Skip to content

Decision 011: Make System Model + Feedback Discipline Explicit

Date: 2026-02-06 Status: Accepted

Context

The skill library already guides architecture, resilience, observability, and verification well, but several systemic-thinking behaviors are still implicit:

  • Problem models are often held in the operator's head instead of externalized.
  • Trade-offs are discussed but not always captured as comparable options with reversal triggers.
  • Metrics are specified, but metric-to-decision linkage and review rituals are not consistently required.
  • Second-order effects, blast radius, and system dynamics (delay/accumulation/drift) are not explicit enough.
  • Outputs are optimized for engineers but not consistently translated for decision stakeholders.

This creates risk of locally strong decisions that are hard to align, review, and adapt over time.

Decision

Update core workflow/spec/architecture/observability/plan/finish guidance to require lightweight explicit artifacts for non-trivial work:

  1. System sketch: boundary, time horizon, actors/incentives, key flows, and top constraints.
  2. Decision table: options, objectives, known downsides, and kill/reversal criteria.
  3. Measurement ladder: decision, leading indicators, lagging outcomes, instrumentation source, review ritual/owner.
  4. Failure propagation map: what breaks next, what breaks silently, and organizational cascade points.
  5. Dynamics check: what accumulates, expected delays, and balancing loops.
  6. Stakeholder translation: concise executive packet plus engineering packet for non-trivial delivery reports.

Keep these artifacts minimal and mostly template-driven to avoid skill bloat.

Consequences

  • Positive:
  • More testable and portable system models across teams and agents.
  • Clearer trade-off memory and reversibility discipline.
  • Faster learning loops by connecting telemetry to named decisions and rituals.
  • Better cross-functional alignment with dual-bandwidth reporting.
  • Trade-offs:
  • Slightly higher upfront structure for non-trivial work.
  • More template maintenance across skills and prompt recipes.
  • Compatibility/migration impact:
  • Additive change only; no skill renames or taxonomy changes.
  • Existing prompts remain valid and can adopt new artifacts incrementally.