version: "1.0.0" name: configuration description: Use when a Project Agent reads or changes lore indexes and group memberships, managed presets, game modules, automations, Skills, or Agent profiles through config_read and config_apply. category: configuration agent: general,ide

Configuration

Manage every supported configuration resource through the stable config_read and config_apply tools. Never edit Denova configuration files or internal JSON storage directly.

Required workflow

1. For an unfamiliar resource, call config_read with operation=describe. Omit resource to discover all registered kinds, or provide one kind for its contract.
2. Call config_read(operation=list) to resolve the exact ID and scope. Catalogs use items plus next_cursor; continue with the identical request and returned cursor until truncated=false when the full catalog matters. Before an update or delete, call config_read(operation=get) for the exact ID and retain its latest revision.
3. Read exactly the relevant reference below with read({"path":"skill://configuration/references/<file>.md"}). Do not infer a value shape from another resource.
4. Follow that resource's mutation semantics. Call config_apply for exactly one create, update, or delete and use only names returned by describe.
5. For update and delete, copy the latest revision from config_read; never guess, copy an example placeholder, or reuse a stale revision. Preserve every field or layer section the user did not ask to change.

get accepts any number of IDs. A mixed batch succeeds with existing items, explicit missing_ids, and per-ID failures; an entirely unsuccessful completed batch fails. Large exact reads also return next_cursor. Never treat truncated=true as a complete read.

1. config_apply returns only a compact persistence receipt (resource, operation, id, revision). Read the changed item again and compare the requested fields plus preservation-sensitive fields. Report success only when the effective result matches.

Deletion must be explicitly requested by the user. If a stale-revision conflict occurs, read the current item again, reconcile the requested change with that value, and retry once with the new revision. Do not silently overwrite concurrent changes.

Mutation semantics

Fields returned for inspection such as path, custom, builtin_overridden, invalid, error, timestamps, resolved snapshots, run history, and secrets are host-owned unless the resource reference explicitly says otherwise. Never add an unknown field: configuration values reject unknown keys.

Examples use REVISION_FROM_GET as a visible placeholder. Always replace it with the exact current revision returned for the same resource, ID, and scope.

Reference routing

• style_reference → skill://configuration/references/style-reference.md
• narrative_style → skill://configuration/references/narrative-style.md
• game_planning → skill://configuration/references/game-planning.md
• event_package → skill://configuration/references/event-package.md
• rule_system → skill://configuration/references/rule-system.md
• state_system → skill://configuration/references/state-system.md
• image_preset → skill://configuration/references/image-preset.md
• automation → skill://configuration/references/automation.md
• skill → skill://configuration/references/skill.md
• agent_profile → skill://configuration/references/agent-profile.md
• lore_index, lore_index_membership → skill://configuration/references/lore-index.md

For lore index organization, use these configuration resources. For creating or editing lore bodies, use query_lore_items with detail=full and write_lore_items when available. Do not pass index fields to write_lore_items or edit the backing JSON directly. Index management applies to the current project in both writing and game configuration pages; it does not enable configuration mutations during game play.

These are references, not sub-Skills. Do not pass a sub_skill argument or invoke another Skill name.