The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to guide the agent in executing a compiled third-party binary (anywrite) via shell commands to interact with the Anytype local API. While it requires the user to provide an absolute path and explicitly forbids executing binaries within the current workspace (e.g., ./dist/anywrite), it remains a high-capability command execution surface.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data ingested from the Anytype API (e.g., chat messages, object bodies, and search results).
Ingestion points: API responses from commands like objects get, search, and chat messages in SKILL.md.
Boundary markers: The instructions recommend resolving names to IDs and verifying state after writes, but lack explicit delimiters for external content returned by the CLI.
Capability inventory: The skill allows the agent to execute shell commands (anywrite CLI) with broad capabilities (52 endpoints), including creating, updating, and deleting data.
Sanitization: No specific sanitization or escaping mechanisms are described for handling data fetched from the API before it enters the agent's context.
[DATA_EXFILTRATION]: The skill interacts with sensitive local data including the Anytype knowledge base and local configuration files located at ~/.anywrite/config.json. There is a risk of data exposure if the agent is prompted to exfiltrate content from these resources, although the skill instructions explicitly advise against printing the API key.

anywrite