The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface within the alert root-cause analysis (RCA) procedures by ingesting data from external Slack and Splunk logs.
Ingestion points: In references/alert-rca.md, the agent is instructed to fetch Slack channel history and alert payloads (e.g., "Error Breakdown" text) which may contain content from downstream service logs or user-controlled inputs.
Boundary markers: The instructions provide specific parsing logic and structural steps for the RCA process, acting as implicit boundaries for the agent's behavior.
Capability inventory: The skill has access to network operations (Adobe APIs, Slack, Splunk) and local filesystem operations via the upload_asset and download_output tools.
Sanitization: While the procedure directs the agent to filter and group alerts based on specific fields, there is no explicit mention of sanitizing the raw text before it is processed by the agent.
[COMMAND_EXECUTION]: The skill manages filesystem permissions through a specific tool designed to interact with the OS sandbox.
Evidence: references/asset-input.md mandates the use of the request_folder_access tool when encountering EPERM or permission denied errors. This tool triggers a native folder picker to grant the agent access to a local directory, which is a standard security mechanism for local MCP servers.
[DATA_EXFILTRATION]: The skill performs network operations to official Adobe endpoints and common developer infrastructure (Slack and Splunk).
Evidence: The skill utilizes tools such as run_workflow_submit, splunk_rw_search, and slack_reply_to_thread. These operations involve transmitting session IDs, workflow definitions, and diagnostic summaries to external services, which is consistent with the skill's primary purpose for workflow management and service monitoring.

run-workflow