win-loss-decision-signal
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes qualitative records (sales notes, prospect interviews) to generate analysis, which constitutes a potential surface for indirect prompt injection.
- Ingestion points: External deal records including sales notes and interview transcripts accessed during the Create or Update modes.
- Boundary markers: Absent; there are no explicit instructions to the agent to treat input data as untrusted or to use delimiters.
- Capability inventory: The skill is permitted to read and write files within local project directories such as
win-loss-signals/andgtm/win-loss/. - Sanitization: No explicit sanitization, validation, or filtering of the qualitative input is defined in the instructions.
- [DATA_EXFILTRATION]: The skill operates entirely on the local file system. It reads and writes to specific organizational paths (e.g.,
win-loss-signals/,competitive-intelligence/) but does not perform any network operations or external data transmissions.
Audit Metadata