The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for the agent to install its core engine by downloading and executing a shell script directly from the author's GitHub repository via a pipe to shell (e.g., curl -fsSL https://raw.githubusercontent.com/Rheinmir/orca-graph/main/install.sh | bash).
[COMMAND_EXECUTION]: The skill executes arbitrary shell commands provided in user-defined PLAN.md files through its run, verify, and QC functions, allowing for the execution of subprocesses based on untrusted data.
[PERSISTENCE]: The skill mentions a background daemon process (watch) intended to monitor task execution and manage leases, which can maintain code execution across sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted instructions from PLAN.md files which are used to control execution flow and define shell commands.
Ingestion points: Processes PLAN.md files to extract task metadata, dependencies, and execution scripts.
Boundary markers: Uses specific markdown headers and key-value pairs (e.g., **Depends:**, **Verify:**) to identify and isolate task properties.
Capability inventory: Possesses capabilities to execute shell commands, monitor system processes, and perform file system operations within the specified store directory.
Sanitization: Implements an optional --strict mode that monitors for unauthorized file changes during task execution using Git diffs and attempts to revert them if they were not declared in the task specification.

orca-graph