Worktale Plugin for Claude Code

AI session tracking for your daily work journal.

Git captures the what. This plugin captures the why and the how.

When you activate /worktale in Claude Code:

1. After every commit, the agent appends a 1–2 sentence narrative note to your daily Worktale journal.
2. At session end, a SessionEnd hook parses the transcript and records aggregate metrics to Worktale: provider, model, tokens, cost, tools used, MCP servers, and duration.

All data stays local. Metrics live in your Worktale SQLite DB; cloud sync is opt-in via worktale publish.

/plugin marketplace add worktale/worktale-plugin
/plugin install worktale-plugin@worktale-plugin

Requires the Worktale CLI v1.4.0+:

npm install -g worktale@latest
cd your-repo
worktale init

/worktale

The agent confirms narration is active. After every commit, it runs:

worktale note "Fixed race condition in job queue — claim query wasn't using SELECT FOR UPDATE"

At session end, the hook automatically runs:

worktale session add \
  --provider anthropic \
  --tool claude-code \
  --model <detected-model> \
  --input-tokens <sum-from-transcript> \
  --output-tokens <sum-from-transcript> \
  --cost <computed-from-rates> \
  --tools-used Read,Edit,Bash,... \
  --mcp-servers <detected> \
  --duration <session-seconds>

worktale today      # Today's sessions + narrative
worktale session list
worktale session stats
worktale dash       # Interactive TUI

The hook reads transcript_path from the SessionEnd payload, parses the JSONL, and sums usage.input_tokens, usage.output_tokens, usage.cache_read_input_tokens, and usage.cache_creation_input_tokens across every assistant message. Cost is computed from a built-in Anthropic rate table (Opus/Sonnet/Haiku 4.x). Unknown models report $0 for cost but full token counts.

MIT