The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from external Orca sessions, creating a surface for indirect prompt injection.\n
Ingestion points: The skill reads JSON session data via the orca CLI tool within scripts/mo-watchdog.sh.\n
Boundary markers: Structural boundaries are enforced using jq schema validation in the validate_target and stable_snapshot functions, which verify field types and presence to prevent malformed data from influencing behavior.\n
Capability inventory: The skill can execute commands to send messages to sessions (orca orchestration send, orca terminal send) and write state digests to the local filesystem for deduplication.\n
Sanitization: The classification logic operates on structured fields rather than arbitrary text, and message deduplication via SHA-256 digests prevents redundant delivery of potentially malicious content.\n- [COMMAND_EXECUTION]: The skill executes the orca CLI tool to interact with backend session state.\n
Evidence: scripts/mo-watchdog.sh invokes various orca subcommands to list, show, and send data to sessions.\n
Context: Commands are executed with quoted arguments and input is constrained by pattern matching to ensure session IDs conform to expected prefixes such as ctx_ or term_.

mo-watchdog