The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the hal-vault CLI tool to perform secret management operations such as initialization, storage, retrieval, and searching.
[EXTERNAL_DOWNLOADS]: The instructions provide commands to install the hal-vault tool via Homebrew (brew install ofoxai/tap/hal-vault) or Go (go install github.com/ofoxai/hal-vault/cmd/hal-vault@latest). These resources are hosted on the vendor's official GitHub organization and represent standard installation procedures for the tool.
[PROMPT_INJECTION]: As the skill handles arbitrary strings provided by the user (secrets), it contains an inherent attack surface for indirect prompt injection. This is mitigated by the skill's core design and instructions:
Ingestion points: Data enters the system via hal-vault add and hal-vault update commands which read from standard input.
Boundary markers: The skill uses a 'safety contract' that mandates secrets only be referenced by their masked forms in any output visible to the agent or user.
Capability inventory: The skill uses shell execution to interact with the vault CLI.
Sanitization: The hal-vault tool itself masks values by default. The instructions explicitly forbid the agent from using the --reveal flag in any way that would output the raw value to the chat or logs.

hal-vault