feat(skills): enterprise-analyst rework of authoring skill prompts

[ops-gaurav]

Summary

Reworks the phase-authoring skill prompts to behave like a senior analyst / architect / tech lead, across all three adapters (Anthropic, OpenAI Codex, Cursor — kept byte-consistent).

Four operating principles now run through the authoring skills:

• Investigate before asking — explore the repo (manifests, related modules, tests) and research external services/libraries before asking the user anything
• Ask informed questions, in one batch — 3–5 focused questions max, each with 2–3 concrete options and a recommended default
• MVP first — P0 set alone must describe a shippable product; deferred ideas are recorded in Future Iterations, not dropped
• Plain professional documents — readable by someone joining cold; inapplicable sections say why instead of filling boilerplate

Per-skill changes

Skill	Change
spec-requirements	Discovery step, batched gap questions, Summary / Goals & Non-Goals / Out of Scope & Future Iterations sections
spec-design	Codebase-convention discovery, open decisions presented as option sets, Design Decisions record, Future Considerations, right-sized sections
spec-tasks	Traceability table (F-XX/NF-XX → tasks), walking-skeleton Phase 1, S/M sizing, per-task Done when: criteria, testing inside each phase
spec-implement	Fix: now reads design.md before building; orients on codebase conventions; verifies (tests + done criterion) before flipping checkboxes
spec-new	Captures spec type + one-line summary, proposes kebab name, duplicate-spec check
spec-research	Scouts ambiguous scope first, then asks batched informed questions
review checklists	Updated for new structures; added MVP-fit and Fit & Simplicity groups

Also updates docs/generated-artifacts.md and normalizes Claude-specific tool names in the Cursor rules to generic phrasing.

Notes

• Stacked on feat(spec-review, spec-revise): emit review file + add revise skill #9 (feat/review-skill-generate-output) — the review-file/revise content these prompts reference lives there. Retarget to main after feat(spec-review, spec-revise): emit review file + add revise skill #9 merges.
• Housekeeping skills (approve, status, switch, update-task, sync) unchanged; spec-sync parsing format untouched.
• Verified adapter sync: anthropic ↔ openai-codex trees identical; cursor .mdc bodies regenerated from skill bodies.

🤖 Generated with Claude Code