The Agent Skills Directory

Command Execution: The skill instructs the agent to execute local commands for environment verification and project state management.
Evidence: The agent is directed to run python3 -m setup_preflight for pre-flight checks and gh api to verify the existence of remote tags.
Indirect Prompt Injection: The skill ingests untrusted external data, such as planning issue bodies and build configurations, which could contain malicious instructions aimed at overriding agent behavior.
Ingestion points: SKILL.md specifies reading planning-issue bodies and external build-config files (release-build.md, release-management-config.md).
Boundary markers: The skill includes explicit instructions to treat external content as data only and to flag any potential injection attempts.
Capability inventory: The agent possesses the capability to write back to external systems by posting comments via gh issue comment.
Sanitization: Instructions are provided to the agent to ensure external content is treated strictly as data, though formal sanitization of the input string is not specified.
Dynamic Instruction Loading: The skill allows for the application of overrides from local files, which modifies the agent's behavior and logic at runtime based on external file content.
Evidence: The "Adopter overrides" section instructs the agent to consult .apache-magpie-local/release-rc-cut.md and .apache-magpie-overrides/release-rc-cut.md and apply any agent-readable overrides found within.

rc-cut