Of the 3 deferred risks from the dual-stack browser install, only the…
Decision
Of the 3 deferred risks from the dual-stack browser install, only the SKILL.md description-drift check earns “implement next session” priority (it’s the sole Layer-4 soft-routing risk). Daemon-orphan cleanup and Python 3.13 pin re-trigger are formally accepted as deferrable (bounded blast radius, error-playbook covers re-trigger).
Rationale
First-principles decomp showed conflict-freeness rests on 4 layers — Layers 1-3 (binary, browser, state) are OS-machine-enforced; only Layer 4 (autoloader keyword routing) is heuristic. Description drift is the ONLY failure that erodes Layer 4 silently — once descriptions diverge, the autoloader misroutes and dual-stack becomes pure overhead. The other 2 risks have bounded blast radius: daemon orphan = ~150MB RAM per zombie (recoverable manually); Python pin re-trigger = error-playbook catches it via symptom grep on next install attempt. Pristine-sweep cron extension to verify the explicit cross-reference strings (“route to browser-use” / “route to playwright-cli”) in both SKILL.md descriptions is cheap, deterministic, and addresses the only soft-layer failure. Confidence calibrated 0.88 because the asymmetric-severity reasoning is well-supported by the layer decomposition and existing pristine-sweep infra.
Alternatives Rejected
Outcome
Pending