sales-methodology-implementer-skill

Pass

Audited by Gen Agent Trust Hub on Mar 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill mandates a file system write operation. It directs the agent in SKILL.md to save a JSON file containing session metrics to ~/.claude/skill-analytics/last-outcome-sales-methodology-implementer.json within the 'Emit Outcome Sidecar' section.
  • [PROMPT_INJECTION]: The skill architecture presents an indirect prompt injection surface.
  • Ingestion points: Untrusted user input enters the agent context via sales methodology parameters and deal-specific data in SKILL.md.
  • Boundary markers: The skill does not define delimiters or use instructions to ignore embedded commands within the ingested data.
  • Capability inventory: The agent has the capability to write files to the local filesystem as specified in SKILL.md.
  • Sanitization: No evidence of input validation, escaping, or sanitization is provided for the data processed or written to the analytics file.
  • [PROMPT_INJECTION]: Deceptive metadata is present. The config.json file attributes the author as 'Tim Kipper', which is inconsistent with the established vendor identity of 'ScientiaCapital'.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 22, 2026, 06:51 PM
Security Audit — agent-trust-hub — sales-methodology-implementer-skill