v1.4.4

Changes

RLM Skills

• terraphim-rlm: add LLM Configuration section (Ollama > OpenRouter > proxy auto-detection, set_llm_client() injection)
• deterministic-rlm-review: add LLM provider Prerequisites section
• kg-rlm-ingest: KG-first, RLM-last pipeline. terraphim-agent extract for deterministic matching; rlm_query optional fallback with budget gate
• rlm-skills-launch: 2026-05-18 LLM bridge wiring update (Refs terraphim-ai #1744)

Why KG-first

terraphim-agent extract uses Aho-Corasick automata for zero-cost concept matching. Only novel concepts need RLM. Works without any LLM provider.

v1.3.0: Judge 5-Level Verdict, Escape Hatches, ZDP SRD, Rust Skills

What's New in v1.3.0

Judge 5-Level Verdict Ladder (#8)

Extended the judge verdict system from binary GO/NO-GO to a 5-level ladder:

• BLOCK -- serious correctness/data/security/deploy risk
• FIX FIRST -- likely safe after targeted fixes
• SAFE TO COMMIT -- ready, minor caveats
• SAFE TO DEPLOY AFTER RUNTIME CHECK -- committable but deploy needs runtime verification
• INSUFFICIENT EVIDENCE -- reviewer lacked evidence to make a reliable call

Escape Hatches for Disciplined Skills (#7)

Added "Skip This Skill When..." sections to all 7 disciplined-* skills to prevent bloat on trivial changes.

ZDP v2.3 SRD Support (PR #52)

Software Release Definition integration across the V-model with maturity tracking, IOC validation, and testability checks.

New Skills (38 total)

• learning-capture -- PostToolUse hook for capturing failed commands
• rust-ci-cd -- Rust CI/CD pipeline patterns
• rust-observability -- Observability and tracing
• rust-performance -- Performance optimization
• rust-development -- Idiomatic Rust development

Installation

curl -fsSL https://raw.githubusercontent.com/terraphim/terraphim-skills/main/scripts/skills.sh | bash

Full Changelog: v1.2.0...v1.3.0