The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where user-provided goals or external tool outputs can influence the execution of shell commands and downstream skills.
Ingestion points: The goal parameter in SKILL.md and the output of the ao search and ao lookup tools in Step 2.
Boundary markers: Absent. The skill does not use delimiters or instructions to the agent to treat the goal string as untrusted data.
Capability inventory: The skill executes shell commands using ao, bd, and local scripts in the scripts/ directory, and it invokes other skills using the Skill() mechanism.
Sanitization: No sanitization or escaping of the goal string is specified before its interpolation into command strings such as ao search "<goal keywords>".
[COMMAND_EXECUTION]: The skill relies on shell command execution to interface with local CLI tools (ao, bd) and utility scripts (scripts/checkpoint-commit.sh, scripts/log-telemetry.sh) for telemetry and state management. While these are likely intended for administrative tasks, they serve as the execution vector for the identified injection surface.

discovery