The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's core protocol utilizes the shell 'eval' command to execute logic defined in the 'Verify' and 'Guard' configuration fields. Evidence found in 'references/autonomous-loop-protocol.md' shows the use of 'eval "$VERIFY_CMD"' and 'eval "$GUARD_CMD"' to perform measurements and stability checks during each iteration.
[REMOTE_CODE_EXECUTION]: The skill architecture is designed to execute arbitrary code strings provided in its configuration. The 'references/metric-library.md' documentation provides several examples that utilize potentially dangerous execution methods, including 'node -e', 'python3 -c', and network-capable utilities like 'curl' and 'wrk' for metric collection.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes external data sources (git logs, diffs, and scoped files) to determine its next optimization steps without proper sanitization.
Ingestion points: Phases 1 (Review) and 2 (Ideate) in 'references/autonomous-loop-protocol.md' involve reading repository history and file contents matching the user-defined 'Scope'.
Boundary markers: The documentation does not specify the use of delimiters or explicit instructions to ignore embedded commands in the processed data.
Capability inventory: The skill possesses the ability to modify the filesystem, perform git operations, and execute arbitrary shell commands via 'eval'.
Sanitization: No sanitization or validation of external content is performed before it is processed by the agent to generate new code or commands.

autoresearch