The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest natural conversation in real time to extract requirements and generate code. This creates a surface for indirect prompt injection where an adversary could provide instructions within the conversation to influence the logic of the generated agent.
Ingestion points: Stage 1 (Listen) extracts structured signals from natural conversation as described in SKILL.md.
Boundary markers: The instructions do not specify explicit text delimiters for the conversation intake, though it uses structured intake YAML as an intermediary.
Capability inventory: The pipeline generates and executes code via platform adapters (ServiceNow, Salesforce, Snowflake, etc.) and performs network operations via the Anthropic API as seen in references/agent-templates.md.
Sanitization: The skill includes mandatory de-identification for PHI/PII using regex patterns in references/industry-patterns.md and requires human approval gates for all irreversible actions in SKILL.md.
[COMMAND_EXECUTION]: The skill describes a pipeline (enterprise-agent-builder and enterprise-agent-shipper) that generates and potentially executes Python, JavaScript (ServiceNow), and Apex (Salesforce) code. While the provided templates in references/agent-templates.md are static, the intended primary purpose involves dynamic code generation.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to well-known enterprise platforms and services for integration purposes.
ServiceNow: Connects to *.service-now.com for incident management as seen in references/agent-templates.md.
Salesforce: Uses the simple-salesforce library to interact with Salesforce APIs.
Microsoft Graph: Connects to graph.microsoft.com for Teams and Azure integration.
Okta: Interacts with *.okta.com for IAM provisioning.
Epic FHIR: Connects to clinical instances for healthcare data.
Anthropic: Calls the Claude API (anthropic.Anthropic) for LLM reasoning.

dynamic-enterprise-agent