v1.5.2

What's Changed

• Sync published main package optional platform dependencies to the release version.
• Keep package smoke fallback on the supported external provider configuration.

Verification

• npm run verify
• GitHub Actions CI passed on main for 08820bf

v1.5.1

What's Changed

• Fix package smoke tests to use the supported external provider configuration.
• Bump npm packages to 1.5.1 for a clean publish from the fixed pipeline.

Verification

• npm run verify
• GitHub Actions CI passed on main for f748be5

v1.5.0

What's Changed

• Add Distill local-model onboarding with local as the default provider.
• Add automatic local server startup and reuse for local LLM calls.
• Add MLX backend for Apple Silicon and llama.cpp backend for Linux, Windows, and Intel macOS.
• Add runtime auto-install paths and Hugging Face/model download warmup with percentage progress in onboarding.
• Add local concurrency controls, config/env overrides, README documentation, and test coverage.

Verification

• npm run verify
• 114 pass, 1 skip, 0 fail
• build and release check passed

v1.4.9

Changes

• Make inline variable guidance always active in distill prompts.
• Add delta Dict behavior: each response defines only newly introduced variables and avoids restating old definitions.
• Enforce a substitution pass after Dict or inline variables are defined.
• Sync distill skill copies and tests.

Verification

• npm run verify

v1.4.8

Changes

• Make persisted DSL memory empty by default.
• Promote only explicit thread-local inline variables after more than five uses in learn-thread transcripts.
• Evict learned DSL entries when absent from the next thread transcript.
• Keep AR-1 prefix grammar as prompt/skill rules instead of stored DSL memory.
• Update docs, skill copies, tests, and package version to 1.4.8.

Verification

• npm run verify

v1.4.7

Release v1.4.7

• Learn inline DSL variables like workspace=#w3 and version=#v1 from distill output.
• Inject active DSL/dict memory into later prompts so every run can reuse learned terms.
• Strengthen /distill skill instructions for fixed S/C/D/R/O/N/P prefixes, Dict use, and every-thread transcript learning.
• Keep unsafe variable targets out of persisted DSL memory.

v1.4.6

Release v1.4.6

• Adopt AR-0/AR-1 distill response grammar with fixed S/C/D/R/O/N/P prefixes.
• Update distill skill copies, onboarding instructions, docs, and DSL built-ins.
• Sync package manifests to 1.4.6.

v1.4.5

Summary

• Add thread-end DSL learning with distill dsl learn-thread --stdin.
• Persist reviewed transcript-learned DSL candidates through the existing lifecycle.
• Update /distill skill and README docs for thread-end learning.

Verification

• bun test
• npm run build
• npm run verify
• npm pack --dry-run ./packages/cli

v1.4.4

Release v1.4.4.

distill v1.4.3

• Add compact DSL auto-learn memory for /distill output.\n- Use short built-in DSL keys with one-letter aliases and numeric macros.\n- Inject compact active DSL memory into normal distill prompts.\n- Add DSL learn/promote commands and update onboarding/skills for Codex and Claude.\n\nVerification: bun test; npm run verify; npm pack --dry-run ./packages/cli.