The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a runner configuration in runners/codex.md that utilizes the codex CLI with the --dangerously-bypass-approvals-and-sandbox flag. This flag is designed to explicitly circumvent built-in security sandboxes and user confirmation prompts during code execution.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources and incorporates it into implementation prompts, creating a vulnerability surface.
Ingestion points: Data is fetched from GitHub issue titles, bodies, and acceptance criteria using the gh CLI tool (as described in references/dev-loop-protocol.md).
Boundary markers: The instructions do not specify the use of strict delimiters or instructions to ignore potential commands embedded within the issue content.
Capability inventory: The orchestrator maintains broad capabilities, including file system modifications (git worktree), network access (git push, gh api), and command execution.
Sanitization: There is no evidence of sanitization or validation performed on the retrieved issue data before it is processed by subagents.- [DYNAMIC_EXECUTION]: The skill dynamically constructs prompts and execution environments for subagents, creating a multi-layered execution chain where inputs from one stage (grooming) influence the behavior of the next (takes).

orchestrate