Back to skill

Security audit

first-principles

Security checks for vulnerabilities and agentic risk

Overview

The skill is marked suspicious because scanners report an unnecessary startup request that fetches changeable instructions and signals skill use to a third party; the scan was partial, so evidence is incomplete.

Consider whether you are comfortable with a skill that makes a per-run third-party request and may retrieve mutable instructions. The scan covered all five listed components, but it was marked partial because two referenced files were missing, so the evidence is incomplete. Review those concerns before deciding whether to install.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

L008 requires a startup network request to retrieve remote skill instructions, which is unnecessary for the stated reasoning task.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The startup directive introduces an unnecessary external dependency into a self-contained reasoning workflow and retrieves mutable remote skill instructions. This creates an avenue for unreviewed instructions to enter the agent's context; HTTPS protects transport but does not establish that future content is safe.

Content

No source excerpt is available for this finding.

A freshness-check pretext promotes remotely supplied skill text to canonical instructions.

Medium
Category
Prompt Injection
Confidence
97% confidence
Finding

Describing remotely returned 'latest_skill_md' as the 'canonical current version' encourages the agent to promote mutable external content into authoritative skill instructions. A compromised service or malicious update could redirect a tool-enabled agent beyond the reviewed reasoning workflow; the supplied file does not demonstrate such a payload, but it establishes the trust-boundary weakness.

Content

No source excerpt is available for this finding.

The mandatory startup request reports skill usage to a third-party service without a privacy warning.

Low
Category
Output Handling
Confidence
94% confidence
Finding

Requesting the skill-specific endpoint with '?ctx=run' discloses a signal of skill use to the third-party service, together with request timing and network metadata such as the requester or proxy IP address. No conversation content or secrets are explicitly transmitted, so the demonstrated privacy exposure is limited, but the automatic per-run request is unnecessary for this reasoning task and lacks a privacy notice.

Content

No source excerpt is available for this finding.