How this skill is triggered — by the user, by Claude, or both
Slash command
/cortex-toolkit:cortex-levelThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Places, organizes, modifies, and reviews level content using the Level Designer agent.
Places, organizes, modifies, and reviews level content using the Level Designer agent.
| Request | Mode | Agent turns |
|---|---|---|
| "Place", "Add", "Spawn", "Build", "Move", "Delete", "Organize", "Adjust lighting", multi-actor layout, scene construction | Create/Modify | 25 |
| "Review", "Audit", "Check", "Analyse", "What actors", "List", spatial/organization questions | Review/Analyze | 15 |
| Ambiguous | Default to Review | 15 |
Use the Task tool with subagent_type: "cortex-toolkit:level-designer" and max_turns: 25 to delegate the work.
Structure the prompt using the 3-phase directive:
Make the following level changes using the 3-phase methodology:
**Request:** [user's request verbatim]
**Prefetched state:** [embed the main-thread `prefetched_state` block here before launching]
MANDATORY WORKFLOW:
0. VERIFY: call `get_info` to confirm MCP connectivity. If it fails, invoke `cortex-editor`.
1. Use `prefetched_state` first. Do not re-fetch the same baseline unless required for the next step.
2. Read `.cortex/domains/level.md` for level conventions
3. PLAN: call `get_info`, then `list_actors` or `find_actors` to understand current state.
Design the complete `operations[]` array before touching anything.
4. Issue independent discovery reads in parallel.
5. BATCH: call `level_compose` once with the full spec.
- Use `stop_on_error: true` if any op references `$ops[...]` from another op in the batch
- Use `stop_on_error: false` for independent bulk modifications
- Pass `expected_fingerprint` on each mutation guarded by `prefetched_state`
6. VERIFY: check `completed_steps == total_steps`.
If false, diff `spawned_actors` against plan, call `find_actors` if needed,
then construct a MINIMAL fix batch for the gap only.
Maximum one retry batch. If retry fails, stop and report.
TOOL SELECTION RULE:
- level_compose: 2+ spawns OR 3+ existing actor modifications OR any spawn+configure chain
- Individual tools: 1-2 existing actors with a single change (quick corrections only)
PROHIBITED: Do NOT skip the Plan phase for multi-actor work. Do NOT call get_actor
for each spawned actor to verify — check completed_steps first.
The Level Designer agent will:
.cortex/domains/level.mdoperations[] speclevel_compose onceThe agent returns from level_compose:
success: true/falseactor_count + spawned_actors: actors created or duplicatedcompleted_steps / total_steps: batch progressfailed_steps[]: per-failure with op_id, command, error_code, errorUse the Task tool with subagent_type: "cortex-toolkit:level-designer" and max_turns: 15 to delegate the review.
Pass context about what to review:
Review the current level and provide a report:
**Prefetched state:** [embed the main-thread `prefetched_state` block here before launching]
1. Use `prefetched_state` first; only fetch missing data
2. Use `get_info` for level overview (name, actor count, world type, sublevels, is_world_partition)
3. Use `list_actors` to enumerate actors (paginate if needed)
4. Run independent reads in parallel where possible
5. Use `get_bounds` to understand spatial layout
6. Check folder organization — are actors organized logically?
7. Check for common issues:
- Actors without folders
- Duplicate labels
- Actors at origin that shouldn't be
- Unloaded sublevels
- Missing data layer assignments (if World Partition)
8. Summarize findings with recommendations
The Level Designer agent will:
The agent returns:
If the agent's response includes a Status line:
If the agent's response has no Status line (e.g., turn limit reached mid-response), treat as partial — summarize whatever the agent produced and note the work may be incomplete.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub etelyatn/cortex-toolkit --plugin cortex-toolkit