feat(catalog): expand 17 core skills and 11 subagents to enterprise contract-first specifications (v1.1.2)

[sagarrathi16]

Summary

This PR expands 17 core skills and 11 subagents across the indium-agentkit catalog into full enterprise, contract-first specifications (averaging 200–450+ lines per file).

It introduces structured JSON input/output schemas, mathematical formulas, systematic phase-by-phase execution workflows, anti-pattern catalogs, reference Python scripts, and automated verification rules across the entire agentkit inventory.

Package version is bumped to 1.1.1 in package.json.

---

Key Changes

🛠️ 17 Core Skills Expanded (skills/)

All 17 core skills were expanded to contract-first execution specifications (~250–450 lines each):

1. accessibility-audit: WCAG 2.1/2.2 AA & AAA contrast math (L = 0.2126R + 0.7152G + 0.0722B), Playwright + axe-core runner, touch target bounds.
2. author-agentkit-content: YAML frontmatter validator, Cursor rule generator, catalog sync rules.
3. delegate-work: Bounded subagent delegation packets (codex_delegate.py), read-only isolation, single-agent fallback guarantees.
4. estimate-work: Work Breakdown Structure (WBS) 4-layer model, PERT 3-point calculation (E = (O + 4M + P) / 6), 95% confidence intervals (CI 95%).
5. llm-eval-harness: RAG Triad evaluation metrics (Faithfulness, Context Precision/Recall, Cosine Similarity), LLM-as-a-judge position swap.
6. load-testing-suite: k6 & Locust performance testing, Little's Law queuing math (L = λ · W), latency SLA bounds (P95 < 200ms).
7. mobile-release-safety: iOS Privacy Manifests (PrivacyInfo.xcprivacy), Android Target SDK (>= 34), dSYM/ProGuard mapping upload.
8. onboard-to-codebase: Developer orientation generator, repository anatomy scanner, 15-minute time-to-first-green-build SLA.
9. plan-change: Implementation plan validator (REQUIRED_PLAN_HEADINGS), acceptance criteria, non-goals bounds.
10. prototype-spike: Time-boxed throwaway spikes, scratch directory isolation (scratch/), latency/memory benchmarking.
11. refactor-code: Atomic refactoring transformations, test assertion immutability, cyclomatic complexity reduction.
12. release-notes: Keep a Changelog 1.0.0 formatting, Conventional Commit categorization (feat, fix, BREAKING CHANGE), Python git changelog generator.
13. resolve-merge-conflicts: Three-way diff inspection (:1:base, :2:ours, :3:theirs), zero leftover conflict markers (Count("<<<<<<<") = 0), lockfile regeneration.
14. review-change: Read-only multi-dimensional code reviews, severity scoring (BLOCKER to NITPICK), Python diff analyzer.
15. test-first-change: TDD Red-Green-Refactor execution loops, explicit Red phase failure verification, minimal production changes.
16. verify-and-ship: Pre-ship verification pipeline, secret scanning, Cursor rule generation, catalog sync, atomic git commits.
17. write-documentation: Technical documentation authoring (README, ADR, API Reference), executable code snippet validation, relative link verifier.

🤖 11 Subagents Deepened (agents/)

All 11 subagents were expanded to 195–272+ lines each with JSON input/output schemas, 5-phase systematic workflows, anti-pattern catalogs (Bad vs Good), and severity matrices:

• accessibility-checker (222 lines)
• doc-writer (221 lines)
• estimator (210 lines)
• explorer (207 lines)
• frontend-builder (202 lines)
• llm-evaluator (194 lines)
• migration-planner (223 lines)
• mobile-specialist (217 lines)
• release-engineer (204 lines)
• reviewer (228 lines)
• verifier (195 lines)