Back to skill

Security audit

Evez Daw Agent

Security checks for vulnerabilities and agentic risk

Overview

This music-generation skill is broadly coherent, but it exposes an unauthenticated network service that can read and write local files without adequate safeguards.

Install only if you are prepared to run it in a restricted local sandbox. Do not expose port 9112 to a network, and avoid running it with access to sensitive files until it binds to localhost by default, adds authentication, validates paths under fixed input/output directories, and enforces request and render limits.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
evez_daw.py:230
Finding

Unauthenticated Network Exposure of Local File Operations

Content
View full analysis
0 else {} if p=="/api/render": audio=render_track(b.get("preset","breakcore_170"),b.get("bars",4), b.get("bass","wobble"),b.get("bass_freq",55),b.get("fx_chain"),b.get("bpm")) fn=b.get("filename",f"evez_{b.get('preset','track')}_{int(time.time())}.wav") path=BASE_DIR/"output"/fn; path.parent.mkdir(parents=True,exist_ok=True) sf.write(str(path),audio,SR) self.j({"filename":fn,"path":str(path),"duration":len(audio)/SR,"bpm":b.get("bpm",PRESETS.get(b.get("preset","breakcore_170"),{}).get("bpm",170))}) elif p=="/api/drumkit": m=gen_drumkit(b.get("name","evez_breakcore"),b.get("style","breakcore")) self.j(m) elif p=="/api/chop": sp=b.get("sample_path","") if sp and os.path.exists(sp): a,sr=sf.read(sp); sl=chop(a.astype(np.float32),b.get("slices",8),b.get("gate",0.5)) pat=b.get("rearrange"); r=rearrange(sl,pat) if pat else np.concatenate(sl) fn=b.get("filename",f"chopped_{int(time.time())}.wav") path=BASE_DIR/"output"/fn; sf.write(str(path),r,SR) self.j({"filename":fn,"path":str(path),"slices":len(sl)}) else: self.j({"error":"sample not found"},400) s=HTTP ...[truncated 2425 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
evez_daw.py:210
Finding

Path Traversal Allows Writes Outside Intended Output Directories

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
evez_daw.py:192
Finding

Unbounded Request and Synthesis Parameters Permit Denial of Service

Content
View full analysis
0 and i0 else {} if p=="/api/render": audio=render_track(b.get("preset","breakcore_170"),b.get("bars",4), b.get("bass","wobble"),b.get("bass_freq",55),b.get("fx_chain"),b.get("bpm")) fn=b.get("filename",f"evez_{b.get('preset','track')}_{int(time.time())}.wav") path=BASE_DIR/"output"/fn; path.parent.mkdir(parents=True,exist_ok=True) sf.write(str(path),audio,SR) self.j({"filename":fn,"path":str(path),"duration":len(audio)/SR,"bpm":b.get("bpm",PRESETS.get(b.get("preset","breakcore_170"),{}).get("bpm",170))}) elif p=="/api/drumkit": m=gen_drumkit(b.get("name","evez_breakcore"),b.get("style","breakcore")) self.j(m) elif p=="/api/chop": sp=b.get("sample_path","") if sp and os.path.exists(sp): a,sr=sf.read(sp); sl=chop(a.astype(np.float32),b.get("slices",8),b.get("gate",0.5)) pat=b.get("rearrange"); r=rearrange(sl,pat) if pat else np.concatenate(sl) fn=b.get("filename",f"chopped_{int(time.time())}.wav") path=BASE_DIR/"output"/fn; sf.write(str(path),r,SR) ``` ### Technical Analysis The server trusts the HTTP `Content-Length` header and reads that m ...[truncated 2459 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The skill description materially understates behavior by omitting HTTP server/API exposure and overstating capabilities around sample-free operation and voice synthesis. Security review and deployment decisions often rely on the manifest; when network exposure and external file handling are undeclared, reviewers may approve a skill into environments where it opens listening services or processes user-supplied paths unexpectedly, increasing attack surface.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The /api/chop endpoint accepts a user-supplied sample_path and passes it directly to os.path.exists and sf.read, allowing a remote client to cause the service to open arbitrary local files. In a network-exposed process, this creates an arbitrary local file read primitive against any audio-readable file the process can access, which is unjustified for the stated autonomous music-generation purpose.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill advertises code-backed functionality that can write files, but the manifest does not declare any tool scope or permissions boundaries. That creates an authorization and review gap: operators and users cannot easily tell that the skill may modify the filesystem, increasing the risk of unexpected file creation or overwrite when the skill runs.

Content

No source excerpt is available for this finding.

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · evez_daw.py (reported line 165)May include surrounding context.

python
"breakcore_170": {"bpm":170,
        "kick":[1,0,0,1,0,0,1,0,0,1,0,0,1,0,1,0,1,0,0,0,1,0,0,1,0,1,1,0,0,0,1,0],
        "snare":[0,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,1,0,1,0,0,0,1,1,0,1],
        "hat":[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1],
        "clap":[0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,1,0,0,0]},
    "dubstep_140": {"bpm":140,
        "kick":[1,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0],

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · evez_daw.py (reported line 185)May include surrounding context.

python
"breakcore_170": {"bpm":170,
        "kick":[1,0,0,1,0,0,1,0,0,1,0,0,1,0,1,0,1,0,0,0,1,0,0,1,0,1,1,0,0,0,1,0],
        "snare":[0,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,1,0,1,0,0,0,1,1,0,1],
        "hat":[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1],
        "clap":[0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,1,0,0,0]},
    "dubstep_140": {"bpm":140,
        "kick":[1,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0],

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill creates directories and writes multiple WAV files plus a manifest.json in gen_drumkit, and this function is invoked both from the /api/drumkit endpoint and at startup. There is no confirmation prompt, no user-facing log describing the file creation, and no inline warning comment or docstring disclosing that the skill will modify the filesystem.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The manifest describes a DAW-style music generation capability, but does not mention operating as a network-accessible service. Running an HTTP server on 0.0.0.0 materially expands capability beyond local audio synthesis and sample processing.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The render and chop endpoints take a client-supplied filename and join it directly with BASE_DIR/'output'/fn before writing, enabling path traversal via values like '../../somewhere/file.wav'. A remote caller can therefore overwrite or create files outside the intended output directory with process privileges, which is a genuine filesystem write vulnerability.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest describes autonomous music generation from pure code and explicitly says no samples are needed, but this endpoint reads an arbitrary audio file from a user-supplied sample_path and performs sample chopping/rearrangement. That adds a file-based sample-processing behavior not aligned with the stated pure-code, no-sample-needed positioning.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The service reads arbitrary local file paths provided by the client without any access control or path restriction. Because the server listens on 0.0.0.0 and exposes this over HTTP, the skill context makes the issue more dangerous: a remote party can probe or consume local media files from the host environment and potentially leverage parser attack surface in audio libraries.

Content

No source excerpt is available for this finding.