DJRHails/pi-interactive-subagents
Interactive subagents for pi — spawn, orchestrate, and manage sub-agent sessions in cmux terminals
Interactive subagents for pi — spawn, orchestrate, and manage sub-agent sessions in cmux terminals
Open-source CT600 Company Tax Return preparation — GOV.UK Design System, React and FastAPI. Replicates taxpipe.co.uk; demo, does not file with HMRC.
Deterministic multi-agent orchestration: plain JS scripts fan out LLM subagents — control flow is code, only the leaves are model-driven
Economist-style chart theme for matplotlib/seaborn
Automatically configure new devices with all the nice-to-haves
Browser automation CLI for AI agents
Transparent sops-backed encryption for git — plaintext worktree, encrypted history, per-host SSH keys
VS Code extension: CriticMarkup rendered as a unified git diff
Static player for oxbow worlds (generated branching adventures) — GitHub Pages
pi extension: redact env secrets before they reach the LLM provider, restore them in the response (palst-style markers, session-scoped)
VS Code extension: LLM 3-way merge for editor-vs-disk save conflicts
On-Disk Input-keyed Cache — disk-backed memoization with pydantic-aware encoding
Replication of 'When Bad Data Leads to Good Models' (Li et al., ICML 2025, arXiv:2505.04741) — toxic pretraining data, linear representations, and inference-time detox
Read and write Firefox saved passwords by driving the browser's own NSS Secret Decoder Ring.
Estimating frontier base model sizes
Benchmark for chain-of-thought legibility: a monitor reconstructs FreeCell game state and proposed lines from another model's reasoning trace, scored by the engine
Weak-to-Strong Generalization via Direct On-Policy Distillation
Bidirectional sync between a Slidev markdown deck and Google Slides as native, editable objects
Auto-names Pi sessions with AI-generated summaries
Inference API for many LLMs and other useful tools for empirical research
Pure-Python reader for Chromium localStorage, SessionStorage, IndexedDB, and the underlying LevelDB / V8 / Blink formats. No native deps. MIT.
Perform entire codebase refactors in a way that is reproducible, testable and reviewable. Convert to FastAPI Annotated in one step.
Native macOS micro-UI for scripts and agents — sub-50ms WKWebView windows with bidirectional JSON communication
BrierBench: OpenReward environment for evaluating LLM forecasting with time-weighted Brier scoring