The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze untrusted data from external sources, specifically email bodies and Delivery Status Notifications (DSNs), to classify bounce reasons and automate response actions.
Ingestion points: The skill explicitly directs the agent to "Pull the DSN from the bounced lead's message history and take the first [45].\d+.\d+ code in the body" (found in SKILL.md).
Capability inventory: The triage process leads to the execution of management scripts (scripts/tag-inboxes.ts and scripts/set-warmup.ts) that modify inbox configurations.
Boundary markers: There are no instructions providing delimiters or warnings to ignore potentially malicious instructions embedded within the ingested email content.
Sanitization: The skill lacks explicit sanitization steps for the DSN content before it is used to drive triage decisions.
[COMMAND_EXECUTION]: The skill defines shell command patterns for the agent to follow when retiring or modifying inboxes, involving the execution of local TypeScript scripts.
Evidence: Commands such as npx tsx scripts/tag-inboxes.ts --ids=<id> --add-tag=retired and npx tsx scripts/set-warmup.ts --mode=disable --ids=<id> are provided in the "Retire workflow" section of SKILL.md.
[DYNAMIC_EXECUTION]: The skill utilizes npx tsx to execute local TypeScript scripts at runtime. While these scripts appear to be internal tools for the agent's environment, they represent the execution of dynamic content to perform administrative tasks.

deliverability-incident-response