The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes shell commands using the maestro CLI (e.g., maestro ralph next, maestro ralph complete, maestro ralph retry) to manage workflow sessions and fetch step definitions.
[COMMAND_EXECUTION]: The skill employs a 'self-invocation chain' where it calls itself (Skill(maestro-ralph-execute)) at the end of successful steps, creating an autonomous loop that continues until a session is completed or paused.
[COMMAND_EXECUTION]: The skill implements 'inline execution,' instructing the agent to treat the standard output of the maestro ralph next command as direct instructions to be followed. This is a form of dynamic instruction execution based on local tool output.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the way it processes external data.
Ingestion points: The skill reads data from .workflow/.maestro/*/status.json, state.json, and various workflow artifacts such as conclusions.json, TASK-*.json, verification.json, and error traces.
Boundary markers: While the skill uses <goal_context> tags to wrap some injected goal data (for specific protocol versions), it does not consistently apply delimiters or 'ignore' instructions to all ingested artifact data.
Capability inventory: The skill has access to powerful tools including Bash, Read, Write, Edit, Glob, and Grep.
Sanitization: There is no evidence of content validation or sanitization for the artifacts and log files before their content is interpolated into the agent's context or used to drive execution decisions.

maestro-ralph-execute