v1.8 Hardening

Release v1.8

Capability-Aware Planning and Token Efficiency: composable pipeline architecture replacing monolithic skill loading, per-agent token budgets, capability manifests, cache-aware ordering, and research compression. Six capabilities that transformed skill loading from a simple list into a sophisticated 6-stage pipeline.

Structured Data

Release Metadata

Version	`v1.8`
Release Era	Hardening
New Capabilities	6
Cumulative Capabilities	21
Cumulative Tests	—

Era Context

Hardening Era

Pipeline architecture, security hardening, spec alignment, and ecosystem robustness.

Progression Metrics

Capabilities	21 of 139
Era Progress	v1.8 of v1.8–v1.10
Growth	+6 caps — largest since v1.7

Capabilities Introduced

Architectural Decisions

Composable pipeline architecture replaces monolithic skill loading — 6-stage Score/Resolve/ModelFilter/CacheOrder/Budget/Load pipeline; each stage is independently testable and replaceable
Per-agent token budget profiles with critical/standard/optional tiers — graceful degradation under pressure; optional skills shed first, critical skills never dropped
Cache-aware skill ordering — cacheTier metadata maximizes prompt cache efficiency; frequently-used stable skills ordered first to exploit cache locality
10-20x research compression with staleness detection — reduces context load from research documents; extraction preserves key findings without full-text embedding

Breaking Changes

None.