The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell command execution to interact with Git repositories and a custom orchestration tool.
Evidence: Uses git -C '<path>' rev-parse and git -C '<root>' status for repository management and validation.
Evidence: Invokes the ai-coord CLI tool for orchestration operations including bundle start, wait, and done.
Evidence: Instructs agents to run 'validation commands' discovered within repository configuration files such as justfile, package.json, AGENTS.md, or CLAUDE.md. This executes commands defined by the content of the target repositories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external repositories and other skills to define its execution plan, creating a surface for injection attacks.
Ingestion points: The skill reads contents from target repositories (AGENTS.md, CLAUDE.md, justfile, package.json), external skill definitions (SKILL.md), and a user-provided --repos-file input.
Boundary markers: The instructions include requirements for a 'strict read-only boundary' for discovery agents and an 'authority boundary' for implementation agents to restrict their capabilities.
Capability inventory: Capabilities include file writes within a defined scope, execution of git and ai-coord commands, and running validation scripts found in the processed repositories.
Sanitization: The skill does not explicitly describe sanitization or validation of the extracted commands or instructions before they are executed or interpolated into sub-agent briefs.
[EXTERNAL_DOWNLOADS]: The skill specifies a requirement for the ai-coord CLI tool as a mandatory external dependency for its orchestration and delegation features.

repo-fanout