When Rich asks for any kind of prompt (launch-prompt / dispatch prompt / template / paste block / boilerplate / methodology-codifying artefact): default to comprehensive + fully detailed. Do NOT restrict size.
Per Rich-directive 2026-05-23: “i want prompts to be comprehensive and fully detailed. do not restrict their size” — explicit override of the prior ≤30-line short-paste-block rule.
How to apply:
- Default to detailed methodology coverage. Decision Matrix entries / anti-patterns / pre-condition gates / done criteria / §13.5 author-side mirror + §13.6 grep-verify / honest-substrate-read declarations / Phase-N substrate cross-references / race-resolution bookkeeping — all stay even if they push the artefact past 200/300/500 lines.
- Don’t ask “short or comprehensive?” — assume comprehensive. If brevity is genuinely needed, Rich will say so.
- Don’t apologise for length or attempt to “tighten” via removing content. Tightening = removing redundancy + sharpening language, not deleting load-bearing methodology.
- Companion methodology files (refined-end-of-turn-directive.md §1; substrate-search-locations.md 15+2 steps; CASCADE-Q-TEMPLATE.md §B.0; arch-state §15 amendment registry) remain canonical sources — the prompt can still cite them rather than restating verbatim, but the prompt itself stays full-detail on its own scope.
Secondary rule (independent of size):
Do NOT factor Rich’s personal storage tooling into prompt-workflow advice. Storage (Google Docs / clipboard managers / snippet tools / etc.) is Rich’s concern. Don’t mention them when devising or running prompts. Per Rich-directive 2026-05-23T~10:00 BST: “please forget about using google docs as part of devising and running prompts”.
Why the rule flipped:
The prior rule was anchored on a single incident 2026-05-23 — launch-prompt-request-template.md v1.0 at 274 lines was rejected as “too complicated”. I generalised that to “≤30-line paste blocks”. The generalisation was wrong.
Subsequent same-session evidence (batch-imp-22 v1.0/v1.1/v1.2 at ~300 lines + batch-imp-23 v1.0/v1.1/v1.2 at ~310+ lines + arch-state A-285/A-286/A-287 narratives at ~500-700 words each) all landed without complaint — Rich works comfortably with long detailed prompts.
The 274-line template rejection was almost certainly about shape (Rich wanted a paste BLOCK, I produced a methodology MANIFESTO) + the friction of self-applying a 7-step PART 2 to author each new prompt — NOT about raw length. The length-generalisation was a misread.
This memory now codifies the corrected posture: comprehensive + fully detailed by default; no size restriction; storage tooling stays out of scope.
Historical incident (preserved for forensic record):
- 2026-05-23 —
launch-prompt-request-template.mdv1.0 (274 lines; 7-step PART 2 + 12-section required-structure + 12 anti-patterns + 5-criterion verification + 5 §A-E sections) was deleted post-rejection. Initial memory captured this as a “≤30 lines” rule. Rich-directive same day 2026-05-23T~17:30 BST corrected: prompts ARE meant to be comprehensive; the 274-line template was rejected on different grounds. - Subsequent prompts (batch-imp-22 v1.0 → v1.2; batch-imp-23 v1.0 → v1.2) all landed at 290-310+ lines without size complaint.
Related: feedback-compound-launch-prompts-amplify-context-degradation-risk (different layer — single-action vs compound multi-action scope; orthogonal to length).