The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions explicitly direct the agent to operate in 'AUTONOMOUS MODE' and 'Do NOT ask questions'. This is a directive to bypass standard user interaction, confirmation steps, and oversight throughout the execution of the integration pipeline.
[COMMAND_EXECUTION]: The skill executes project-specific commands such as test suites, build processes, and linters (e.g., npm test, npm run build) during its validation phase. Since these commands are defined within the project being analyzed (e.g., in package.json), it could lead to the execution of arbitrary code if the project contains malicious script definitions.
[DATA_EXFILTRATION]: The discovery phase involves scanning for the presence and values of sensitive environment variables, including STRIPE_SECRET_KEY, GOOGLE_CLIENT_ID, and AUTH_SECRET. While the script does not contain explicit network exfiltration logic, it harvests these credentials into the agent's active context.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted project metadata (frameworks, existing integrations) to determine its logic flow and routing. Maliciously crafted project files could influence the agent's behavior or trigger unintended sub-skill execution.
Ingestion points: Processes content from package.json, requirements.txt, .env, and other configuration files (SKILL.md).
Boundary markers: Absent. The skill lacks delimiters or instructions to ignore potential commands embedded within the data it scans.
Capability inventory: Executes sub-skills, runs shell commands for testing/building, and writes telemetry data to ~/.claude/projects/.
Sanitization: No evidence of sanitization or validation of extracted project metadata before its use in command execution or routing decisions.

integrate