build-first-reuse-default — operating philosophy

Census id: F6 · Family: F — rules corpus + generated-rule tooling + shipped audit scripts · Provenance: framework pin b069c59328aa8e08671d3a19fd25efa618dfc1c5 (staging FETCH_HEAD, read 2026-09-11; re-anchored from census pin a1337cb301). Every file:line below is at that pin. Raw reference drafted by stage E4 — factual, unpolished by design.

What it bans

Reinventing what upstream already ships, and macro-level parallel evolution. .claude/rules/build-first-reuse-default.md:15 «# Build-first, reuse-default — operating philosophy»; the core contract: .claude/rules/build-first-reuse-default.md:32 «Every capability proposed for this project resolves into ONE of seven verdicts:» (the verdict table :34-42: ADOPT / ADOPT VOCABULARY / ADAPT / REFERENCE / KEEP NARROW / BUILD / REJECT), and .claude/rules/build-first-reuse-default.md:44 «Default = ADOPT or REFERENCE. BUILD only after §3 mechanism confirms a documented problem-class mismatch or load-bearing gap.» Class header: :17 «> Class: A — companion principle test shipped at packages/core/principles/11-build-first-reuse-default.test.ts (#75, 2026-05-17).»

Never (fires)

BUILD chosen without the prior-art mechanism, or adoption refused out of vanity. Named anti-patterns: .claude/rules/build-first-reuse-default.md:84 «- #parallel-evolution-creep — building parallel to existing production tool because per-commit decisions never composed at scope level. Counter: this rule (macro check).»; :85 «- #own-stack-blind-spot — believing «our problem is unique» because we never surveyed upstream. Counter: §3 mechanism mandatory before BUILD.»; :86 «- #adoption-shame — refusing to adopt because «we can do it our way» (vanity ≠ technical reason). Counter: explicit verdict justification, not «we prefer to build».» The BUILD bar in full: :41 «| BUILD | Write ourselves; no upstream / fundamental misfit / load-bearing gap | Confirmed via §3 mechanism that no production-grade upstream candidate exists for our problem-class |»

Always (clean)

A capability commit carrying one of the seven verdicts with the §3 mechanism behind it — six layers, quoted: .claude/rules/build-first-reuse-default.md:71 «1. Prior-art SSOT trailer required on capability commits (already enforced via .husky/pre-push).» … :75 «3. DeepWiki ask_question MCP tool for repository-level inquiry … Use ≥3 phrasings.» and :76 «4. WebSearch ≥3 phrasings on problem-domain term — for general state-of-art surveys («what frameworks exist for problem-class Z?»). Counters context7-only training-data bias.» The tooling caveat names what does NOT count: :80 «> Tooling caveat: context7 MCP is intentionally excluded from this list.»

Enforcement channels

• Rule-index row: .claude/rules/00-rule-index.md:14 «| build-first-reuse-default.md | A | any capability commit / new-capability proposal. | paths:(7), edit-time inject |»
• Edit-time inject, 7 globs: .claude/rules/build-first-reuse-default.md:12 «» with inject text at :13 (7-verdict reminder, «BUILD requires SSOT consult + DeepWiki+WebSearch ≥3 phrasings»).
• Companion principle test (CI, shipped): packages/core/principles/11-build-first-reuse-default.test.ts (#75; named at :17 and :93).
• Pointer carriers (post-2026-08-08 re-scope, :17/:126): «always-on pointer carriers: session-bootstrap digest invariant (1) + CLAUDE.md per-commit gate.»
• Not backend-rendered — no FF diagnostic applies (FF7001/FF7002, packages/core/diagnostics/registry.ts:321/:329, are backend render refusals); honest status: path-scoped inject + CI principle test + husky pre-push prior-art trailer.