The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including a local adoption-config.md file and the crm_ai_usage database table. This data is interpolated into a report template and a Gmail draft. This creates a surface for indirect prompt injection if the source data (such as feature names or user identifiers) is maliciously crafted.
Ingestion points: Data is pulled from the crm_ai_usage table via SQL queries and from a local adoption-config.md file in the skill's environment.
Boundary markers: The skill uses {{PLACEHOLDERS}} within a markdown template (references/adoption-report-template.md) to delimit data, but does not specify explicit escaping or sanitization logic for the interpolated content.
Capability inventory: The skill possesses the ability to read from Supabase SQL, create Google Docs, move files in Google Drive, update CRM deal notes via MCP, and create Gmail drafts.
Sanitization: No explicit sanitization or validation of the telemetry data content is described, relying on the LLM's default handling of interpolated strings.
[COMMAND_EXECUTION]: The documentation includes a command npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill consult-adoption-reporter. This command downloads and installs code from a remote GitHub repository. While this is the standard mechanism for skill distribution and the repository belongs to the skill's author, it represents the execution of remote code during the setup phase.
[REMOTE_CODE_EXECUTION]: The skill relies on external MCP tools (Supabase, Google Workspace, Zynkr) to perform its functions. While these tools are scoped and authenticated, the skill's logic involves executing SQL queries and API calls against these remote services based on instructions derived from the input and telemetry data.

consult-adoption-reporter