Give me a description of how the session management in the opencode project @packages/opencode/ syncs the git repositories with session events in order to roll back changes if nessecary.
How does this handle multiple session in the same repoository making changes?
Are they using worktrees? Are they rolling their own repository management with git libraries or similar?
Called the list tool with the following input: {"path":"/home/jm/data/code/opencode/packages/opencode/"}
/home/jm/data/code/opencode/packages/opencode/
migration/
20260127222353_familiar_lady_ursula/
migration.sql
snapshot.json
20260211171708_add_project_commands/
migration.sql
snapshot.json
20260212005636_add_session_agent_model/
migration.sql
snapshot.json
src/
bun/
index.ts
registry.ts
bus/
bus-event.ts
global.ts
index.ts
test/
acp/
agent-interface.test.ts
event-subscription.test.ts
agent/
agent.test.ts
cli/
tui/
transcript.test.ts
github-action.test.ts
github-remote.test.ts
import.test.ts
config/
fixtures/
empty-frontmatter.md
frontmatter.md
markdown-header.md
no-frontmatter.md
weird-model-id.md
agent-color.test.ts
config.test.ts
markdown.test.ts
file/
ignore.test.ts
path-traversal.test.ts
ripgrep.test.ts
fixture/
lsp/
fake-lsp-server.js
fixture.ts
ide/
ide.test.ts
lsp/
client.test.ts
mcp/
headers.test.ts
oauth-browser.test.ts
memory/
abort-leak.test.ts
patch/
patch.test.ts
permission/
arity.test.ts
next.test.ts
plugin/
auth-override.test.ts
codex.test.ts
project/
project.test.ts
provider/
copilot/
convert-to-copilot-messages.test.ts
copilot-chat-model.test.ts
amazon-bedrock.test.ts
gitlab-duo.test.ts
provider.test.ts
transform.test.ts
question/
question.test.ts
server/
session-list.test.ts
session-select.test.ts
session/
compaction.test.ts
instruction.test.ts
llm.test.ts
message-v2.test.ts
prompt-missing-file.test.ts
prompt-special-chars.test.ts
prompt-variant.test.ts
retry.test.ts
revert-compact.test.ts
session-agent-model.test.ts
session.test.ts
skill/
discovery.test.ts
skill.test.ts
snapshot/
snapshot.test.ts
storage/
json-migration.test.ts
tool/
snapshots/
tool.test.ts.snap
fixtures/
large-image.png
models-api.json
apply_patch.test.ts
bash.test.ts
external-directory.test.ts
grep.test.ts
question.test.ts
read.test.ts
registry.test.ts
skill.test.ts
truncation.test.ts
util/
filesystem.test.ts
format.test.ts
iife.test.ts
lazy.test.ts
lock.test.ts
timeout.test.ts
wildcard.test.ts
bun.test.ts
keybind.test.ts
permission-task.test.ts
preload.ts
scheduler.test.ts
.gitignore
AGENTS.md
Dockerfile
README.md
bunfig.toml
drizzle.config.ts
package.json
parsers-config.ts
sst-env.d.ts
tsconfig.json
Give me a description of how the session management in the opencode project @packages/opencode/ syncs the git repositories with session events in order to roll back changes if nessecary.
How does this handle multiple session in the same repoository making changes?
Are they using worktrees? Are they rolling their own repository management with git libraries or similar?
Called the list tool with the following input: {"path":"/home/jm/data/code/opencode/packages/opencode/"}
/home/jm/data/code/opencode/packages/opencode/
migration/
20260127222353_familiar_lady_ursula/
migration.sql
snapshot.json
20260211171708_add_project_commands/
migration.sql
snapshot.json
20260212005636_add_session_agent_model/
migration.sql
snapshot.json
src/
skill/
index.ts
skill.ts
worktree/
index.ts
index.ts
sql.d.ts
test/
acp/
agent-interface.test.ts
event-subscription.test.ts
agent/
agent.test.ts
cli/
tui/
transcript.test.ts
github-action.test.ts
github-remote.test.ts
import.test.ts
config/
fixtures/
empty-frontmatter.md
frontmatter.md
markdown-header.md
no-frontmatter.md
weird-model-id.md
agent-color.test.ts
config.test.ts
markdown.test.ts
file/
ignore.test.ts
path-traversal.test.ts
ripgrep.test.ts
fixture/
lsp/
fake-lsp-server.js
fixture.ts
ide/
ide.test.ts
lsp/
client.test.ts
mcp/
headers.test.ts
oauth-browser.test.ts
memory/
abort-leak.test.ts
patch/
patch.test.ts
permission/
arity.test.ts
next.test.ts
plugin/
auth-override.test.ts
codex.test.ts
project/
project.test.ts
provider/
copilot/
convert-to-copilot-messages.test.ts
copilot-chat-model.test.ts
amazon-bedrock.test.ts
gitlab-duo.test.ts
provider.test.ts
transform.test.ts
question/
question.test.ts
server/
session-list.test.ts
session-select.test.ts
session/
compaction.test.ts
instruction.test.ts
llm.test.ts
message-v2.test.ts
prompt-missing-file.test.ts
prompt-special-chars.test.ts
prompt-variant.test.ts
retry.test.ts
revert-compact.test.ts
session-agent-model.test.ts
session.test.ts
skill/
discovery.test.ts
skill.test.ts
snapshot/
snapshot.test.ts
storage/
json-migration.test.ts
tool/
snapshots/
tool.test.ts.snap
fixtures/
large-image.png
models-api.json
apply_patch.test.ts
bash.test.ts
external-directory.test.ts
grep.test.ts
question.test.ts
read.test.ts
registry.test.ts
skill.test.ts
truncation.test.ts
util/
filesystem.test.ts
format.test.ts
iife.test.ts
lazy.test.ts
lock.test.ts
timeout.test.ts
wildcard.test.ts
bun.test.ts
keybind.test.ts
permission-task.test.ts
preload.ts
scheduler.test.ts
.gitignore
AGENTS.md
Dockerfile
README.md
bunfig.toml
drizzle.config.ts
package.json
parsers-config.ts
sst-env.d.ts
tsconfig.json
I'll delegate this analysis to the analyzer agent to investigate the session management and git repository synchronization in the opencode project.
