chore(skills): tier-1/tier-2 split + sharpen content + README intro

[HumanBean17]

Why

Skills under skills/<name>/ underperformed docs/AGENT-GUIDE.md as a reference. When the agent loaded a single SKILL.md it often got generic boilerplate (Q-class ritual without an inline taxonomy, "uses all 5 tools" headers even when the skill only calls find), and it could not opt into just navigation listings without also dragging in the multi-step workflows. Two changes address that:

1. Layout: split the flat tree so users can install per tier.
   • skills/tier-1/ — 11 single-purpose listings (one MCP chain each)
   • skills/tier-2/ — 4 multi-step workflows that compose Tier 1 with bounded recursion and explicit stop conditions
2. Content: each SKILL.md rewritten to be self-sufficient.
   • frontmatter description sharpened for auto-discovery (concrete user phrasings)
   • per-skill Tools used line (only the tools that skill actually calls)
   • Reasoning preamble with inline Q-class taxonomy: semantic → search, structured → find/resolve, inspect → describe, walk → neighbors
   • Microservice scoping caveat surfaced inline (microservice is not validated; an invalid name returns an empty result, not an error)
   • Recovery section per skill, including the two-attempt stop rule (after two failed attempts on the same intent, stop and report)
   • Worked example showing the reasoning preamble + the MCP calls

README.md gets a new Why this exists section above the install block, covering the five differentiators that distinguish this from grep / ctags / vector-only RAG:

• Hybrid RAG + GraphRAG (not either-or)
• Java-tuned role model (CONTROLLER / SERVICE / REPOSITORY / CLIENT / PRODUCER / MAPPER / DTO)
• Ranking specialized for Java (role + microservice + FQN aware)
• Cross-service resolution + system-level navigation (HTTP_CALLS / ASYNC_CALLS)
• Brownfield annotations with exclusive priority

What changed

Area	Change
skills/tier-1/*	11 SKILL.md rewritten; new layout
skills/tier-2/*	4 SKILL.md rewritten; new layout
skills/README.md	Index split per tier, philosophy block referencing the 5 differentiators, install instructions per tier
README.md	New ## Why this exists section between the GPS metaphor and the install block
AGENTS.md	"Two audiences, two skill trees" table + Where-to-look entry updated to point at skills/tier-1/ and skills/tier-2/
tests/test_agent_skills_static.py	Validator refactored to walk both tier dirs and assert no skills remain at skills/<name>/

No code changes. No MCP behavior changes. No new dependencies.

Propose amendments (captured here)

This PR amends two decisions from propose/completed/AGENT-SKILLS-AND-COMMANDS-PROPOSE.md. Both are intentional and called out so future readers know to read this PR alongside the propose.

1. /producers is a Tier 1 skill (symmetric to /clients). Locked propose enumerated 10 Tier 1 skills; /producers was added in PR feat(skills): add 14 navigation and workflow skills + static validator #225 to keep outbound-side symmetry. Tier 1 ships at 11 in this PR.
2. Skills layout changes from skills/<name>/ to skills/tier-N/<name>/. Locked propose specified the flat layout. Splitting per tier lets users opt in to listings without workflows (or the reverse).

Tests

• pytest tests/test_agent_skills_static.py — 140 passed (was 140 before; same coverage, refactored asserts)
• pytest tests/ — unaffected (no code changes)

Out of scope

• Tier 3 (CLI helpers) — already out of scope per the locked propose.
• docs/AGENT-GUIDE.md — skills and AGENT-GUIDE are explicitly two separate artifacts; AGENT-GUIDE is unchanged.
• .agents/skills/ — developer-facing workflow skills, different audience.