The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow where the agent autonomously reads codebase files, including project-specific conventions (e.g., CONVENTIONS.md). This creates an attack surface where malicious instructions embedded in a project's source code could potentially influence the agent's behavior during autonomous sessions.
Ingestion points: The tool reads the local codebase and specific context files like CONVENTIONS.md (documented in SKILL.md).
Boundary markers: The skill does not provide instructions on using delimiters or markers to separate untrusted codebase content from agent instructions.
Capability inventory: The documented tool has the capability to modify files, create git commits, and execute arbitrary shell commands via the --lint-cmd and --test-cmd flags.
Sanitization: No sanitization or validation steps are provided for the content read from the codebase before it is processed by the agent.
[COMMAND_EXECUTION]: The skill highlights the use of autonomous execution modes (--yes-always) coupled with arbitrary shell commands for linting and testing (--lint-cmd, --test-cmd). While this is a core feature of the tool, it represents a high-capability execution environment.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions to install the aider-chat package from the Python Package Index (PyPI). aider-chat is the standard and well-known package for this utility.

aider