The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from Basecamp comments, messages, card descriptions, and boosts, which are used to drive agent actions. Ingestion points: External data enters the agent context via Basecamp recordings, chat lines, and boost metadata. Boundary markers: The skill instructions tell the agent to distinguish between 'operators' and 'participants,' treating participant text as input rather than authoritative instructions. Capability inventory: The skill dispatches background agents with capabilities to execute shell commands via the basecamp CLI, git, gh, and repository-specific scripts, as well as the ability to write files and interact with remote APIs. Sanitization: The connector script bin/connect attempts to mitigate malicious input by re-fetching data directly from the Basecamp API to ensure authenticity before passing it to the agent stream.- [COMMAND_EXECUTION]: The skill executes local scripts and CLI utilities to manage the bridge and perform tasks. Evidence: Invokes bin/connect, bin/setup, and bin/ci within the vendor's repository. It also uses basecamp, git, gh, and tailscale CLI tools to perform operations.- [EXTERNAL_DOWNLOADS]: The skill may download the agent connector source code from the vendor's official GitHub repository if it is not found locally. Evidence: Directs the user to clone the basecamp/basecamp-local-agent-connector repository to set up the runtime.- [DATA_EXFILTRATION]: The skill configures a public network endpoint using Tailscale Funnel to allow the Basecamp service to send webhooks to the local machine. Evidence: Uses tailscale funnel to expose local services to the internet for webhook delivery, creating a network surface for external traffic to reach the local environment.

basecamp-connect