The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides multiple mechanisms to execute shell commands as part of its coordination workflow. Tools such as pd watch --exec, pd with-lock, and the 'fleet' background agent system are designed to run local scripts or commands in response to events like git commits or pub/sub messages.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by allowing agents to share data (notes, tuples, and pheromones) that is subsequently ingested by other agents. This creates a vector where malicious or compromised agents could influence the behavior of others.
Ingestion points: Agents ingest data from shared sources via pd sitrep, pd notes, pd salvage, pd pheromone sniff, and pd tuple rd (as seen in SKILL.md and schemas/mcp-tool-catalog.md).
Boundary markers: The instructions do not specify or recommend the use of delimiters or 'ignore' instructions when processing data originated from other agents.
Capability inventory: The skill possesses significant capabilities including arbitrary command execution (pd watch), background agent spawning (pd spawn), and the installation of git hooks (pd fleet init).
Sanitization: There is no evidence of sanitization, validation, or filtering of the content within notes or tuples before they are presented to or processed by other agents.
[EXTERNAL_DOWNLOADS]: The architecture.html file fetches the Mermaid.js library from the well-known cdn.jsdelivr.net service. This is used neutrally to render architectural diagrams within the provided documentation and follows established safe practices for documentation assets.

port-daddy