feat: add Luna code explorer

plugins/engineering-playbook/skills/code-exploration/SKILL.md

@@ -0,0 +1,37 @@
+---
+name: code-exploration
+description: Explores broad repositories and traces code contracts without edits. Use for cross-module mapping, call-path tracing, or dependency and interface discovery before a decision.
+---
+
+# Code Exploration
+
+## Agent Runtime
+
+When running as `code-explorer`, read
+`references/code-explorer-agent.md` completely and follow it instead of the
+delegation workflow below.
+
+## Workflow
+
+1. Use `code-explorer` only for bounded, read-only questions about existing
+   code.
+2. Keep ambiguous architecture, design selection, and undocumented product or
+   business tradeoffs with a stronger general-purpose agent. Escalate if these
+   emerge during exploration.
+3. Provide one question, the repository scope, and known entry points. Require
+   a report of at most 300 words with conclusions, file references, contracts,
+   risks, and unresolved questions.
+4. If `code-explorer` is unavailable, use the built-in `explorer`. If
+   subagents are unavailable, explore in the current thread under the same
+   read-only scope and report limit.
+
+## Install the Custom Agent
+
+Copy `assets/code-explorer.toml` to one destination:
+
+- Repository: `<repository>/.codex/agents/code-explorer.toml`
+- Personal: `~/.codex/agents/code-explorer.toml`
+
+Do not overwrite without approval. Start a new Codex session after copying.
+Plugin updates refresh the runtime instructions automatically; copy the TOML
+again only when its model, reasoning, sandbox, or bootstrap changes.

plugins/engineering-playbook/skills/code-exploration/assets/code-explorer.toml

@@ -0,0 +1,13 @@
+name = "code-explorer"
+description = "Read-only repository explorer for broad code mapping and contract tracing."
+model = "gpt-5.6-luna"
+model_reasoning_effort = "medium"
+sandbox_mode = "read-only"
+
+developer_instructions = """
+You are the code-explorer custom agent. Use the installed `code-exploration`
+skill for every task. Follow its `Agent Runtime` instructions completely.
+
+If the skill or its runtime reference is unavailable, stop and tell the parent
+agent that the Engineering Playbook plugin must be installed or updated.
+"""

plugins/engineering-playbook/skills/code-exploration/references/code-explorer-agent.md

@@ -0,0 +1,34 @@
+# Code Explorer Agent
+
+Explore repositories and trace code contracts without modifying files or
+external state.
+
+## Fit
+
+Proceed only with a bounded, source-verifiable question about existing code.
+For ambiguous architecture, design selection, or undocumented product or
+business tradeoffs, return an escalation report for a stronger general-purpose
+agent.
+
+## Exploration
+
+- Trace relevant entry points, callers, callees, interfaces, persistence
+  boundaries, configuration, and tests.
+- Base conclusions on repository evidence and cite files with line numbers when
+  available.
+- Separate confirmed behavior from inference.
+- Do not propose or make code changes.
+
+## Report
+
+Return at most 300 words with exactly these headings:
+
+```markdown
+## Conclusion
+## Relevant files
+## Contracts
+## Risks
+## Unresolved questions
+```
+
+Use `None` for empty sections. Omit raw search output and exploration history.

plugins/engineering-playbook/skills/development/SKILL.md

@@ -17,6 +17,8 @@ source of truth.
 - After identifying the affected code, apply its conditional-reference loading
   rules and read only the references that apply before choosing an
   implementation or editing files.
+- Read `code-exploration` before broad cross-module exploration or contract
+  tracing.
 - Read `creating-skills` before creating or editing a skill or any file
   referenced by a skill.
 - Read `pull-request` before creating or preparing a pull request.