The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external email content, which serves as an untrusted data source for the agent.
Ingestion points: The agent is configured with a native email trigger in infra/agents/engager.ts that wakes the agent on replied events, and it utilizes the listEmailEvents capability to read thread history.
Boundary markers: The system instructions in infra/agents/engager.prompt.ts do not define explicit delimiters or instructions for the agent to ignore potential malicious prompts embedded within lead emails.
Capability inventory: The agent has access to sendEmail and listEmailEvents native actions, as well as memory and context tools.
Sanitization: There is no evidence of sanitization or filtering of the email body content before it is processed by the language model.
[DYNAMIC_EXECUTION]: The evaluation script evals/contract.mjs utilizes dynamic imports to load and inspect the resource graph.
Evidence: await import("../infra/agents/engager.ts?contract=${Date.now()}"); in evals/contract.mjs.
Context: This is a local import used for development-time resource validation and does not involve loading code from external or untrusted sources.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of specific CLI tools and SDKs provided by the platform vendor.
Evidence: SKILL.md references the installation of @cargo-ai/cli and dependency on @cargo-ai/cdk.
Context: These are official resources from the skill author (Cargo AI) and are necessary for the deployment and management of the infrastructure defined in the skill.

agentic-engagement