claude-skill-orchestration-audit

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive filesystem paths to discover other installed skills.
  • Evidence: SKILL.md contains instructions to 'Glob ~/.claude/skills/*/SKILL.md' to discover skills for auditing.
  • Context: While this is necessary for the skill's primary purpose, accessing the user's home directory configuration files is a sensitive operation.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data (the content of other skills) through a sub-agent without sufficient isolation.
  • Ingestion points: SKILL.md reads the content of any SKILL.md file found in the globbed path and passes it to the 'Skill Analyzer Agent' in Step 2.
  • Boundary markers: None. The instructions in references/agent-skill-analyzer.md do not include delimiters or warnings to ignore instructions embedded within the analyzed skill content.
  • Capability inventory: The orchestrator has the ability to read/write files and launch sub-agents. The 'Agent' tool is used to execute the analysis logic.
  • Sanitization: No sanitization or escaping of the target skill's content is performed before it is processed by the LLM.
  • [COMMAND_EXECUTION]: The skill includes functionality to modify the local filesystem, specifically other skill files.
  • Evidence: The '--fix' flag in SKILL.md allows the agent to 'apply with Edit after user confirms' changes to the audited skills.
  • Context: This capability allows for modification of other agent logic, which can be risky if combined with an indirect prompt injection attack.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 03:46 PM