plan
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files to inform the planning process.
- Ingestion points: The agent reads
$PluginPath/.ideas/creative-brief.mdand$PluginPath/.ideas/parameter-spec.mdin Step 1.1 to define DSP architecture and complexity. - Boundary markers: There are no instruction-level delimiters or 'ignore embedded instructions' warnings when reading these external files, increasing the risk of the agent following instructions hidden within the plugin ideas.
- Capability inventory: The skill executes shell commands (
apc standards), runs PowerShell scripts (state-management.ps1), and writes multiple markdown files to the local file system. - Sanitization: No sanitization, escaping, or validation of the ingested file content is performed before the data influences the decision-making process or is used in project status updates.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute specific CLI tools (
apc standards) and PowerShell modules (state-management.ps1). While these are intended for legitimate project configuration, an attacker could potentially influence the arguments or logic passed to these tools if they manage to inject malicious instructions into the source project files.
Audit Metadata