task-observer: docs/ai/writing-docs updates from 2026-05-13

[divad12]

Target

global-doc: docs/ai/writing-docs.md

Observations addressed

• Observation 2 (dotfiles/log.md): For judgment-heavy behavior, rationale is part of the instruction — when compressing agent instructions, preserve short rationale paragraphs for rules that depend on model judgment → changed: added bullet to Contract section stating that for hard rules and judgment-heavy behaviors, the failure-mode rationale must be preserved even when compressing
• Observation 3 (dotfiles/log.md): Code guardrails need failure-mode rationale — compressed hard rules should retain what goes wrong without them and what thought pattern is a stop signal → changed: same bullet above covers both; models need the why to recognise when a rule applies under pressure, not just the what

Skipped from same log: Observation 1 (already in AGENTS.md: "remove redundancy without softening corrective guardrails") and Observation 4 (already in writing-docs.md: "Lead with the positive action").

---

Generated by Claude Code