The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent on how to use Aider, a tool designed to read source code, generate edits, and execute shell commands. This creates a surface where an agent could potentially be influenced by malicious instructions embedded within a third-party codebase it is assigned to edit.
Ingestion points: The agent reads repository content through Aider's /add and /read commands, as well as the tree-sitter based repo-map (SKILL.md, references/R1-architecture.md).
Capability inventory: The documented tool has the capability to write to the local filesystem and execute shell commands via /run or /test (SKILL.md, references/R2-sop-workflow.md).
Boundary markers: The SOP highlights the /add command as a 'security boundary' (write-set) to limit which files the LLM can modify (SKILL.md Section 2.2).
Sanitization: The skill relies on the 'Human-in-the-loop' REPL design of Aider for manual review of proposed changes and command execution results.
[COMMAND_EXECUTION]: The skill provides documentation and examples for using Aider's command execution capabilities, such as /run <cmd> and /test <cmd>. These are legitimate features of the pair-programming tool intended for running linters, tests, and build scripts (SKILL.md Section 4).
[EXTERNAL_DOWNLOADS]: The documentation contains links to Aider's official website (aider.chat) and its GitHub repository (github.com/Aider-AI/aider). These are well-known, trusted sources for the tool described in the SOP and do not represent a security risk.

agentsop-aider