snowflake-v2-progressive-disclosure-for-skill-files
Monolithic skill files (1000+ lines) load their full token cost on every invocation even when only one phase executes. Splitting into a 50-line entry point that references phase-specific files (loaded on demand) achieves 95%+ reduction in baseline token cost. The session-close skill went from 1230 lines to a 54-line entry point with 8 phase reference files — same capability, fraction of the cost per session.
Related
- 2026-04-04-oracle-001-self-architecture-analysis
- enterprise-capability-expansion-5-pillars-from-digital-employee-analysis
- docker
- clawteam-openclaw-multi-agent-swarm-evaluation
- oracle
- protocol-file-splitting-threshold-400-lines
- snowflake-v2-pattern-for-oversized-skill-files
- snowflake-v2-progressive-disclosure-for-large-skill-files
- snowflake-v2-skill-pattern-reduces-entry-point-token-cost-95
- snowflake-v2-progressive-disclosure-pattern-for-large-skill
- progressive-disclosure-skill-architecture-pattern