Worktale Plugin for OpenAI Codex

AI session tracking for your daily work journal — with real token + cost capture.

When active, Codex:

1. Per commit: the agent appends a 1–2 sentence narrative note to your daily Worktale journal via worktale note.
2. At session end: a Stop / SessionEnd hook parses the Codex session JSONL files at ~/.codex/sessions/YYYY/MM/DD/*.jsonl and automatically records provider, model, input tokens, cached tokens, output tokens, computed cost, and duration to your local Worktale DB.

Codex DOES persist token usage to disk (see CodexMonitor's prior art — same source). The plugin reads those rollouts and computes cost from a built-in OpenAI rate table.

# 1. Clone this repo
git clone https://github.com/worktale/worktale-codex-plugin.git

# 2. Copy the skill so Codex picks it up
mkdir -p ~/.codex/skills/worktale
cp -r worktale-codex-plugin/skills/worktale/* ~/.codex/skills/worktale/

# 3. Register the SessionEnd hook
mkdir -p ~/.codex
cp worktale-codex-plugin/hooks/hooks.json ~/.codex/hooks.json

# 4. Make sure the hook script is somewhere stable
mkdir -p ~/.codex/hooks
cp worktale-codex-plugin/hooks/session-track.mjs ~/.codex/hooks/
# Then update ~/.codex/hooks.json to point at the absolute path:
# "command": "node /home/you/.codex/hooks/session-track.mjs"

Requires the Worktale CLI v1.4.0+:

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

Codex fires the Stop hook after every turn. The script:

1. Scans ~/.codex/sessions/ for recent .jsonl rollouts.
2. Skips files modified in the last 5 minutes (treats them as still-active).
3. Skips files already recorded (tracked in ~/.worktale/codex-processed.json).
4. For each newly-finished session: parses the JSONL, sums total_token_usage deltas across token_count events, extracts model from turn_context events, computes cost against the built-in OpenAI rate table (GPT-5/4o/4.1/o3/o4 family + Codex-mini), and shells out to worktale session add with the aggregate.

The trade-off vs. Claude Code's hook: Codex doesn't fire a session-end event, only turn-end. So the very last session of a run gets recorded the next time Codex starts (or by manually running the script). Everything else is captured the same day.

The script ships with current published OpenAI rates (USD per 1M tokens):

Unknown models report tokens but $0 for cost. Update the table in hooks/session-track.mjs if rates change.

In your Codex session, mention the worktale skill (or activate it by argument). Codex will:

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

That's it. At session end, the hook runs the equivalent of:

worktale session add \
  --provider openai \
  --tool codex \
  --model gpt-5 \
  --input-tokens 13000 \
  --output-tokens 2300 \
  --cost 0.2955 \
  --duration 300

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

MIT