The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands including go test, hive x assess (file, replay, watch, drive, scenario), and mise run. These commands are used for replaying fixtures, diagnosing rules, and running integration tests.
[DATA_EXFILTRATION]: The hive x assess watch --record command captures raw bytes from a terminal pane and writes them to a local file. The skill explicitly notes that these recordings can contain sensitive information such as 'source, output, paths, and secrets'. While it does not exfiltrate this data to a remote network, it stores potentially sensitive data in an unencrypted local format.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and processing raw terminal output recordings to tune classification rules.
Ingestion points: Raw terminal content is ingested via hive x assess watch --record and processed from files in test/calibration/sequences/ and internal/domain/terminal/assess/testdata/.
Boundary markers: No specific boundary markers or 'ignore' instructions are mentioned for the ingested terminal data.
Capability inventory: The skill has the capability to execute arbitrary Go tests and repository-specific CLI tools (hive, mise) which can interact with the host or container environment.
Sanitization: There is no mention of sanitizing or escaping the captured terminal content before it is processed by the assessment rules or replayed.
[COMMAND_EXECUTION]: The drive command utilizes tmux respawn-pane -k, a destructive operation that kills and replaces the current process in a target pane. The skill describes this as 'destructive to the target pane'.
[COMMAND_EXECUTION]: The skill describes a safety gate that restricts mutating commands to container environments but provides a bypass flag (--allow-host) which allows these commands to interact directly with the host's tmux server.

cli-autocalibrate