The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions and templates for the agent to construct and execute shell commands using the claude CLI tool to perform delegated implementation and review tasks.
[PRIVILEGE_ESCALATION]: The skill directs the agent to use the --permission-mode bypassPermissions and --dangerously-skip-permissions flags. These flags are specifically designed to suppress interactive user approval prompts for sensitive operations such as file system modifications, repository commits, and remote actions. This effectively escalates the agent's autonomy by removing established safety constraints.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates a multi-step delegation workflow where one agent's instructions influence the behavior of another, and the results are then ingested back into the primary context.
Ingestion points: Data enters the context during the verification phase in SKILL.md when the agent reads the current worktree and run summary.
Boundary markers: The instructions lack formal data boundary markers for the ingested changes, relying instead on natural language instructions for the user to inspect the delta.
Capability inventory: The skill allows for file edits, commits, pushes, and the opening of artifacts as described in the assignment modes in references/collaboration-patterns.md.
Sanitization: There is no automated sanitization or filtering of the content produced by the delegated run before it is evaluated by the agent.

working-with-claude