The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install additional components using the command npx -y skills add ElZaWarudo/krt, which fetches external code from the author's vendor repository.
[COMMAND_EXECUTION]: The orchestration instructions include shell commands for asset management and environment preparation, such as directory creation (mkdir) and file copying (cp).
[COMMAND_EXECUTION]: The worker role is granted the capability to execute arbitrary verification commands discovered or generated during the planning phase of the pipeline.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection (Category 8) as it ingests untrusted planning data to drive its implementation and execution sub-agents.
Ingestion points: The agent processes content from docs/roadmaps/, docs/brainstorms/, and docs/plans/ during its orchestration loop.
Boundary markers: The skill does not employ explicit delimiters or delimiters to isolate ingested planning artifacts from the system instructions.
Capability inventory: The worker sub-agent configuration in assets/codex-agents/compound-worker.toml permits workspace-wide write access and shell command execution.
Sanitization: There is no implemented sanitization or validation of the roadmap or plan artifacts before they are used to generate code implementation commands.

krt-compound-master