sequence-load-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests prospect data from an external JSON state file, which introduces a potential surface for indirect prompt injection if the source data is manipulated.
- Ingestion points: Reads lead data (names, titles, companies, emails) from
~/.claude/skill-state/prospect-refresh/last-run.jsonin Stage 1. - Boundary markers: The skill performs specific validations on
run_date, email formats, and vertical categories, but does not employ delimiters or explicit instructions to ignore embedded commands within other free-text fields liketitleorcompany. - Capability inventory: The skill possesses extensive capabilities including reading and writing local files (
~/.claude/skill-state/,~/.claude/skill-runs/,~/.claude/skill-analytics/) and executing network operations via various HubSpot and Apollo MCP tools. - Sanitization: Sanitization is limited to format and value-set validation for specific fields, rather than a comprehensive escaping strategy for all untrusted input data processed by the CRM and outreach tools.
Audit Metadata