Token Efficiency Sprint

This release is all about making every token count — hyperfocused mode, context compression, and method-level surgery work together to reduce token consumption by 10–75x.

Hyperfocused Mode — 1 Tool, 99.5% Less Schema

--mode hyperfocused registers a single SAP(action, target, params) tool instead of 122 individual tools.

All safety controls (read-only, allowed-ops, package restrictions) work identically across all modes.

Context Compression — 7-30x Per Dependency

GetSource auto-appends compressed dependency contracts (public API signatures only). One MCP call returns source + surrounding context. 10 regex patterns extract dependencies, contracts fetched in parallel, classes compressed to PUBLIC SECTION only.

Method-Level Surgery — 95% Source Reduction

Read/write individual methods without pulling entire classes:

SAP(action="read",  target="CLAS ZCL_FOO", params={"method": "CALCULATE"})
SAP(action="edit",  target="CLAS ZCL_FOO", params={"method": "CALCULATE", "source": "..."})

vsp handles fetch, splice, syntax check, push internally. Context compression scopes to the method's own dependencies.

ABAP LSP

vsp lsp --stdio provides real-time syntax checking and go-to-definition — zero token cost (runs via editor LSP channel, not MCP).

Mode Unification

SAP_TOOL_MODE removed. Now one axis: SAP_MODE=focused|expert|hyperfocused.

Combined Token Budget

Platforms

Linux x64, ARM64, x86, ARM | macOS Intel, Apple Silicon | Windows x64, ARM64, x86

Credits

• Filipp Gnilyak — hyperfocused mode concept
• Lars Hvam — abaplint ABAP parser