The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted content from Slack messages and Freshdesk ticket metadata to drive its triage logic and autonomous resolution workflow. It lacks instructions for sanitizing these inputs or using boundary markers to prevent malicious instructions embedded in tickets from influencing agent behavior.\n
Ingestion points: Slack monitored channels (#customer-support, #product-diagnosis-support) and Freshdesk ticket API responses (subject, narrative, and requester fields).\n
Boundary markers: Absent; untrusted user content is processed directly to populate fields like 'User Problem' and 'Code Investigation'.\n
Capability inventory: Execution of shell commands (curl), Git operations (worktree creation, branch pushing), GitHub PR creation, and Notion database writes.\n
Sanitization: No sanitization or validation of Slack/Freshdesk content is mentioned before it is interpolated into investigations or used to trigger autonomous PR creation.\n- [COMMAND_EXECUTION]: The skill provides instructions to execute shell commands using variables sourced from external data. Specifically, it uses a ticket ID from Slack/Freshdesk mirrors in a curl command: curl -s -u "$FRESHDESK_API_KEY:X" "https://$FRESHDESK_DOMAIN/api/v2/tickets/<ticket-id>?include=requester". If the ticket ID contains shell metacharacters, it could lead to command injection.\n- [CREDENTIALS_UNSAFE]: The skill references the use of FRESHDESK_API_KEY and FRESHDESK_DOMAIN from the environment. While it correctly warns against storing these in the prompt or repository, using them in shell interpolation can lead to credential exposure in environment logs or through the aforementioned command injection vector.

support-ticket-triage