The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from various external sources including forums (Reddit), social media posts, product reviews, and public GitHub issues (references/demand-signals.md). This presents a vulnerability surface for indirect prompt injection where an attacker could embed malicious instructions in these public posts to influence the agent's behavior during the qualification or report generation phases.
Ingestion points: Discovery phases in all branches ingest data from external platforms; Step 3 in Demand-signal discovery specifically targets public forums and posts.
Boundary markers: The skill instructs the agent to 'Quote minimally, paraphrase by default' and use 'source-based openers', which may reduce accidental command adherence but does not fully eliminate the risk.
Capability inventory: The skill identifies leads and prepares data for handoff to other skills like suede-cold-email, which could be exploited to send malicious outreach.
Sanitization: Step 6 in demand-signals.md includes a command to 'Escape every value and keep the output self-contained' when generating reports, which is a positive security practice.
[EXTERNAL_DOWNLOADS]: The skill recommends and utilizes several well-known external SaaS platforms and APIs for data enrichment and verification, including Apollo, Clay, ZoomInfo, Clearbit, BuiltWith, and Truelist (references/data-sources.md). Per security policy, these are recognized as well-known technology services and are documented neutrally without escalating the verdict.

suede-prospecting