The Agent Skills Directory

Indirect Prompt Injection: The skill ingests untrusted data from external issue trackers, comments, and linked web pages (Step 1). While the instructions explicitly state to treat this content as data and ignore embedded directives (Boundary markers), the skill's capabilities—including modifying production code, running test suites, and drafting pull requests—create a potential surface for injection attacks. This is mitigated by the requirement for explicit human confirmation for every state-changing action.
Command Execution: The skill executes various shell-based operations, including Git commands for branch management and committing, and project-specific runtime commands for executing tests (Steps 0, 3, 5, 7, 9). These operations depend on configurations provided within the adopter's project environment.
Dynamic Execution: During its pre-flight phase, the skill invokes a local Python module (setup_preflight) from a hidden directory (.apache-magpie-local). This represents the dynamic execution of local scripts to verify the project's setup and configuration before the main workflow begins.

fix-workflow