task_id: ses_3ab0f8f2cffeCyb3Un2qYTyOio (for resuming to continue this task if needed)
<task_result>
Based on my comprehensive analysis of the OpenCode project, here's a detailed report on how session management syncs with git repositories for rollback purposes:
---
# OpenCode Session Management & Git Sync Architecture
## Architecture Overview
OpenCode uses a sophisticated multi-layer architecture for session management and git synchronization:
### Core Components
1. **Session System** (`src/session/`) - Manages conversation sessions, messages, and parts
2. **Snapshot System** (`src/snapshot/`) - Git-based state tracking for rollback
3. **Worktree Module** (`src/worktree/`) - Git worktree management for isolation
4. **Project Instance** (`src/project/instance.ts`) - Context provider for session operations
5. **Event Bus** (`src/bus/`) - Event-driven communication between components
---
## 1. Session Management & Git Sync
### Snapshot-Based Rollback System
The snapshot system is the heart of the rollback mechanism. It creates a **separate git repository** for tracking file states:
```typescript
// src/snapshot/index.ts - gitdir() function
function gitdir() {
const project = Instance.project
return path.join(Global.Path.data, "snapshot", project.id)
}
```
**Location:** `~/.opencode/data/snapshot/<project-id>/`
### Key Snapshot Operations
| Operation | Git Command Used | Purpose |
|-----------|------------------|---------|
| `track()` | `git write-tree` | Creates a git tree object representing current state |
| `patch(hash)` | `git diff --name-only` | Gets files modified since snapshot |
| `revert(patches)` | `git checkout <hash> -- <file>` | Restores individual files to snapshot state |
| `restore(snapshot)` | `git read-tree && git checkout-index` | Full restoration to snapshot state |
| `diffFull(from, to)` | `git diff` + `git show` | Computes detailed file diffs between snapshots |
### Session Event Tracking
Sessions track changes through **messa...Based on the analyzer's comprehensive investigation, here's how the system works:
OpenCode uses a clever dual-repository approach:
.git directory for version control~/.opencode/data/snapshot/<project-id>/ for tracking file statesThe snapshot system uses git tree objects (not commits) to track state:
git write-tree - Creates a tree hash representing current file stategit checkout <hash> -- <file> - Restores individual filesgit read-tree + git checkout-index - Full restorationSession messages contain step-start and step-finish parts with snapshot hashes, plus patch parts listing modified files. Rollback collects patches after the revert point and restores files using git checkout.
Sessions do NOT use worktrees for isolation. Instead, they share the same working directory with:
src/file/time.ts) - Tracks when each session last read each fileThis is optimistic concurrency control - they rely on file modification time checks and write serialization rather than git-level locking.
The src/worktree/ module creates separate git worktrees for isolated development environments (like testing branches), stored at ~/.opencode/data/worktree/<project-id>/<name>/. These are not used for session rollback - they're for creating parallel development sandboxes.
They use native git commands through Bun's shell syntax ($), not a JavaScript library:
// Example: Creating a snapshot
const hash = await $`git --git-dir ${git} --work-tree ${Instance.worktree} write-tree`
// Example: Restoring a file
await $`git --git-dir ${git} --work-tree ${Instance.worktree} checkout ${hash} -- ${file}`
| Aspect | Approach |
|---|---|
| Rollback tracking | Separate git repo with tree objects |
| Session isolation | File-level locks + read time tracking |
| Worktrees | For dev sandboxes, not sessions |
| Git operations | Native commands via Bun shell |
| Event coordination | Bus system for cross-component sync |
The design is elegant: using git's tree objects avoids needing branches/commits in the snapshot repo, and native git commands are simpler and more reliable than JavaScript git libraries.