Back to skill

Security audit

Edvisage Agent Ops

Security checks for vulnerabilities and agentic risk

Overview

This monitoring skill is coherent, but it directs the agent to persist broad action and cost logs without clear privacy limits, retention, or redaction controls.

Install only if you want the agent to keep persistent records of its activity. Before using it on sensitive projects, configure or instruct the agent to avoid logging prompts, secrets, personal data, credentials, full paths, API endpoints, private resource names, and detailed notes, and periodically clear the logs.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:38
Finding

Persistent Logging of Potentially Sensitive Operational Metadata Without Safeguards

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:38-52
Vulnerability Type: Persistent plaintext storage of potentially sensitive operational metadata
Risk Level: Medium

markdown
Maintain a running daily cost log in your memory files:
text
## Daily Cost Log — [DATE]
| Time | Model | Tokens (in/out) | Est. Cost | Task Type | Description |
|------|-------|-----------------|-----------|-----------|-------------|

2. Action Logging

Every significant action gets logged with:

  • Timestamp: When it happened
  • Action type: Post, reply, API call, file read, file write, skill use, tool use
  • Target: Where/what was affected
  • Outcome: Success, failure, or partial
  • Cost: If applicable
text

### Technical Analysis

The skill instructs the agent to maintain persistent cost and action records in its memory files. The action records include file operations, API calls, tool usage, affected targets, descriptions, and outcomes. These fields can contain sensitive filenames, filesystem paths, resource identifiers, user context, API operation details, or other operational metadata.

No requirements are provided for user consent, data minimization, access control, encryption, redaction, retention periods, or secure deletion. Persisting this information in general-purpose agent memory can expose it to future sessions, other installed skills, or any process with access to that memory.

This is an insecure data-handling practice rather than evidence of deliberate exfiltration. The audited project contains no script or instruction that transmits the stored records to the promotional domains or another remote service.

### Attack Path

1. A user or automated task causes the agent to read a sensitive file, invoke an API, use a tool, or operate on a sensitive target.
2. Following the skill instructions, the agent records the action type, target, description, and outcome in a persistent memo
...[truncated 1155 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit user opt-in before enabling persistent operational logging.
  2. Store logs in a dedicated, access-controlled location rather than general-purpose agent memory.
  3. Default to aggregated metrics, such as total token usage and cost, instead of recording individual file, API, and tool targets.
  4. Redact credentials, authorization headers, tokens, personal data, URL query strings, sensitive paths, and resource identifiers before storage.
  5. Define a short retention period and automatically delete records after expiration.
  6. Provide commands to inspect, export, disable, and securely erase collected records.
  7. Apply least-privilege filesystem permissions and encryption at rest where supported.
  8. Document exactly what is collected, where it is stored, who can access it, and how long it is retained.
  9. Avoid logging action descriptions or notes unless necessary; use predefined event categories and sanitized identifiers.
  10. Add explicit instructions prohibiting other skills or reporting features from transmitting logs without separate, informed authorization.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (4)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README explicitly promotes logging every action, tracking costs by model, and generating daily summaries, but it provides no warning about what agent activity data may be collected, how long it is retained, or whether sensitive prompts, tool outputs, or user data could be captured. In an agent skill context, this omission is security-relevant because operators may install the skill expecting observability while unintentionally enabling broad telemetry over potentially sensitive agent interactions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly tells the agent to track its own costs and log its actions, but it does not warn that those logs may capture sensitive operational data such as prompts, targets, file names, API endpoints, or user-derived context. In an agent environment, broad action logging can create secondary data exposure through retained memory files, later retrieval, or accidental inclusion in reports.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction to maintain a running daily cost log in memory files creates persistent storage of operational data without informing the user or limiting what may be stored. Persistent natural-language logs often outlive the original task and can be surfaced to future prompts, tools, or users, increasing confidentiality risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill directs the agent to keep persistent logs of significant actions, targets, outcomes, and notes in daily memory files, which creates a durable record of agent behavior in natural language. That record can unintentionally retain sensitive context and later leak through summaries, memory recall, compromise of storage, or reuse by other agent workflows.

Content

No source excerpt is available for this finding.