The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon instructions provided in external data files generated or managed by the 'ax' tool environment.\n
Ingestion points: The agent reads task playbooks from .ax/tasks/*.md and command agendas via ax dojo agenda --json.\n
Boundary markers: The skill instructions do not specify any delimiters or warnings to ignore potentially malicious instructions embedded within these external briefs or agendas.\n
Capability inventory: The agent is authorized to execute shell commands via the ax CLI tool, perform git operations such as git worktree add, and write various files (proposals, reports, goals) to the local file system.\n
Sanitization: The skill lacks explicit logic for sanitizing or validating the content of the briefs or agendas before the agent processes the embedded instructions.\n- [COMMAND_EXECUTION]: The skill requires the execution of multiple local CLI commands as part of its core functionality.\n
Evidence: It frequently invokes ax (axctl) subcommands (e.g., ax dojo agenda, ax improve verdict, ax routing tune) and git commands (e.g., git worktree add). These tools are essential for the skill's stated purpose of managing the 'ax' graph and performing automated development tasks.

dojo