The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The agent processes untrusted data from activity logs (emails, notes, and meeting summaries) to generate follow-up drafts and Slack digests.
Ingestion points: The agent reads the subject and body fields from the gtm_activities model as defined in infra/models/gtm-activities.ts and infra/agents/nudger.prompt.ts.
Boundary markers: The prompt interpolation does not include explicit delimiters or instructions to ignore potential commands embedded within the activity data.
Capability inventory: The agent's capabilities are restricted. It has read-only access to most models and is only permitted to write to the deal_nudges ledger and post to Slack using a locked channelId defined in infra/agents/nudger.ts. It cannot send emails or modify CRM data.
Sanitization: No specific sanitization or filtering of the activity log content is implemented before it is processed by the language model.

stalled-deal-nudge