v3.0.0

v3.0.0 — Single Compress Tool Architecture

A major release featuring a complete architectural overhaul of the Dynamic Context Pruning system.

Breaking Changes

• The previous 3-tool system (distill, compress, prune) has been replaced with a single compress tool. All context management now flows through one unified interface.

New Features

• Decompress / recompress commands — Built-in commands for managing compression blocks
• Subagent support — experimental.allowSubAgents enables compression awareness in sub-agent contexts
• Custom prompts — experimental.customPrompts allows user-defined prompt overrides
• Flat schema option — compress.flatSchema for simplified tool schema presentation
• Protected user messages — protectUserMessages prevents user messages from being compressed
• Protected tools with glob patterns — Configure which tool outputs are preserved during compression using glob matching
• Visual progress bar — Real-time compression activity indicator
• Configurable turn nudges — Fine-tune when and how compression nudges appear
• Strengthened manual mode — More robust manual compression control

Improvements

• Reduced cache invalidation — Significantly decreases how often DCP causes cache invalidation, fixing Anthropic-related cache issues without needing manual mode
• Infinite conversations — Conversations can now last almost indefinitely; the previous tool pruning approach limited this by leaving user/AI messages untouched
• Simplified model behavior — The model no longer needs to choose between 3 context management tools, reducing decision complexity and improving reliability

Contributors

• @spoons-and-mirrors

Migration

This is a major version bump. Users upgrading from v2.x should review the updated README for the new single-tool configuration. The compress tool now handles all context management operations that were previously split across multiple tools.

Full Changelog: v2.1.8...v3.0.0

v2.2.9-beta0 - New config options and context compression improvements

What's Changed

• Added protectUserMessages config option to prevent user messages from being compressed.
• Added flatSchema config option for simplified tool schema injection.
• Added wildcard/glob pattern support for protectedTools configuration.
• Gated compress tool behind a one-shot manual trigger.
• Fixed: insert injected text parts before tool parts in assistant messages.
• Fixed: include cache.write in system prompt token calculation.
• Fixed: strip hallucinated tags from merged sub-agent results.
• Consolidated user and assistant turn nudges into a single prompt.
• Show raw LLM summary in showCompression notification.

Full Changelog: v2.2.8-beta0...v2.2.9-beta0

v2.2.8-beta0 - Compress workflow and prompt/runtime updates

What's Changed

• Added /dcp decompress and /dcp recompress command workflows with better compression status and notification behavior.
• Added custom prompt overrides.
• Improved compress robustness with graceful fallback handling and fixes for nested compression protected-tool injection.

Full Changelog: v2.2.6-beta0...v2.2.8-beta0

v2.2.6-beta0 - Compression nudge bounds and ID consistency

What's Changed

• Added max/min nudge limits for compress guidance.
• Tightened message-id injection for ignored user messages and first subagent prompt messages.
• Added subagent safety guidance to system prompts.
• Bumped package metadata to 2.2.6-beta0.

Config Changes

• Rename compress.contextLimit -> compress.maxContextLimit.
• Rename compress.modelLimits -> compress.modelMaxLimits.
• Add optional lower-bound settings: compress.minContextLimit and compress.modelMinLimits.

"compress": {
  "maxContextLimit": "80%",
  "minContextLimit": 30000,
  "modelMaxLimits": {
    "openai/gpt-5.3-codex": 120000,
    "anthropic/claude-sonnet-4.6": "75%"
  },
  "modelMinLimits": {
    "openai/gpt-5.3-codex": 30000,
    "anthropic/claude-sonnet-4.6": "20%"
  }
}

Defaults:

• compress.maxContextLimit: 100000
• compress.minContextLimit: 30000

Full Changelog: v2.2.5-beta0...v2.2.6-beta0

v2.2.5-beta0 - Subagent context and compression state improvements

What's Changed

• Added experimental subagent support:

"experimental": { "allowSubAgents": true }

• /undo properly undoes (?) compressions done in the undone messages
• Bumped package metadata to 2.2.5-beta0.

Full Changelog: v2.2.3-beta0...v2.2.5-beta0

v2.1.8 - Version bump

What's Changed

• fix(hooks): append system prompt to last output part
• fix: invert text part rejection logic to target claude models
• chore: add python artifacts to gitignore

Full Changelog: v2.1.7...v2.1.8

v2.2.3-beta0

Beta release 2.2.3-beta0

v2.2.2-beta0 - beta prerelease

What's Changed

• Added configurable nudgeForce behavior for turn-level compression prompting.
• Restored user/assistant turn nudge prompts and simplified direct prompt imports.
• Simplified user-facing config wording for compression settings in docs/schema.

Full Changelog: v2.2.0-beta0...v2.2.2-beta0

v2.2.0-beta0 - beta branch release

What's Changed

• Strategy shift to a single user-facing context tool: compress.
• Context management is less surgical and more focused on preserving high-signal work grouped by task.
• Significantly improved cache invalidation, should be a very minor impact now on all providers especially anthropic
• To try this prerelease, install @tarquinen/opencode-dcp@beta.

v2.1.7 - Prune origin tracking and sync improvements

What's Changed

• Track prune origins to improve undo reconciliation behavior
• Add sync message flow and rename reconcile to sync origins
• Prevent compress tool summary modification during prune
• Add write/edit to default protected tools and skip already-pruned IDs

Full Changelog: v2.1.6...v2.1.7