skills/yonatangross/orchestkit/freeze/Gen Agent Trust Hub

freeze

Fail

Audited by Gen Agent Trust Hub on Oct 2, 2026

Risk Level: HIGHDYNAMIC_CONTEXT_INJECTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [DYNAMIC_CONTEXT_INJECTION]: The skill utilizes Claude Code's dynamic context syntax (!command) in SKILL.md to execute a script whenever the skill is invoked. It passes the $ARGUMENTS variable, which contains user-supplied input, directly into a shell command: node "${CLAUDE_SKILL_DIR}/scripts/freeze-guard.mjs" arm "${CLAUDE_PROJECT_DIR}" "${CLAUDE_SESSION_ID}" "$ARGUMENTS". Because this input is interpreted by the shell, a malicious user or an indirect prompt injection attack could execute arbitrary shell commands by using metacharacters such as backticks (`), command substitutions ($()), or command separators (;, &&).
  • [COMMAND_EXECUTION]: The vulnerability in the dynamic context mechanism facilitates arbitrary command execution on the host system. The skill's documentation explicitly warns that characters like $, backticks, or double quotes in a directory name would be interpreted by the shell, yet no technical validation or sanitization is implemented in the shell command construction to prevent exploitation.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests untrusted data via $ARGUMENTS which is then used in a high-privilege shell execution context.
  • Ingestion points: The SKILL.md file ingests user-supplied arguments through the $ARGUMENTS variable in the dynamic context invocation.
  • Boundary markers: There are no boundary markers, escaping, or delimiters applied to the user input within the shell command template.
  • Capability inventory: The skill invokes shell commands (node) that can be manipulated to execute arbitrary payloads.
  • Sanitization: No technical sanitization or validation of the $ARGUMENTS content is performed before interpolation into the shell command.
  • [METADATA_POISONING]: The skill's metadata identifies the author as 'OrchestKit', which is inconsistent with the expected author context 'yonatangross'. This discrepancy in author identification can be misleading to users regarding the source and security posture of the skill.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 2, 2026, 03:46 AM