feat(AI-82): port chapter generation prompts into SKILL.md Step 3

[dastratakos]

Summary

Replace the Step 3 TODO in skills/stage-chapters/SKILL.md with the full chapter generation prompt, ported from the Stage monorepo's analysis-agent.ts and narrative-agent.ts system prompts. After this change, an agent following the skill produces real Stage-quality chapters with clustering, narration, and key-change extraction in a single pass.

Changes

• 3a Clustering rules: causal grouping, multi-file chapters, moves/refactors as single chapters, foundation-first ordering, hunk ordering within chapters
• 3b Self-validation rules: every hunk in exactly one chapter, identified by exact (filePath, oldStart) from @@ headers
• 3c Narration rules: action-oriented titles (max 8 words), 2–3 sentence impact-first summaries with causal links, markdown formatting
• 3d Key change rules: judgment-call questions only, empty array when nothing needs human input, tight lineRefs with OLD/NEW column guidance, all good/bad examples from narrative-agent.ts
• 3e Output format: chapter object shape matching ChaptersFileSchema

Testing

• Verified all rule categories from the acceptance criteria are present (clustering, self-validation, narration, key changes, output format)
• Confirmed output shape aligns with ChaptersFileSchema in packages/cli/src/schema.ts and hunkReferenceSchema/lineRefSchema in packages/types/src/chapters.ts

---

[gemini-code-assist]

Code Review

This pull request replaces a placeholder in SKILL.md with comprehensive instructions for clustering and narrating code changes into 'chapters,' defining rules for grouping hunks, narration style, and identifying key changes. A review comment suggests ensuring the output format uses standard JSON instead of jsonc to avoid parsing errors if comments are included in the generated output.