self-reflectionlisted

<!-- @harness-posture: portable — prose self-application checklist over repo artefacts; no harness primitives --> # Self-reflection — recommendation discipline gate > **Authoritative for:** skill activation conditions (frontmatter `description`); §1.7 forward+backward checklist summary; output contract for discipline-introducing recommendations; pointers to cold references. > **NOT authoritative for:** the §1.7 rule itself — see [`.claude/rules/phase-research-coverage.md §1.7`](../../rules/phase-research-coverage.md). Project goal — see [README.md#why-this-exists](../../../README.md#why-this-exists). ## Why this skill exists Three documented occurrences of the same shape in 2026: 1. **PR #16** — `EXECUTION-PLAN §1` silently re-defined the project goal in the doc that was supposed to _prevent_ drift. 2. **Prior session (4 turns)** — research about applying discipline-from-start, with the assistant repeatedly applying «defer until consumer pain» framing — the _opposite_ of project's thesis, in a session whose subject was project's thesis. 3. **2026-05-09 L3 generated-docs research** — recommendation about doc-authority discipline failed forward+backward checks across 6 existing project disciplines; gap surfaced via reviewer pushback, not via own self-audit pass. Same root cause: **meta-cognitive blindspot** — the agent of analysis is not also the object of analysis. When the assistant reasons about discipline X, attention loads subject domain + prior art + trade-offs but