magpie-setup
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads framework snapshots and dependencies from official Apache Software Foundation sources, including the
apache/magpieandapache/comdevGitHub repositories, as well as ASF distribution servers for signed releases. - [REMOTE_CODE_EXECUTION]: Executes internal scripts and tools (e.g.,
permission-auditandagent-guard.py) from the downloaded framework snapshot using Python anduv run. - [COMMAND_EXECUTION]: Manages project structure through various system commands, including symlink creation (
ln -s), file system operations (cp,rm,mkdir), and the installation of a gitpost-checkouthook. - [DATA_EXFILTRATION]: Ingests repository metadata (such as issue and pull request counts/dates) via the
ghCLI to provide 'fit signal' recommendations for skill family selection. The skill explicitly instructs the agent to treat this data as non-executable input. - [PROMPT_INJECTION]: Supports 'agentic overrides' where instructions in
.apache-magpie-overrides/*.mdfiles are read at runtime to customize framework skill behavior. The skill also implements a self-update mechanism that reloads its own instructions in-flight after updating its committed source code. - [SAFE]: The skill utilizes platform capabilities (capability:platform) to perform legitimate configuration management tasks, such as updating sandbox allow-lists in
.claude/settings.local.jsonwith user confirmation.
Audit Metadata