registry / Credential exfiltration / @theronap/agnoclast-mcp
@theronap/agnoclast-mcp
Connect your AI assistant to Cortex — your org's projects, activity, gaps, and directives, scoped to you.
This report shows the evidence for @theronap/agnoclast-mcp@0.9.293. 7 of 131 known versions are blocked or flagged. The newest version, 0.9.308, is allowed. See all versions.
AI Security Review
scanned 2d ago · by lpm-firewall-aiThe server automatically reads a Codex login id_token and sends it to the package's service during startup when confidential-brain seats exist. This is an unsolicited transfer of another service's credential, so the package is malicious.
Static reason
No blocking static signals were detected.; source matched previously finalized malicious package; routed for review
Trigger
Starting the MCP server with a configured Agnoclast token when the account has confidential-brain seats and Codex is the nearest agent.
Impact
Exposes a credential from another service to Agnoclast without a separate user request to send that token.
Mechanism
Reads ~/.codex/auth.json, uses its id_token as proof, then posts that proof to the confidential-brain account endpoint.
Attack narrative
When the MCP server starts, it checks whether the user's Agnoclast account has confidential-brain seats. If so, and Codex is the nearest agent, it reads the OpenAI id_token from ~/.codex/auth.json and posts it to Agnoclast as account proof. The token is sent automatically during startup, before a confidential-brain tool call.
AI rationale
The server automatically reads a Codex login id_token and sends it to the package's service during startup when confidential-brain seats exist. This is an unsolicited transfer of another service's credential, so the package is malicious.
Evidence
lib/server.mjslib/confidential_session.mjslib/diagnose.mjs~/.codex/auth.json
Network endpoints1
agnoclast.com/api/brains/confidential/mine
Decision evidence
public snapshotAI called this Malicious at 94.0% confidence as Malware with low false-positive risk.
Evidence for block
- At server startup, if the Agnoclast account has confidential-brain seats, the code gathers an AI-agent proof and sends it to Agnoclast.
- For Codex, the proof is the id_token read from ~/.codex/auth.json.
- The proof is sent in a POST to /api/brains/confidential/mine; the default service host is agnoclast.com.
Evidence against
- The package has no npm lifecycle scripts.
- The background close-out job is off by default and installed only through an explicit command.
Behavioral surface
ChildProcessCryptoEnvironmentVarsFilesystemNetwork
HighEntropyStringsUrlStrings
Source & flagged code
15 flaggedlib/capture.mjsView file
•matchType = normalized_sha256
matchedPackage = @theronap/agnoclast-mcp@0.9.285
matchedPath = lib/capture.mjs
matchedIdentity = npm:QHRoZXJvbmFwL2Fnbm9jbGFzdC1tY3A:0.9.285
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.
lib/capture.mjsView on unpkg1import { readFileSync, accessSync, constants, openSync, mkdirSync } from 'fs'
L2: import { spawn, execFileSync } from 'child_process'
L3: import { homedir } from 'os'
...
L21: if (!res.ok) return null
L22: const j = await res.json().catch(() => ({}))
L23: return Array.isArray(j.types) ? j.types : null
...
L35: try {
L36: if (resolve(String(cwd)) === homedir()) return 'general'
L37: } catch { /* unresolvable path — fall through to basename */ }
...
L45: //
L46: // Handles the four remote forms git emits: scp-like ssh (git@github.com:o/n.git), ssh://, https://,
L47: // git://. The HOST CHECK IS LOAD-BEARING, not cosmetic: this string becomes a clearance key, and
Low
Weak Crypto
Package source references weak cryptographic algorithms.
lib/capture.mjsView on unpkg · L1lib/editors/antigravity.mjs
Findings
13 High3 Medium5 Low
HighTrigger Reachable External Ai Agent Control Surface Mutationlib/channel.mjs
HighTrigger Reachable Persistencelib/closeout.mjs
HighSemantic Analysis Limitedbin/cortex-mcp.mjs#virtual:normalized:round1
HighKnown Malware Source Similaritylib/confidential_session.mjs
HighKnown Malware Source Similaritylib/capture.mjs
HighKnown Malware Source Similaritylib/edge_extract.mjs
HighKnown Malware Source Similaritylib/graphify_sync.mjs
HighKnown Malware Source Similaritylib/obligations_worker.mjs
HighKnown Malware Source Similaritylib/confidential_guard.mjs
HighKnown Malware Source Similaritylib/obligation_review.mjs
High
Affected versions and remediation
This report applies to @theronap/agnoclast-mcp@0.9.293.
The versions list below shows the verdict for each recorded version.
- Avoid installing @theronap/agnoclast-mcp@0.9.293. 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.