The Agent Skills Directory

[PROMPT_INJECTION]: The skill interacts with an external data source (HashiCorp Vault KV engine). While the skill documentation claims secret values are never displayed to the user, the agent's ingestion of this untrusted external data represents a surface for indirect prompt injection.
Ingestion points: Vault KV secrets via read_secret and inject_secret (SKILL.md).
Boundary markers: The documentation includes a 'CRITICAL' warning that secret values are never displayed, which serves as a prompt-level constraint (SKILL.md).
Capability inventory: The skill allows reading, writing, and deleting secrets, as well as injecting them into unspecified integrations (SKILL.md).
Sanitization: No explicit sanitization logic is provided in the manifest to handle potentially malicious instructions embedded within secret values.
[DATA_EXPOSURE]: The skill is explicitly designed to handle sensitive credentials. It implements a protection mechanism where only metadata is returned in responses, which is a security best practice for secret management tools.
[CREDENTIALS_UNSAFE]: The documentation references VAULT_TOKEN and VAULT_ADDR as required prerequisites. These are standard requirements for HashiCorp Vault and the skill appropriately guides the user to provide them via environment variables rather than hardcoding them.

vault-secrets