Skip to content

security: LLM meta-analysis can suppress high-severity static findings via prompt injection #59

@AbhiramDwivedi

Description

@AbhiramDwivedi

Problem

meta_analyzer.apply_filter() keeps a static finding only if the LLM confirms it (is_vulnerability=True, confidence >= 0.6). Since the LLM's input includes attacker-controlled skill content, a prompt-injection payload can make the LLM drop even a CRITICAL/HIGH static finding — a false negative in a security gate. Affects all providers.

Proposed fix

Never let LLM filtering silently drop CRITICAL/HIGH static findings (preserve + tag llm-unconfirmed, surfaced in JSON); keep LLM false-positive filtering for MEDIUM/LOW.

Opened to follow CONTRIBUTING.md (issue-first); fix is in #54.

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions