The Agent Skills Directory

[SAFE]: The skill establishes a rigid protocol for status reporting within an autonomous agent framework. By mandating a machine-parseable block, it enhances the transparency and observability of the agent's internal loop states.
[COMMAND_EXECUTION]: To gather the necessary metrics for its status reports, the skill instructs the agent to execute standard development commands, including running a test suite and performing a git diff to track file modifications. These actions are limited in scope to the agent's workspace and are essential for its primary function.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface due to its data ingestion requirements.
Ingestion points: The skill reads external data from IMPLEMENTATION_PLAN.md to track task completion, file modification logs from version control, and output from test execution to calculate status metrics.
Boundary markers: While the skill provides clear delimiters for its output (---RALPH_STATUS---), it does not specify explicit boundary markers or safety instructions for the input files it processes.
Capability inventory: The agent's capabilities in this context are limited to reading local files and executing pre-defined testing and version control commands. No network access or arbitrary file writing is involved.
Sanitization: The resulting status block is highly structured, using predefined enums and integers for most fields, which significantly restricts the impact of any potentially injected content. The free-text RECOMMENDATION field is strictly limited to a single line.

ralph-status