Back to skill

Security audit

CEO Capabilities Catalog

Security checks for vulnerabilities and agentic risk

Overview

This skill is a simple capabilities catalog, but it asks to read unspecified local workspace memory that could expose private context.

Review before installing if your workspace memory may contain private project, customer, workflow, or personal context. Use only when you are comfortable with the agent reading local memory files for summaries, or narrow the instructions to explicit allowed files and fields.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:15
Finding

Overly Broad Access to Local Workspace Memory

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 15–21
Vulnerability Type: Excessive access to potentially sensitive local workspace data
Risk Level: Medium

Vulnerable Code Snippet:

markdown
- Current connected workflow highlights if known from local workspace memory

## Source of truth

Read from:

- Local workspace memory files

Technical Analysis

The skill instructs the agent to read unspecified “local workspace memory files” and use workflow information found there when producing a capabilities summary. This grants the task a broader data-access scope than is necessary to inventory installed skills.

The instruction does not constrain access to specific paths, files, fields, or data types. It also does not require explicit user consent before inspecting memory that may contain conversation-derived state, operational details, personal information, or unrelated workflow context. Although SKILL.md line 28 prohibits exposing secrets, that instruction does not prevent excessive collection or disclosure of other sensitive information.

The issue is classified as T05: Unauthorized Access and Privilege Escalation because the skill directs the agent to cross a least-privilege boundary by accessing local workspace information beyond the minimum data required for its stated cataloging purpose. It does not itself grant operating-system privileges; exploitation is limited to files already readable by the executing agent.

Attack Path

  1. A user or untrusted prompt invokes the skill by requesting an inventory of the agent’s capabilities.
  2. Following the skill instructions, the agent searches unspecified local workspace memory files.
  3. Those files may contain sensitive workflow details, conversation-derived state, personal data, or information unrelated to the capability inventory.
  4. The agent incorporates some of that information into the generated summary.
  5. The requesting party re ...[truncated 545 chars]
Remediation
View remediation

Remediation Suggestions

  • Remove access to local workspace memory from the default capability-inventory workflow.
  • Use installed SKILL.md files and the dedicated capabilities-inventory.md backup as the primary, narrowly scoped sources.
  • If workspace memory is genuinely required, obtain explicit user consent before accessing it.
  • Define an allowlist of exact files, directories, and fields that may be read.
  • Apply data minimization by extracting only capability names and non-sensitive descriptions.
  • Exclude conversation content, credentials, API keys, personal data, customer information, and unrelated workflow details.
  • Add a mandatory redaction step before presenting any information derived from local memory.
  • Record the sources used in the resulting summary so users can verify that access remained within the approved scope.
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 (1)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 23)May include surrounding context.

md
- Installed `SKILL.md` files when needed