The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/kanban.sh executes several local scripts and vendor-provided binaries to manage the task lifecycle. Key operations include running controller_resolver.py for path resolution, install_requirements.sh for environment initialization, and invoking the yylo-ledger or juno-kanban executables from a virtual environment.\n- [DYNAMIC_EXECUTION]: The skill employs the eval command within scripts/kanban.sh to process the output of a local Python helper script (controller_resolver.py). This is used to dynamically inject environment variables into the shell session based on project configuration.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could contain instructions designed to influence the agent.\n
Ingestion points: User-provided $ARGUMENTS in SKILL.md and external task records retrieved via yy task start in references/implement.md.\n
Boundary markers: Instructions tell the agent to "Preserve task references and directives literally", which provides context but may lead to the agent following embedded instructions found within the task data.\n
Capability inventory: The skill can execute shell commands through the kanban wrapper, modify product code in the worktree, and interact with the git ledger.\n
Sanitization: There is no documented mechanism for sanitizing or filtering instructions that may be embedded in the externally assigned task records.\n- [EXTERNAL_DOWNLOADS]: The skill includes a dedicated script for environment setup that likely downloads external dependencies. The scripts/kanban.sh file executes install_requirements.sh when the Python virtual environment (.venv_juno) is missing, which is a common pattern for installing remote packages.

ralph-loop-yylo