Records per-session tokens, cost, model, provider, tool usage, and duration from OpenCode into your local Worktale journal.

OpenCode pre-computes cost on every assistant message. This plugin listens for message.updated events and tool.execute.after events, aggregates them per session, and shells out to worktale session add after 60 seconds of session inactivity.

Cost is read directly from OpenCode — no rate table guesswork.

npm install -g worktale@latest    # requires v1.4.0+
cd your-repo
worktale init

mkdir -p .opencode/plugins
curl -o .opencode/plugins/worktale.ts \
  https://raw.githubusercontent.com/worktale/worktale-opencode-plugin/main/src/index.ts

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    "./.opencode/plugins/worktale.ts"
  ]
}

That's it. Next time you run OpenCode in this project, the plugin loads automatically. Sessions get recorded ~60s after they go idle.

mkdir -p ~/.config/opencode/plugins
curl -o ~/.config/opencode/plugins/worktale.ts \
  https://raw.githubusercontent.com/worktale/worktale-opencode-plugin/main/src/index.ts

Then add to ~/.config/opencode/opencode.jsonc:

{
  "plugin": [
    "~/.config/opencode/plugins/worktale.ts"
  ]
}

1. Each assistant message.updated event carries the message's full snapshot — the plugin replaces any previous state for that messageID (handles streaming updates correctly without double-counting).
2. Each tool.execute.after event adds the tool name to the session's tool set.
3. Both kinds of event reset a 60-second debounce timer for that session.
4. When the timer fires (no new events for 60s), the plugin sums tokens + cost across all messages in the session and runs:

worktale session add \
  --provider <providerID> \
  --tool opencode \
  --model <modelID> \
  --input-tokens <input + cache.read + cache.write> \
  --output-tokens <output + reasoning> \
  --cost <pre-computed by OpenCode> \
  --duration <session-seconds> \
  --tools-used <comma-separated>

Sessions with fewer than 100 total tokens are skipped.

worktale today                    # today's commits + AI sessions
worktale session list             # recent sessions
worktale session stats --days 30  # cost & token rollup
worktale dash                     # interactive TUI

• Multiple records per session: if a single OpenCode session has natural pauses longer than 60s between bursts of activity, each burst becomes a separate Worktale record. The cumulative cost is the same; just split across rows. Use worktale session stats to roll up.
• Active-on-shutdown loss: if you close OpenCode mid-debounce (within 60s of the last assistant message), that session's record never gets written. Adjust FLUSH_DELAY_MS in the plugin if this matters.
• Per-project state: OpenCode loads the plugin per-project, so state isn't shared across concurrent projects. Each project's sessions are tracked independently.

Set WORKTALE_PLUGIN_DRY_RUN=1 in the OpenCode environment. Instead of calling worktale, the plugin logs what it would have recorded. Useful for verifying the wiring without polluting your journal.

MIT