Ygg v0.6.5

ygg v0.6.5

Released 2026-08-30.

v0.6.5 is a tool-loop performance, context-accounting, provider-correctness,
and model-aware terminal-controls release. It uses the provider's advertised
Codex context more fully, reduces avoidable tool-turn latency, makes /context
reflect the agent's semantic accounting, and keeps provider transport and model
metadata behavior truthful. The command and binaries remain ygg and
ygg-host; ygg --version reports ygg 0.6.5.

Highlights

Faster, safer tool turns

• Responses Lite keeps its explicit serial wire contract while the host overlaps
  only explicitly parallel-safe pure or workspace-read calls. Shell and mutation
  effects remain ordered.
• Codex routes can use provider-advertised tool-call batching without weakening
  host-side effect ordering, reducing avoidable tool-loop latency.
• The default bounded model-visible tool-result allowance is now 50 KiB instead
  of 16 KiB. Output remains policy-controlled and bounded.

More useful context accounting

• Authenticated Codex routes use the provider's full advertised context window
  by default for in-context learning. Set compaction.max_active_tokens (for
  example 272000) to constrain the active working set.
• /context now reads the agent's provider-reconciled semantic breakdown instead
  of reconstructing transport-shaped estimates, and its capacity grid adapts to
  large model windows.
• The default compaction threshold and advertised output ceiling continue to
  respect the actual remaining context capacity.

Immediate answers and model-aware controls

• /answer [instruction] requests a final response from evidence already
  gathered. Idle runs start without tools; active runs persist the directive at
  the next safe boundary and expose no tools afterward.
• Slash completion, model selection, session resume, and picker focus use the
  active model's adaptive accent consistently without recolouring stored prompt
  provenance.

Provider data and transport correctness

• GPT-5.6 Luna and Terra pricing now matches OpenAI's published standard costs,
  including long-context tiers. GPT-5.4 Pro and GPT-5.5 Pro Codex pricing is
  preserved, the rejected base gpt-5.6 alias is excluded, and checked-in model
  snapshots are refreshed together.
• A preferred Responses WebSocket is retired after a pre-generation
  connection-lifetime failure so the request can retry through HTTP. A
  post-generation disconnect remains terminal.

Updating

Installer-channel users can update in place:

ygg update --check
ygg update

Or run the version-pinned installer directly:

curl --proto '=https' --tlsv1.2 -LsSf \
  https://github.com/skaft-software/ygg/releases/download/v0.6.5/install-ygg.sh | sh

Then restart Ygg and verify:

ygg --version
ygg --help
ygg doctor

Cargo-channel users can install the immutable tag with Rust 1.86 or newer:

cargo install --locked \
  --git https://github.com/skaft-software/ygg \
  --tag v0.6.5 \
  --bins \
  ygg-coding-agent

Packages and compatibility

• Core crates and first-party packages target Ygg 0.6.5 exactly.
• Executable-extension API 0.2 remains the wire contract; matching
  ygg-browse, ygg-mcp, ygg-subagents, ygg-web-search, and Ygg Serve
  archives are published with the release.
• Existing local configuration, credentials, sessions, goals, extensions, and
  Serve data remain in place across installer upgrades.
• Existing explicit compaction thresholds remain readable; the built-in default
  continues to use the full advertised window unless an active-token cap is
  configured.

Known limits

• Prebuilt targets remain GNU/Linux x86-64, macOS x86-64, and macOS Apple
  silicon; Linux musl is not included.
• Live provider/audio acceptance is optional and is recorded as a waiver when
  not explicitly selected for the release workflow.
• Theme selection remains disabled; the compiled model-aware default is the only
  runtime theme.
• npm, Homebrew, PyPI, GHCR, universal MCP compatibility, and universal Pi
  compatibility are not part of this release.