Back to skill

Security audit

Session Sync Cloud

Security checks for vulnerabilities and agentic risk

Overview

This skill is a cloud memory-backup tool, but it makes strong security and retention claims that the implementation does not uphold.

Review this skill carefully before installing. It is intended to back up memory to your own cloud bucket, but the implementation does not provide the advertised end-to-end encryption and can overwrite local files during restore based on cloud manifest data. Do not use it for sensitive memory unless encryption, path validation, restore confirmation, and retention behavior are fixed and verified.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/run.py:194
Finding

Sensitive memory files are uploaded without the documented encryption

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/run.py:116
Finding

Cloud-controlled manifest paths allow arbitrary file overwrite during restore

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/run.py:164
Finding

Unescaped cloud metadata causes stored script injection in the dashboard

Content
View full analysis
{ts}{files}{size:.1f}Restore" ``` ### Technical Analysis The timestamp is derived from cloud object names returned by the configured S3-compatible service. It is inserted directly into two distinct parser contexts: - An HTML table cell - A quoted JavaScript argument inside an inline event handler No HTML escaping, JavaScript string encoding, or strict timestamp validation is applied. An attacker with bucket-write access can create a manifest object whose filename contains HTML or quote characters. When `generate_dashboard()` processes the object, those characters become executable markup or JavaScript in `memory/sync-dashboard.html`. Because the value appears inside an inline `onclick` attribute, escaping only HTML would not be sufficient. The implementation must avoid inline script construction or safely bind validated values through DOM APIs. ### Attack Path 1. An attacker gains permission to create objects under the configured manifest prefix. 2. The attacker creates an object with a filename designed to break out of the table cell or the quoted `onclick` argument. 3. The victim invokes `sync_now`, which lists cloud manifests and calls `generate_dashboard()`. 4. The malicious filename is written verbatim into `memory/sync-dashboard.html`. 5. The victim opens the generated dashboard in a browser. 6. The injected markup or JavaScript executes in the dashboard's browser context, either immediately or when the crafted restore button is selected. ### Impact Assessment The attacker can execute JavaScript in the context a ...[truncated 500 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This is a dangerous trust mismatch: the skill promises encrypted cloud backups, automated sync, retention, and a dashboard, but the analyzed behavior reportedly does not actually enforce encryption, does not truly automate scheduling, and incompletely prunes retained data. Users may rely on these claims and upload highly sensitive memory data under false assumptions, leading to plaintext exposure, excess retention, and unsafe recovery expectations.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill description promises encrypted cloud backup, but the implementation uploads manifests and file contents to S3 without any application-layer encryption step. If the bucket, endpoint, IAM policy, provider-side encryption settings, or transport assumptions are misconfigured or compromised, sensitive memory contents and metadata can be exposed, making this a material confidentiality issue as well as a deceptive security claim.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill performs or implies sensitive capabilities including reading local memory files, writing restored data, and sending data to remote cloud storage, but it does not declare an explicit overall tool scope or allowed-tools boundary. In a backup skill handling conversation history and memory state, missing scope declarations weakens reviewability and consent, making unintended data access or exfiltration materially more dangerous.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Restore writes into local memory state but the documentation does not clearly warn that existing files may be overwritten or rolled back. In this context, memory files are operational state for the agent, so an uninformed restore could destroy newer context, corrupt workflows, or reintroduce stale state.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is designed to transmit the entire memory directory, which may contain sensitive conversation history, notes, and agent state, yet it does not present a strong user-facing privacy warning at the point of configuration or use. Given the highly sensitive nature of memory data and the cloud upload context, lack of explicit disclosure increases the risk of accidental exfiltration and unsafe deployment.

Content

No source excerpt is available for this finding.

Cloud Storage Exfiltration

Medium
Category
Data Exfiltration
Confidence
55% confidence
Finding

Data is uploaded to cloud storage (S3 / GCS / Azure Blob). This may be a legitimate backup or exfiltration to an external bucket. Manual review is recommended.

Content

Scanner excerpt · scripts/run.py (reported line 94)May include surrounding context.

python
prefix = config.get("path_prefix", "")
    key = f"{prefix}{MANIFEST_PREFIX}/{timestamp}.json"
    body = json.dumps(manifest, indent=2)
    s3.put_object(Bucket=bucket, Key=key, Body=body.encode("utf-8"), ContentType="application/json")
    log_event("manifest_uploaded", {"bucket": bucket, "key": key, "files": len(manifest.get("files", {}))})

def list_manifests(config: dict) -> List[dict]:

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The description says the skill includes a web dashboard to browse backups, which implies an accessible web interface. This code only generates a local HTML file in the memory directory, and its JavaScript calls a /restore endpoint that is not implemented anywhere in this file, so the claimed dashboard capability is not actually delivered by the code shown.

Content

No source excerpt is available for this finding.

Cloud Storage Exfiltration

Medium
Category
Data Exfiltration
Confidence
55% confidence
Finding

Data is uploaded to cloud storage (S3 / GCS / Azure Blob). This may be a legitimate backup or exfiltration to an external bucket. Manual review is recommended.

Content

Scanner excerpt · scripts/run.py (reported line 216)May include surrounding context.

python
if e.response["Error"]["Code"] != "404":
                raise
        # Upload
        s3.upload_file(
            Filename=str(MEMORY_DIR / rel_path),
            Bucket=bucket,
            Key=s3_key,

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest advertises versioned retention for 30 days, but the pruning logic only deletes entries under the manifests prefix. Backup file objects under the files/ prefix are never removed or version-pruned here, so historical backup data can persist indefinitely beyond the declared retention window.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The restore path can overwrite all local memory files once invoked, with no CLI confirmation, dry-run, backup, or integrity verification beyond trusting the selected manifest. In a skill handling persistent memory state, accidental or coerced invocation can cause destructive loss of local data and replacement with potentially stale or attacker-controlled remote content.

Content

No source excerpt is available for this finding.