Writing on code-level context, effect propagation, and keeping agent-authored code accountable. A new post lands each week.
File-backed rules that allow, deny, or require approval per action and actor — evaluated on every ledger mutation and emitted to the audit stream.
Read more →Every ledger mutation and policy decision, appended to a tamper-evident log where each entry references the prior one's hash — replayable and independently verifiable.
Read more →Append-only decisions, hazards, and rationale attached to a symbol — surviving renames, file moves, and the person who wrote them leaving.
Read more →Seventeen effect categories, declared per symbol and pushed transitively through the call graph, so an agent knows what a change reaches before it makes it.
Read more →A diff tells you what changed. It never tells you why the code exists, what it touches, or who approved it — and that's exactly the context an agent needs.
Read more →