Provenance

Every file here was written or rewritten for this repository, in second person, with nothing personal left in it. This page records where the ideas came from, as history.

• Core rules (working-style, decisions-and-plans, delegation, transcript-hygiene, voice-and-format, research-and-verification) grew out of a personal rules directory built over a year of running coding agents, most of it in Cursor and the last two months in Claude Code, across a strategy game (Sovereign), an executive-assistant app (Consul), a governance MCP server (Cortex), and volunteer data work.
• conciseness and verification were first written as guest-contributor rules for a fork of an open-source data-pipeline library, where the cost of a sloppy PR is someone else’s time.
• secrets and the migration-safety skill came from a volunteer team’s tooling, where staging held real data.
• The testing stance came from a product-ideas planning repo; the autonomy stance from the governance server’s own rules and from a Codex configuration.
• The worktree-per-agent and spike-contract skills came from the strategy game’s repo instructions (Sovereign), where several agents worked the same checkout and spikes had to be read against measured numbers.
• The design-loop skill is derived from dream-loop (MIT); see THIRD_PARTY_NOTICES.md.
• The plan-card validator and the read-only hook were written for this harness after too many permission prompts in plan mode and too many plans nobody could review in one screen.

The audit that decided what to bring in and what to leave out is summarised in field-scan.md’s properties: anything that was true of one person, one project, or one employer stayed out.