The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to circumvent platform security controls. It advises the agent that if a command is blocked by a sandbox (resulting in runtime_access_denied), it should "re-run it with escalated permissions". This encourages the agent to seek higher privilege levels to bypass environment restrictions.
[DYNAMIC_EXECUTION]: The skill implements a pattern where the primary instructional logic is not contained within the skill file itself but is fetched dynamically at runtime. The agent is instructed to run ORCA skills get orca-ask to retrieve the "complete, version-matched guide" and then follow those instructions, which allows for the execution of instructions that were not present during the initial security review.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands using various local binaries (orca, orca-ide, orca-dev) determined by environment variables like ORCA_CLI_COMMAND and ORCA_DEV_REPO_ROOT.
[INDIRECT_PROMPT_INJECTION]: The skill's core functionality is to ingest untrusted data from a human user through prompt windows (text, select, number). Since the agent is instructed to block and wait for these answers to determine its next actions, it creates a surface for indirect prompt injection where a user could provide instructions that override the agent's current task or exploit its command execution capabilities.

orca-ask