The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted text from Facebook Marketplace buyer messages.
Ingestion points: Buyer messages are extracted from the Facebook DOM using browser-harness within the triage (facebook-marketplace-buyer-inbox-triage.md) and heartbeat (facebook-marketplace-inbox-heartbeat.md) components.
Boundary markers: The skill frequently references docs/facebook-marketplace-inbox-policy.md and requires explicit user approval for outbound messages. It uses a "Safety-guard both sides" approach to validate content.
Capability inventory: The skill has the ability to execute shell commands and perform browser automation (navigation, data extraction, and form filling) via the browser-harness tool.
Sanitization: All inbound buyer messages are passed through facebook-marketplace-safety-guard to detect scam patterns and off-platform requests. Outbound drafts are also scanned before being presented to the user for approval.
[COMMAND_EXECUTION]: The skill uses browser-harness and python3 to automate browser interactions. It executes scripts (Python and JavaScript) to interact with the Facebook Marketplace UI, extract information, and manage the selling process. This high-privilege access is limited to the skill's operational scope.

facebook-marketplace-seller