chore: remove legacy tools/codex-skills (moved to .agents/skills), split filter drawer tests, clean up stale docs

- tools/codex-skills/ removed — skills now live in .agents/skills/
- docs/reference-agents-*.md removed — replaced by .agents/workflow.md
- FilterDrawerContractTest split into Runtime + Template contract tests
- bin scripts and .gitignore updated for .agents/ paths

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

docs/reference-agents-flow.md

@@ -1,42 +0,0 @@
-# Agent Workflow
-
-Letzte Aktualisierung: 2026-03-06
-
-## State Machine
-- `planned`
-- `executing`
-- `review_guard`
-- `review_acceptance`
-- `finalize`
-- `done`
-
-If a reviewer returns `FAIL`, state goes back to `executing` with explicit findings.
-
-## Minimum Handover Artifacts
-- Planner: `plan.json`
-- Executor: `execution-report.json`
-- Reviewer Guards: `review-guards.json`
-- Reviewer Acceptance: `review-acceptance.json`
-- Finalizer: `finalize.json`
-
-## Guard and Gate Binding
-- Planner selects required guard IDs from `agent-system/checks/guard-catalog.json`.
-- Planner selects required quality gate IDs from `agent-system/checks/quality-gates.json`.
-- Planner assigns stable success criteria IDs (`SC-*`) in `plan.json`.
-- Executor must provide guard evidence and gate results by ID.
-- Guard reviewer verifies and reports verdict against the same IDs.
-- Acceptance reviewer verifies and reports verdict against the same `SC-*` IDs.
-
-## Fail Loop Rule
-- No free-text "please improve".
-- Every fail must include:
-  - `id`
-  - `severity`
-  - `file`
-  - expected fix
-
-## Practical Rollout
-- Start with one small issue.
-- Run workflow manually with templates.
-- Stabilize contracts first.
-- Add automation only after 3 to 5 successful runs.

docs/reference-agents-overview.md

@@ -1,29 +0,0 @@
-# Agent Workflow Overview
-
-Letzte Aktualisierung: 2026-03-09
-
-This project uses a multi-role workflow for issue and feature delivery.
-
-Roles:
-- Planner
-- Executor
-- Reviewer (guards and best practices)
-- Reviewer (feature acceptance)
-- Finalizer
-
-Primary goals:
-- predictable handovers
-- strict quality gates
-- fast retry loop from reviewers back to executor
-- explicit guard and gate IDs (`GR-*`, `QG-*`) across all roles
-
-Source of truth:
-- guardrails and planning standards in `tools/codex-skills/`
-- contracts and templates in `agent-system/`
-- guard catalog: `agent-system/checks/guard-catalog.json`
-- quality gates: `agent-system/checks/quality-gates.json`
-
-Entry points:
-- roles: `/docs/reference-agents-roles.md`
-- workflow: `/docs/reference-agents-flow.md`
-- schemas and prompts: `agent-system/`

docs/reference-agents-roles.md

@@ -1,49 +0,0 @@
-# Agent Roles
-
-Letzte Aktualisierung: 2026-03-06
-
-## Planner
-- Input: issue or feature request
-- Output: `plan.json` (see `agent-system/contracts/planner.schema.json`)
-- Must define:
-  - scope in and out
-  - required guard IDs (`GR-*`)
-  - required quality gate IDs (`QG-*`)
-  - success criteria with stable IDs (`SC-*`)
-  - implementation steps
-  - test and acceptance checks
-  - risks and mitigations
-
-## Executor
-- Input: approved `plan.json`
-- Output: `execution-report.json`
-- Must:
-  - implement the plan
-  - run relevant checks
-  - document changed files and commands
-
-## Reviewer Guards
-- Input: code diff and `execution-report.json`
-- Output: `review-guards.json`
-- Focus:
-  - architecture boundaries
-  - security
-  - coding standards
-  - guardrail compliance
-  - explicit validation against required `GR-*` and `QG-*` IDs
-
-## Reviewer Acceptance
-- Input: code diff and plan success criteria
-- Output: `review-acceptance.json`
-- Focus:
-  - feature correctness
-  - edge cases
-  - acceptance criteria coverage
-  - explicit check mapping by `SC-*` criterion IDs from `plan.json`
-
-## Finalizer
-- Input: both review files with PASS verdict
-- Output: `finalize.json`
-- Must:
-  - verify gates are green
-  - prepare merge and commit metadata