Reviewed knowledge · exact delivery · honest evidence

Give agents the guidance your team trusts.

Documentation explains. Honemill operationalizes distilled engineering guidance into reviewed canon. See exactly what reached each registered run — and how a human judged the result.

Works with Claude Code · Codex · Cursor · any MCP agent

$ honemill loop sync --repo api
pr-review · active
recall top 5 · manifest pinned
$ honemill memory inbox --repo api
◇ gotcha · “serialize migrations”
awaiting human review

The flywheel

Sources, agents, and humans propose engineering knowledge. Humans review it. Registered runs receive exact canon revisions, then human verdicts show which guidance held up. Scroll: the stone turns with you.

the flywheel01 / 04 · run
reviewrecallrunjudge

01 / 04Review

Sources propose. Humans decide.

Selected documentation, agents, and humans put distilled conventions, decisions, and gotchas in a review inbox. Only a human can promote a revision into normal recall.

$ honemill memory inbox --repo api
◇ gotcha · “serialize migrations”
human promoted revision r-2

scroll — the stone turns with you

Knowledge · Human reviewed

Canon, not conversation exhaust

Agents, humans, and selected Git documentation can propose conventions, decisions, and gotchas. A human approves the exact revision before normal recall.

Delivery · MCP + CLI

Scoped guidance, every agent

Claude Code, Codex, Cursor, and other MCP clients share the same reviewed canon. Repository and team scope keep guidance inside the database trust boundary you choose.

Evidence · Exact lineage

Know what reached the run

Every prepared run pins revisions, provenance, scope, method, score, rank, manifest, and policy. Human verdicts add evidence without turning observational association into a causal claim.

PHASE 1 IS IMPLEMENTED. The CLI, three-tool MCP server, Git-tracked Markdown importer, exact delivery lineage, recovery envelope, and association report are working. Two-week dogfood acceptance comes next.

Give coding agents only the engineering guidance your team has reviewed.

Get early accessRead the docsBring your own agent — if it speaks MCP, it works.

or write to hello@honemill.ai