fix(rldyour-claudecode): track AGENTS.md and .claude/CLAUDE.md on main as system files

.claude/CLAUDE.md

@@ -0,0 +1,71 @@
+# Claude Code Project Memory - rldyour-claudecode
+
+This is the Claude Code-native deep memory for the rldyour Claude adapter.
+Cross-tool overview lives in `AGENTS.md`; this file keeps Claude-specific
+runtime, hook, subagent, and skill-budget facts concise.
+
+<!-- sync_contract:
+claims:
+  mcp_owner: rldyour-mcps
+  hook_owners: [rldyour-flow, rldyour-serena-mcp]
+  lsp_owner: rldyour-lsps
+  reviewer_transport_marker: RLDYOUR_REPORT_EOF
+  reviewer_report_dir_template: ".serena/reviews/<run_id>/"
+  reviewer_run_id_format: "<UTC-ISO-compact>-<git-short-sha>"
+  claude_code_runtime_pin: "2.1.177"
+  claude_code_feature_floor: "2.1.146"
+  skill_listing_budget_fraction: 0.04
+  max_skill_description_chars: 1536
+  fullrepo_branch: fullrepo
+  plugin_count: 10
+  skill_count: 38
+  command_count: 11
+  subagent_count: 8
+-->
+
+## Runtime
+
+Claude Code runtime pin is `2.1.177`. The compatibility floor is `2.1.146`.
+The pin sources are `package.json`, `references/claude-baseline.json`, and
+`config/mcp-runtime-versions.env`.
+
+## Plugin Inventory
+
+- `rldyour-mcps`: MCP transport owner for 11 active MCP servers.
+- `rldyour-serena-mcp`: Serena code workflow, memory sync, and lifecycle hooks.
+- `rldyour-flow`: SDLC commands, reviewer agents, post-task sync, and Stop
+  lifecycle dispatcher.
+- `rldyour-orchestrator`: macOS cmux orchestrator/worker role skills only.
+- `rldyour-explore`, `rldyour-security`, `rldyour-browser`, `rldyour-design`,
+  `rldyour-lsps`, and `rldyour-rules`: domain plugins.
+
+Current inventory: 10 plugins, 38 skills, 11 slash commands, and 8 subagents.
+
+## Hook Canon
+
+Only `rldyour-flow` and `rldyour-serena-mcp` own hooks. Stop hooks are advisory
+enforcement gates: they compute state and block with guidance, while the main
+workflow performs memory sync, tests, commits, pushes, and fullrepo publication.
+The single registered Claude Stop hook is the Flow dispatcher.
+
+| Event | Owner | Script | Timeout |
+|---|---|---|---|
+| Stop | rldyour-flow | `hooks/stop_lifecycle_dispatcher.sh` | 45s |
+
+## Reviewer Contract
+
+Reviewer subagents use file-first output. Full reports go to
+`.serena/reviews/<run_id>/<reviewer-name>.md`; parent-visible output stays
+compact. The heredoc marker is `RLDYOUR_REPORT_EOF`.
+
+## Skill Budget
+
+Use `skillListingBudgetFraction: 0.04` and `maxSkillDescriptionChars: 1536` in
+user Claude settings so bilingual skill descriptions remain discoverable.
+
+## Boundaries
+
+Claude subagents are not cmux orchestrators. Webwright, Playwright CLI, and
+Chrome DevTools MCP keep their global browser-provider roles. Configure only
+providers listed in the approved active inventory and avoid tool-specific
+negative validator tombstones.

AGENTS.md

@@ -0,0 +1,88 @@
+# AGENTS.md - rldyour-claudecode
+
+rldyour AI CLI configuration for Claude Code: plugin marketplace, MCP/LSP,
+Serena memory, SDLC flow, browser/design workflows, security review, and
+reviewer agents. This repository contains Claude Code-native plugin metadata,
+skills, slash commands, subagents, hooks, scripts, docs, and references; it has
+no runtime application service.
+
+This file is a system file for the Claude Code install surface. It is
+tracked on `main` so that `git clone` followed by the Claude install
+script (or Claude Code's own `.claude/CLAUDE.md` discovery) picks it up
+without depending on the `fullrepo` overlay. The companion file
+`.claude/CLAUDE.md` is also tracked on main for the same reason.
+
+Canonical public description: rldyour AI CLI configuration for Claude Code: plugin marketplace, MCP/LSP, Serena memory, security review, browser/design workflows, and reviewer agents.
+
+<!-- sync_contract:
+claims:
+  mcp_owner: rldyour-mcps
+  hook_owners: [rldyour-flow, rldyour-serena-mcp]
+  lsp_owner: rldyour-lsps
+  reviewer_transport_marker: RLDYOUR_REPORT_EOF
+  reviewer_report_dir_template: ".serena/reviews/<run_id>/"
+  reviewer_run_id_format: "<UTC-ISO-compact>-<git-short-sha>"
+  claude_code_runtime_pin: "2.1.177"
+  claude_code_feature_floor: "2.1.146"
+  skill_listing_budget_fraction: 0.04
+  max_skill_description_chars: 1536
+  fullrepo_branch: fullrepo
+  plugin_count: 10
+  skill_count: 38
+  command_count: 11
+  subagent_count: 8
+-->
+
+## Source Of Truth
+
+- `.claude-plugin/marketplace.json` is the marketplace manifest.
+- `plugins/*/.claude-plugin/plugin.json` are per-plugin manifests.
+- `plugins/rldyour-mcps/.mcp.json` is the single MCP transport owner.
+- `plugins/rldyour-flow/hooks/hooks.json` and
+  `plugins/rldyour-serena-mcp/hooks/hooks.json` are the only hook owners.
+- `config/rldyour-contract.json` and `docs/contract-matrix.md` define the
+  cross-tool contract.
+- `references/claude-baseline.json`, `package.json`, and
+  `config/mcp-runtime-versions.env` pin Claude Code `2.1.177`.
+
+## Native Boundaries
+
+- Use Claude Code plugin, skill, slash-command, subagent, hook, MCP, and LSP
+  surfaces only.
+- Do not copy Codex, OpenCode, or Gemini-native command/config syntax into this
+  adapter except as explicitly marked comparison notes.
+- Browser routing stays global: Webwright for long-horizon workflows,
+  Playwright CLI for low-level screenshots/snapshots/traces, and Chrome
+  DevTools MCP for DevTools/performance/network/console/memory/Lighthouse.
+- cmux orchestration is terminal-session-only. Claude subagents are not cmux
+  orchestrators.
+
+## Hook Lifecycle
+
+| Event | Owner | Script | Timeout |
+|---|---|---|---|
+| Stop | rldyour-flow | `hooks/stop_lifecycle_dispatcher.sh` | 45s |
+
+The single registered Claude Stop hook is the Flow dispatcher above. Stop hooks
+are advisory enforcement gates. They compute sync state and block with guidance;
+the main workflow performs memory sync, validation, commits, pushes, and
+fullrepo publication.
+
+## Validation
+
+Run after changes:
+
+```bash
+python3 scripts/validate_plugin_versions.py
+python3 scripts/validate_contract.py
+python3 scripts/validate_release_state.py
+python3 scripts/validate_instruction_docs.py --require-agent-docs
+python3 scripts/validate_claude_surface_adoption.py
+uv run --with pytest --with pyyaml python -m pytest -q
+```
+
+## Fullrepo Policy
+
+`AGENTS.md`, `.claude/CLAUDE.md`, and `.serena/project.yml` are restored and
+published through `fullrepo`. Normal `main` must not track these agent-only
+files.