The Agent Skills Directory

[COMMAND_EXECUTION]: The relay.mjs script manages the lifecycle of the aider CLI and git processes. It utilizes spawn and execFileSync to run these tools and taskkill (on Windows) or process.kill (on POSIX) to ensure clean termination of process groups.
[DYNAMIC_EXECUTION]: The relay.mjs script dynamically constructs the path to a sibling configuration script (../../delegate-setup/scripts/lane.mjs) and executes it using spawnSync when the --lane option is used. This allows for modular configuration resolution within the toolset.
[EXTERNAL_DOWNLOADS]: The skill relies on the aider-chat package, which is a well-known tool. The documentation guides the user to install it via standard package managers. Aider naturally connects to external AI model providers (such as OpenAI or Anthropic) or local API endpoints to perform its tasks.
[INDIRECT_PROMPT_INJECTION]: The skill processes task briefs and codebase content through the Aider implementer, which creates a potential surface for indirect prompt injection. A malicious file in the repository could attempt to influence Aider's output. The skill explicitly mitigates this by enforcing a manual diff review and gate verification process before code is committed.
Ingestion points: Task briefs provided via file or stdin, and the repository files Aider reads into its context.
Boundary markers: The documentation recommends a structured brief format using headers like GOAL, CONTEXT, and DO NOT TOUCH in references/writing-the-brief.md.
Capability inventory: The skill facilitates file system modifications and execution of project test/lint suites (gates) through Aider in scripts/relay.mjs.
Sanitization: No automated sanitization of the brief is performed; the system relies on the mandatory human review of the generated diffs as the primary gate.

aider-delegate