nWave-ai nw-ab-house-style (KNOWLEDGE)
Kind: KNOWLEDGE (reference). Loaded by nw-ab-create-agent (CREATE) and nw-ab-migrate-monolith (EXTRACT). No forced sequence.
House style
Author caveman-curated: dry, declarative, tables and compact one-line bold-lead lists, zero narrative padding. Agent body lean (~100-200 lines, role+routing+contract); deep knowledge in skills (<5000 tokens each). Exemplar: nw-security-analyst — 104-line agent (93 body), deep knowledge in 2 skills (96+107=203). Byte-exact preservation: ### Example N: headers, NORMATIVE blocks, code/YAML, AskUserQuestion trees, machine content. User-facing templates stay clear, not compressed.
By-construction guarantees
- House style — every created/modified asset is caveman-curated (above).
- Reasoning-mandate injection — every created/modified agent gets the
## Reasoning Mandateblock inserted verbatim. Depth from rigor, never padding. - Opportunistic retro-compression (measured-gain ≥20% or skip) — on MODIFY: measure prose share first. Projected reduction ≥20% → compress with hard invariants (pre-existing blocks/tables byte-identical,
### Example N:+ NORMATIVE preserved, content-pin tests green). Below 20% → SKIP and report the measurement. Hash-pinned skills (md5 baseline intests/build/unit/test_skill_restructuring.py): never silently edit; REPORT the required hash bump, do not bump it yourself.
Reasoning Mandate block (inject verbatim)
## Reasoning Mandate (Caveman)
Verdict-first, tables over prose, evidence-dense, zero narrative. Depth comes from rigor, not padding. State the conclusion, then the supporting evidence; never bury the verdict under exposition.