agent-orchestration-improve-agent
Warn
Audited by Gen Agent Trust Hub on Aug 4, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: In Phase 1.1, the skill defines a tool invocation command:
analyze-agent-performance $ARGUMENTS --days 30. The use of an unquoted$ARGUMENTSplaceholder is a command injection vector. If user-supplied input is interpolated into this command string in a shell context, an attacker can use shell metacharacters (e.g.,;,|,&) to execute arbitrary system commands. - [PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface by instructing the agent to analyze user interaction patterns and feedback in Section 1.2. This data is then used by the
prompt-engineertool in Phase 2 to modify agent behavior. Malicious content within the ingested feedback could influence the generated prompts, leading to compromised or backdoored agent instructions. - Ingestion points: Performance data gathered via
context-managerand user interaction patterns described in Section 1.2. - Boundary markers: Absent. The instructions do not specify the use of delimiters or provide warnings to ignore embedded instructions in the analyzed data.
- Capability inventory: The skill utilizes
prompt-engineer(capable of modifying system prompts),context-manager, andparallel-test-runner. - Sanitization: Absent. No instructions for validating, filtering, or escaping the analyzed feedback are provided.
Audit Metadata