[rust] 0.8.0

Added

• Separate CI/CD pipelines for JS and Rust code
• rust.yml workflow for Rust-specific builds, tests, and releases
• Changelog fragment system in rust/changelog.d/ for tracking changes

Added

• Added --temperature CLI option to override the temperature for model completions (#241)

Fixed

• Fixed Rust CI/CD release pipeline that was silently skipping all releases (#247)
  • Added always() to auto-release and manual-release job conditions to prevent GitHub Actions from skipping them when the changelog-check job is skipped on push events
  • Fixed changelog version regex in create-github-release.mjs to support Rust's `

[js] 0.22.2

fix: update create-github-release.mjs to support both JS and Rust changelog formats (#247)

• Fixed changelog version regex to handle Rust's `

Related Pull Request: #248

---

[js] 0.22.1

Lower compaction safety margin from 85% to 75% to reduce context overflow errors. Add token estimation fallback when providers return 0 token counts. Cap maxOutputTokens to never exceed model context limit.

Related Pull Request: #250

---

[js] 0.22.0

feat: add --temperature CLI option for model completion override (#241)

• Added --temperature flag to JS and Rust CLI implementations
• When set, overrides per-model and per-agent temperature defaults
• When not set, existing behavior is unchanged
• Temperature flows through PromptInput schema and User message to AI SDK
• Priority chain: CLI --temperature > agent config > model defaults

Related Pull Request: #244

---

[js] 0.21.0

feat: replace deprecated qwen3.6-plus-free default with nemotron-3-super-free (#242)

• Default model changed from qwen3.6-plus-free to nemotron-3-super-free (~262K context, NVIDIA hybrid Mamba-Transformer)
• Removed qwen3.6-plus-free from compaction models cascade (free promotion ended April 2026)
• Updated provider priority lists to remove unavailable model
• Moved qwen3.6-plus-free to deprecated/discontinued sections in documentation

Related Pull Request: #243

---

[js] 0.20.2

fix: resolve Agent CLI problems preventing task completion (#239)

• argv.ts: harden getModelFromProcessArgv() with Bun.argv fallback for global installs (oven-sh/bun#22157)
• model-config.js: add diagnostic logging for model resolution, detect silent Bun/yargs argument parsing mismatch
• storage.ts: sanitize null bytes in migration paths before file operations
• prompt.ts: check for completed tool calls BEFORE zero-token check — prevents premature session termination when provider reports zero tokens but model executed tool calls successfully

Related Pull Request: #240

---

[js] 0.20.0

feat: make qwen3.6-plus-free the default model, add compaction models cascade (#232)

• Change default model from minimax-m2.5-free to qwen3.6-plus-free (~1M context)
• Add --compaction-models CLI option for cascading compaction models
• Default cascade: (big-pickle nemotron-3-super-free minimax-m2.5-free gpt-5-nano qwen3.6-plus-free same)
• Add nemotron-3-super-free and qwen3.6-plus-free to free models lists
• Update documentation and provider priority lists

Related Pull Request: #234

---

[js] 0.19.2

Fail immediately when explicit model not found in provider, retry server errors (500/502/503), improve logging for HTTP responses, storage migrations, and verbose diagnostics

Related Pull Request: #233

---

[js] 0.19.0

feat: centralize agent config with lino-arguments, always log resolved config (#227)

• Added lino-arguments for unified env var resolution (case-insensitive, .lenv support)
• Created centralized AgentConfig module as single source of truth for all configuration
• All env vars resolved via getenv() from lino-arguments (CLI args > env vars > .lenv > defaults)
• Always log resolved configuration as JSON at startup for debugging
• Moved all direct process.env reads (MCP, read tool) into centralized Flag module
• --verbose is now the most reliable flag: triple-checked via in-memory, AgentConfig, and env var
• Removed all OPENCODE_* env var support; use LINK_ASSISTANT_AGENT_* exclusively

Related Pull Request: #228

---

[js] 0.18.3

fix: use compaction model (--compaction-model) for summarization to avoid doubling rate-limit pressure on free-tier main models (#223)

---