a Pi resource-discovery event.
Evidence against
- package.json has no preinstall, install, or postinstall hook; its only lifecycle hook is prepublishOnly.
- Dependency rewriting occurs when the user launches chat, not during npm installation.
- Worker subprocesses restrict tools to a read-only allowlist and MCP subprocesses use an explicit environment without inherited secrets.
- No source evidence of secret exfiltration, remote payload loading, destructive behavior, or self-dependency was found.
Behavioral surface
ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShell
HighEntropyStringsUrlStrings
Source & flagged code
4 flaggeddist/src/setup.jsView file
172const entry = fileURLToPath(import.meta.resolve("@earendil-works/pi-coding-agent"));
L173: const { readStoredCredential } = await import(pathToFileURL(join(dirname(entry), "core", "auth-storage.js")).href);
L174: return readStoredCredential(providerId, join(agentDir, "auth.json")) !== undefined;
Medium
Dynamic Require
Package source references dynamic require/import behavior.
dist/src/setup.jsView on unpkg · L172dist/src/integrations/mcp.jsView file
•matchType = normalized_sha256
matchedPackage = psypi@0.3.2
matchedPath = dist/src/integrations/mcp.js
matchedIdentity = npm:cHN5cGk:0.3.2
similarity = 1.000
summary = normalized source hash matched finalized malicious source
High
Known Malware Source Similarity
Source file is highly similar to a previously finalized malicious package; route for source-aware review.
dist/src/integrations/mcp.jsView on unpkgdist/src/adapters/pi/rpc.jsView file
Findings
2 High4 Medium6 Low
HighKnown Malware Source Similaritydist/src/integrations/mcp.js
HighPrevious Version Dangerous Deltadist/src/adapters/pi/rpc.js
MediumDynamic Requiredist/src/setup.js
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
LowNon Install Lifecycle Scripts
LowScripts Present
LowFilesystem
LowHigh Entropy Strings
LowUrl Strings
LowSuspicious Dependency Evidencedist/src/chat.js
Affected versions and remediation
This report applies to psyclaw@0.27.6.
See all versions of psyclaw for other recorded verdicts.
- Review the evidence and your use of psyclaw@0.27.6 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.