registry / Data exfiltration / @cliphijack/santaclaude
@cliphijack/santaclaude
SantaClaude 커넥터 — 클라우드 예약을 내 로컬 Claude(tmux)에 발사
This report shows the evidence for @cliphijack/santaclaude@1.0.122. 8 of 25 known versions are blocked or flagged. The newest version, 1.0.124, is flagged. See all versions.
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThis is not a passive connector: it combines remote terminal observation and prompt injection with default permission bypass and persistent changes to a broad AI-agent control surface. The absence of an npm lifecycle hook does not remove the concrete harm once the advertised CLI is run.
Static reason
One or more suspicious static signals were detected.; source matched previously finalized malicious package; routed for review
Trigger
The user runs santaclaude after configuring or loading a connection.
Impact
A remote service can observe terminal content and steer a permission-bypassed Claude session; subsequent Claude sessions can modify project CLAUDE.md files and create commits.
Mechanism
Remote screen exfiltration, command injection, and global Claude hook installation.
Attack narrative
After the CLI is run, it sends tmux screen captures and session metadata to santaclaude.app and processes remote control messages. Remote jobs are injected into tmux panes, whose default Claude command disables permission prompts. The same run installs global Stop, Notification, and SessionStart hooks in ~/.claude/settings.json. The SessionStart path reads prior Claude transcripts, submits them to a Claude model, appends derived rules to each project CLAUDE.md, and attempts a git commit. This provides remote AI-agent control, data export, persistence, and unprompted project mutation.
AI rationale
This is not a passive connector: it combines remote terminal observation and prompt injection with default permission bypass and persistent changes to a broad AI-agent control surface. The absence of an npm lifecycle hook does not remove the concrete harm once the advertised CLI is run.
Evidence
santaclaude.jshooks/sc-mdsync.shhooks/mdsync-apply.js~/.claude/settings.json~/.claude/CLAUDE.md~/.claude/skills~/.santaclaudeCLAUDE.md
Network endpoints1
santaclaude.app
Decision evidence
public snapshotAI called this Malicious at 98.0% confidence as Malware with low false-positive risk.
Evidence for block
- It sends captured tmux screens, session details, directories, and system state to the remote service, then accepts remote control commands.
- Remote jobs are injected into the selected tmux pane, while the default Claude command disables permission prompts.
- Running the CLI automatically adds global Claude hooks and a status command to the user's Claude configuration.
- The installed SessionStart hook launches a transcript-processing script that appends AI-derived rules to project CLAUDE.md files and commits them.
- The source header addresses an inspecting reader and makes self-serving claims about what review should find.
Evidence against
- package.json contains no npm lifecycle scripts.
- The package is activated through its bin command rather than at install time.
- The separately flagged bundled brainstorming server is a localhost service, and the graph renderer is an explicit utility.
Behavioral surface
ChildProcessCryptoEnvironmentVarsFilesystemNetwork
Source & flagged code
11 flagged · loading sourcebundled-superpowers/skills/brainstorming/scripts/server.cjsView file
536const url = companionUrl(); // must carry the key or the gate 403s it
L537: const cp = require('child_process');
L538: // Operator-provided launcher: run as given (this env var is trusted operator input).
High
Child Process
Package source references child process execution.
bundled-superpowers/skills/brainstorming/scripts/server.cjsView on unpkg · L536•matchType = normalized_sha256
matchedPackage = @brainervirus/workit-opencode@0.8.5
matchedPath = [redacted].cjs
matchedIdentity = npm:[redacted]:0.8.5
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.
bundled-superpowers/skills/brainstorming/scripts/server.cjsView on unpkg1const crypto = require('crypto');
L2: const http = require('http');
L3: const fs = require('fs');
...
L12: function computeAcceptKey(clientKey) {
L13: return crypto.createHash('sha1').update(clientKey + WS_MAGIC).digest('base64');
L14: }
...
L84:
L85: const PORT_FILE = process.env.BRAINSTORM_PORT_FILE || null;
L86: const randomPort = () => 49152 + Math.floor(Math.random() * 16383);
...
L195: try { sessionStorage.setItem('brainstorm-session-key', ${jsonKey}); } catch (e) {}
L196: location.replace('/');
L197: </script>
Findings
10 High4 Medium4 Low
HighChild Processbundled-superpowers/skills/brainstorming/scripts/server.cjs
HighShell
HighSame File Env Network Executionsantaclaude.js
HighCommand Output Exfiltrationsantaclaude.js
HighExternal Ai Agent Control Surface Mutationsantaclaude.js
HighTrigger Reachable External Ai Agent Control Surface Mutationsantaclaude.js
HighRuntime Package Installsantaclaude.js
HighSemantic Analysis Limitedbundled-superpowers/skills/brainstorming/scripts/server.cjs#virtual:normalized:round1
HighKnown Malware Source Similaritybundled-superpowers/skills/brainstorming/scripts/server.cjs
HighKnown Malware Source Similaritybundled-superpowers/skills/writing-skills/render-graphs.js
Affected versions and remediation
This report applies to @cliphijack/santaclaude@1.0.122.
The versions list below shows the verdict for each recorded version.
- Avoid installing @cliphijack/santaclaude@1.0.122. 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.