v0.9.1

Bug Fix

Fix misleading model fallback warning (#18)

The problem: When a configured model wasn't found, the warning said:

[micode] Model "glm/xxx" is not available. Using opencode default.

This was misleading because:

1. The "opencode default" might have ALSO failed validation
2. The actual model being used could be the plugin default (openai/gpt-5.2-codex), not the opencode default
3. Users couldn't tell which model was actually being used

The fix: Warning now shows the actual fallback model:

[micode] Model "glm/xxx" is not available. Using openai/gpt-5.2-codex.

This accurately reflects which model will be used, helping users understand and debug their configuration.

v0.9.0

New Features

• Configurable compaction threshold: Add compactionThreshold to ~/.config/opencode/micode.json to control when auto-compaction triggers (0-1 ratio, default 0.5)
• Model context limits from opencode.json: Automatically reads limit.context from your provider configs instead of relying on hardcoded defaults

Example Configuration

{
  "compactionThreshold": 0.3
}

This is useful for providers with lower context caps (e.g., github-copilot with ~128k tokens).

Closes #9

v0.8.6

Micro-task architecture for parallel execution

Changes

• Planner: generates ONE file per task with dependency batching (5-15 tasks per batch)
• Executor: batch-first parallelism with 10-20 simultaneous implementers
• Implementer: handles single-file micro-tasks, no commits (executor batches commits)
• Reviewer: quick single-file reviews

Fixes

• Prevents planner from running git/ls commands that were causing 10-15 min startup delays

v0.8.5

Features

• feat(reviewer): add actionable fix suggestions to reviews
• feat(commander): add quick-mode for trivial tasks
• feat(implementer): add adaptive behavior for minor mismatches
• feat(planner): add confident gap-filling behavior
• feat(probe): add proactive question philosophy
• feat(agents): expand commander and octto prompts
• octto integration
• feat(octto): expand question schema support
• feat(task): update logger test to use ESM import
• feat(indexing): add milestone artifact indexing
• feat(octto): add browser-based brainstorming system
• feat(indexing): add milestone artifact search tool
• feat(utils): add centralized config utility
• feat(utils): add logger.ts utility for standardized logging
• feat(utils): add errors.ts utility for unified error handling

Fixes

• fix(config): relax model validation
• fix(config): clean up model validation tests
• fix(octto): address review feedback
• fix(brainstormer): make agent decisive instead of question-seeking
• fix(config): validate model overrides against opencode.json
• fix(agents): reduce confirmation prompts
• fix: validate model configuration and fallback to defaults
• fix(pty): add error handling for spawn failures
• fix(context-injector): use camelCase filePath property name
• fix: update integration test to expect new default model

Other

• docs: add agent proactiveness plan
• test(config): clean lint warnings
• test(config): cover model validation
• docs(design): clarify milestone artifact classification
• docs: add milestone artifact indexing design
• refactor: use centralized config, logger, and error utilities
• 0.8.5

v0.8.3

Bug Fixes

• Fix intermittent compaction timeout by fixing race condition
• Set up compaction listener before calling summarize API
• Wait for summary message (message.updated with summary: true) instead of non-existent session.compacted event

v0.8.2

Fixes

• Fix auto-compaction only reducing context by ~10% instead of 90%+ (fixes #8)
• Now waits for session.compacted event before completing compaction

v0.7.6

Changes

• Remove background task feature (background_task, background_output, background_cancel, background_list tools)
• Subagents now run synchronously via the Task tool

v0.7.5

Changes

Other

• refactor(agents): replace background_task with synchronous Task tool

v0.7.4

Changes

Other

• refactor(planner): use synchronous Task tool instead of background_task

v0.7.3

What's Changed

Features

• background-task: Add ctx.metadata() for session navigation UI
• background-task: Add session.idle event handling for completion detection

Refactoring

• prompts: Update all agent prompts to always use background_task (never Task)

Chores

• Remove debug console.log statements

Full Changelog: v0.7.2...v0.7.3