Lentilles Cognitives Multi-Cadres

---

name: cognitive-lensing description: Apply multilingual cognitive frames to re-approach complex tasks with targeted reasoning patterns and bias checks. allowed-tools: Read, Write, Edit, Bash, Glob, Grep, Task, TodoWrite model: sonnet x-version: 3.2.0 x-category: foundry x-vcl-compliance: v3.1.1 x-cognitive-frames:

• HON
• MOR
• COM
• CLS
• EVD
• ASP
• SPC

---

L1 Improvement

• Recast the lensing guide into the Skill Forge section flow with explicit guardrails, patterns, and completion checks.
• Added prompt-architect style constraint extraction and confidence ceilings to prevent overclaiming from speculative frames.

STANDARD OPERATING PROCEDURE

Purpose

Switch reasoning frames (linguistic, disciplinary, or persona-based) to unlock alternative solution paths and mitigate bias in complex tasks.

Trigger Conditions

• Positive: stalled reasoning, need for alternative perspectives, bias detection, or creative exploration.
• Negative/reroute: straightforward prompt rewrites (prompt-architect) or agent creation (agent-creator/agent-creation).

Guardrails

• Keep outputs in English; cite which lens was applied and why.
• Do not fabricate expertise—ground lens choice in task constraints and evidence.
• Limit to 2-3 focused lenses per pass to avoid fragmentation.
• State confidence ceilings explicitly, especially when using speculative or research lenses.

Execution Phases

1. Assessment: Capture task intent, constraints, and observed failure modes; classify hard/soft/inferred constraints.
2. Lens Selection: Choose lenses (e.g., formal proof, UX research, security red-team, socio-technical) mapped to the task.
3. Application: Re-articulate the problem through each lens with targeted heuristics and checks.
4. Synthesis: Compare insights, resolve conflicts, and propose next actions with confidence ceilings.

Pattern Recognition

• Analytical stagnation → apply formal/algorithmic lens.
• User impact unclear → apply UX research or accessibility lens.
• Risky changes → apply safety/security lens to uncover failure paths.

Advanced Techniques

• Use paired lenses (builder vs breaker) to surface hidden assumptions.
• Run time-boxed divergent thinking followed by convergence synthesis.
• Feed lens outputs into prompt-architect for clarity before execution.

Common Anti-Patterns

• Cycling too many lenses without decision.
• Treating lens opinions as facts; neglecting evidence and ceilings.
• Ignoring domain constraints when adopting a lens.

Practical Guidelines

• Name the lens, heuristic, and expected impact in each pass.
• Keep synthesized recommendations actionable and prioritized.
• Capture what changed between lenses for traceability.

Cross-Skill Coordination

• Upstream: prompt-architect to clarify the task.
• Parallel: recursive-improvement to iterate on stuck areas; meta-tools to compose lens outputs into tools.
• Downstream: agent-creation/agent-selector to operationalize chosen approach.

MCP Requirements

• Optional memory MCP to recall past lens effectiveness; tag WHO=cognitive-lensing-{session}, WHY=skill-execution.

Input/Output Contracts

inputs:
  task: string  # required problem statement
  constraints: list[string]  # optional constraints
  target_lenses: list[string]  # optional lens hints (e.g., security, accessibility)
outputs:
  lens_analyses: list[object]  # lens name, insight, risks
  synthesis: summary  # consolidated recommendation
  next_steps: list[string]  # actions derived from lensing

Recursive Improvement

• Run recursive-improvement when lenses disagree or output is indecisive; focus on evidence gaps and decision criteria.

Examples

• Apply security red-team + reliability lens to a new API rollout before deployment.
• Use accessibility + onboarding UX lens to rewrite a complex setup guide.

Troubleshooting

• Lens produces no new insight → select orthogonal lens or consult domain specialists.
• Conflicting recommendations → prioritize by risk, effort, and evidence strength.
• Overconfident claims → restate with ceilings and cite observed data.

Completion Verification

• [ ] Lenses named with rationale and heuristics applied.
• [ ] Synthesis provided with prioritized actions and ceilings.
• [ ] Traceability of changes between lenses captured.
• [ ] Constraints respected; English-only output.

Confidence: 0.70 (ceiling: inference 0.70) - Lensing SOP rewritten with Skill Forge cadence and prompt-architect ceilings.