The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to bypass sandbox security controls. It states: "If it fails with runtime_access_denied, your sandbox blocked the connection: re-run it with escalated permissions". This encourages the agent to seek higher privileges than those originally allocated by the environment.
[COMMAND_EXECUTION]: The skill relies on executing local command-line tools such as orca, orca-ide, or orca-dev to manage environments, terminals, and browsers. It provides logic for selecting the appropriate binary based on environment variables like ORCA_CLI_COMMAND and ORCA_DEV_REPO_ROOT.
[INDIRECT_PROMPT_INJECTION]: The skill dynamically fetches and loads a version-matched guide via the command ORCA skills get orca-cli. This content is treated as instructions for the agent, creating a surface where a local executable or its source could inject malicious instructions into the agent's context.
Ingestion points: The output of ORCA skills get orca-cli is ingested into the session (SKILL.md).
Boundary markers: No boundary markers or "ignore embedded instructions" warnings are used when processing the fetched guide.
Capability inventory: The skill description indicates it can manage terminals, repos, worktrees, and an embedded browser (SKILL.md).
Sanitization: No sanitization or validation of the dynamically fetched content is performed before processing.

orca-cli