pragman-unfck
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill ingests untrusted data from session transcripts and tool outputs, creating a surface for indirect prompt injection.
- Ingestion points: Data is pulled from Codex, Claude Code, and Cursor adapters or manual user excerpts as defined in SKILL.md and references/privacy-and-ingestion.md.
- Boundary markers: Explicit instructions in SKILL.md mandate that transcript content be treated as inert evidence and that any instructions found within transcripts must be ignored.
- Capability inventory: The skill uses the pragman CLI to analyze sessions and modify project or personal configuration files (references/change-contract.md).
- Sanitization: The process includes mandatory redaction, use of aliases for paths, and a two-stage user approval process for all proposed configuration changes.
Audit Metadata