Inspiration
Developers remember the final commit but often lose the reasoning that made it possible: a documentation page that unblocked them, a failed test that narrowed the problem, and the validation that proved the result.
What it does
ShadowLog creates a local, privacy-controlled evidence graph from three opted-in sources:
- Approved technical research in Chrome
- Metadata-only VS Code save activity
- Sanitized terminal outcomes
It connects research → attempt → verified result, then generates standups, commit drafts, and performance-review bullets. Developers can inspect, correct, or delete every event before export.
Privacy by design
ShadowLog never stores source code, page content, URL queries/fragments, raw terminal commands, arguments, secrets, or environment variables. Local SQLite storage, a user-controlled browser allowlist, path redaction, and per-event deletion keep the evidence auditable and private.
How it works
The local Node.js companion stores privacy-safe events in SQLite, groups them into 15-minute work blocks, and assigns confidence based on independent signals such as research, saved-file metadata, and test outcomes. Selected redacted blocks are sent to Gemini 2.5 Flash-Lite for structured updates; an offline deterministic fallback keeps the synthetic replay demo runnable without an API key.
Built with Codex
Codex with GPT-5.6 accelerated the core implementation: the event contract, local SQLite daemon, Chrome and VS Code adapters, redaction rules, real-time dashboard, VSIX packaging, synthetic replay, and regression tests. Gemini powers the runtime text-generation call because no OpenAI API credits are available.
Try it
Run node src/daemon.js --seed, then open http://localhost:4318. The repository includes synthetic replay data, tests, Chrome/VS Code adapter instructions, and a VSIX packaging script for judge testing.
Built With
- chrome
- codex
- gemini-api
- gpt-5.6
- javascript
- node.js
- sqlite
- visual-studio-code-extension