Mandate 2 3 4 Code Static Eval

Mandate 2.3.4 Code Static Evaluation Skill

Mandate

• ID: 2.3.4
• Title: Mitigation of Misinformation and Hallucination

Mitigates

• LLM09 Misinformation
• LLM08 Vector and Embedding Weaknesses
• ASI08 Cascading Failures
• ASI09 Human-Agent Trust Exploitation

Inputs

• Repository root path.
• Tech stack/language map.
• Optional architecture notes if provided.

Workflow

1. Scope relevant code surfaces tied to this mandate.
2. Run mandate-specific static checks and trace source-to-sink/control paths.
3. Extract exact evidence with file/line references.
4. Classify each control as implemented, partial, or missing.
5. Produce partial-confidence findings and list required runtime/process evidence for closure.

Mandate-Specific Static Method

1. Detect RAG or grounding pipeline usage in action-oriented workflows.
2. Verify source attribution capture and confidence metadata.
3. Verify pre-action cross-reference checks against trusted sources for critical decisions.
4. Detect execution paths that allow action without grounding/verification.
5. Verify fallback behavior when sources conflict or confidence is low.
6. Emit findings for ungrounded action paths and weak verification logic.

Output Contract

Return a single structured result with:

• mandate_id, mandate_title
• static_status: implemented|partial|missing
• confidence: low|medium|high (static-only confidence)
• evidence: list of file/line findings
• gaps: unresolved items requiring operational proof
• next_evidence_requests: runtime/process artifacts needed

Guardrails

• Avoid final compliance pass/fail using code alone.
• Mark assumptions explicitly.
• Prefer deterministic checks over heuristics when possible.