I build data pipelines and the tooling around LLMs. Most of what I build never shows up here; the public stuff is whatever stayed useful more than once.
Lately that means bug fixes in LLM and agent libraries. The bugs tend to be the same kind: a layer that reports something other than what the layer under it did.
github/spec-kit: 5 merged PRs on agent dispatch, bundle pins, catalogs, JSON settings
facebookincubator/muse-gadget-sdk: 2 merged PRs on system.runexits andfile.writemodes
Everything else I've had merged
- ai-quality-gate: offline checks for recorded LLM output (schema, grounding, snapshots) that fail CI with an exit code and never call a model. On PyPI as
aiqg. - actgate: an MCP proxy that holds an agent's tool calls until a person approves them, and records each decision in a hash-chained ledger. On PyPI as
actgate. - djinni-market-etl: Djinni jobs and candidate profiles in Postgres, with each vacancy tracked until it closes.
- StableInvoice: USDC escrow invoices on Solana. The client funds a vault up front, and payment goes out in one lump sum once every milestone is accepted.


