sop-dogfooding-continuous-improvement

name: sop-dogfooding-continuous-improvement description: SOP for running continuous improvement cycles via dogfooding and adversarial validation. allowed-tools:

• Read
• Write
• Edit
• Bash
• Glob
• Grep
• Task
• TodoWrite model: sonnet x-version: 3.2.0 x-category: quality x-vcl-compliance: v3.1.1 x-cognitive-frames:
• HON
• MOR
• COM
• CLS
• EVD
• ASP
• SPC

STANDARD OPERATING PROCEDURE

Purpose

Provide a repeatable loop for applying our skills to themselves, measuring improvement deltas, and documenting learnings for continuous quality gains.

Trigger Conditions

• Positive: periodic quality reviews, regression checks after major updates, or requests to improve a specific skill.
• Negative: single execution runs without improvement goals; ad-hoc debugging tasks.

Guardrails

• Confidence ceiling: Add Confidence: X.XX (ceiling: TYPE Y.YY) with ceilings {inference/report 0.70, research 0.85, observation/definition 0.95}.
• Structure-first: Maintain examples/tests demonstrating the improvement loop and convergence criteria.
• Adversarial validation: Include boundary inputs and noisy cases before claiming convergence (<2% delta across two runs).
• Evidence logging: Tag artifacts with WHO/WHY and store metrics for trend analysis.

Execution Phases

1. Plan & Baseline
   • Select the skill and metrics; capture current performance and known gaps.
   • Prepare memory namespace and retrieve prior runs.
2. Self-Application & Iteration
   • Apply the skill to itself or representative tasks; document findings and fixes.
   • Iterate until improvements plateau.
3. Adversarial Probing
   • Inject edge cases to test robustness; log false positives/negatives.
4. Synthesis & Handoff
   • Summarize deltas, remaining risks, and next steps.
   • Update references/resources and state confidence with ceiling.

Output Format

• Baseline metrics and session goals.
• Iteration log with findings, fixes, and deltas.
• Adversarial probe outcomes and adjustments.
• Confidence statement and follow-up plan.

Validation Checklist

• Baseline captured with metrics and scope.
• At least one self-application iteration completed.
• Adversarial probes executed; deltas measured.
• References/resources updated with learnings.
• Confidence ceiling provided; English-only output.

Confidence: 0.71 (ceiling: inference 0.70) - SOP rewritten using Prompt Architect confidence discipline and Skill Forge structure-first dogfooding pattern.