The Agent Skills Directory

[COMMAND_EXECUTION]: The skill accepts and executes arbitrary shell commands provided by the user (the METRIC_COMMAND parameter) to measure performance. These commands are executed repeatedly within an autonomous iteration loop (SKILL.md).
[DYNAMIC_EXECUTION]: The skill autonomously generates and applies modifications to source code files within the user-defined IN_SCOPE directories. The agent then executes this modified code to verify metrics, representing a dynamic code generation and execution cycle (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill supports a WEB_SEARCH feature that enables the agent to access external network resources for research and documentation. This capability allows for the retrieval of data from the public internet during the optimization process (SKILL.md).
[INDIRECT_PROMPT_INJECTION]: The skill's workflow creates an attack surface for indirect prompt injection by processing external data as part of its autonomous loop.
Ingestion points: The agent ingests output from shell commands (the results of METRIC_COMMAND saved in LOG_FILE) and the contents of files specified in the IN_SCOPE parameter (SKILL.md).
Boundary markers: No specific delimiters or boundary markers are employed to isolate and identify potentially untrusted data from the agent's core instructions.
Capability inventory: The skill has access to the terminal, file system modification capabilities, shell execution, and optional network access via web searching (SKILL.md).
Sanitization: The instructions do not specify any sanitization, filtering, or validation steps for the command output or file contents before they are analyzed by the agent (SKILL.md).

autoresearch