Update to 5.5e SRD Oracle

{
  "description": "Explore and use this project's redistributable 5.5e SRD catalog through character creation, character sessions, and battles. Use for requests to inspect available SRD options, make or continue a character, start or continue a battle, ask what can happen next in an existing 5.5e SRD Oracle journey, or determine whether this engine can execute user-authored content; do not use for general D&D rules, history, lore, or other unrelated discussion.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 203
    }
  ],
  "name": "dnd-srd-oracle",
  "skill_md_contents": "---\nname: dnd-srd-oracle\ndescription: Explore and use this project's redistributable 5.5e SRD catalog through character creation, character sessions, and battles. Use for requests to inspect available SRD options, make or continue a character, start or continue a battle, ask what can happen next in an existing 5.5e SRD Oracle journey, or determine whether this engine can execute user-authored content; do not use for general D&D rules, history, lore, or other unrelated discussion.\n---\n\n# Consult 5.5e SRD Oracle\n\nEstablish the user's immediate goal. Use the relevant discovery or read operation before making a stateful choice unless the current MCP result already supplies the required facts.\n\nRead `describe_mcp_workflow` for this connection's tool names and argument patterns; reuse a guide already returned by the same connection. Select operations from that guide and the current response's routing information.\n\nDistinguish browsing installed catalog presence from asking which choices are currently legal for a character. Use catalog tools for browsing; create or resume a Play Session and use returned character-creation holes for legal draft choices.\n\nCreate a Play Session for a new journey. Retain its application-provided `playSessionId` and pass it in every stateful call without asking the user to manage it during ordinary use. Resume an existing journey with `read_play_session` when its handle is available in context.\n\nPublic Play Sessions require standard MCP OAuth and are saved from creation.\nLet the tool's OAuth challenge start account linking; never request a password,\nAPI key, token, or unrelated permission in conversation. Local stdio Play\nSessions need no sign-in and last only for that server process. Use\n`list_saved_play_sessions` for explicit public recovery and never silently\nselect a saved session. Treat `delete_saved_play_session` as immediate and\npermanent.\n\nFor a character-creation continuation, use `read_play_session` first. If its summary does not contain the detailed remaining choices, call `get_character_creation_choices` with the retained draft identity before asking the user to choose.\n\nTreat returned catalog records, character-creation holes and options, battle acts, runtime holes, projections, and operation results as authoritative. Copy their identifiers and typed inputs exactly. Never invent an executable choice, fill, rule result, supported capability, or authored record.\n\nFor Battle setup, follow the returned `battleState` variant. If the result is\n`initialInitiativeSetup`, use the exposed Initiative-swap and setup-finalization\noperations before asking for or presenting active Battle Acts. For an active\nBattle, use the exposed add/remove-combatant operations for supported roster\nchanges and keep the returned occupancy and settlement projections authoritative.\n\nFor Battle Acts, use `chatGptBattleOperations` when the response supplies it to\nselect the named operation for the current subject or Reaction. Copy the current\nsubject into its named Act tool, without a fill, to select it and open any holes.\nAnswer ordinary holes through `provide_battle_input` with a fill and no subject.\nSelect each Reaction through its named operation with an `interruptDecision`\nfill, no subject, and `choice.fills: []`. The pending transaction owns the parent.\n\n`roll_dice` is an optional independent raw-face request. It has no Battle Hole\nor modifier context and does not fill anything automatically. Copy its ordered\nfaces into an ordinary typed fill only when the current Runtime Hole requests\nthat shape; otherwise ask for the missing user/table fact.\n\nPresent relevant returned rules facts faithfully. Ask only for unresolved user decisions or requested rolls. When the current result contains multiple independent meaningful choices, present them together and let the user answer them in one reply; do not serialize independent current holes into one question per turn. Keep dependent choices for a later batch when their options have actually been returned. Apply forced single-option choices automatically and report them. Synthetic names, situations, and narration may frame play, but must not create mechanics or reproduce non-SRD official content.\n\nAfter every operation, report the returned result and relevant state. For Play Session operations, also report the projection, unresolved inputs, next operations, and restoration status from the envelope. Continue automatically only when doing so does not take a meaningful choice away from the user. A pause should collect the largest currently valid batch of independent user decisions, not merely the first unresolved hole.\n\nWhen a Play Session is unavailable, explain that the service cannot access the handle, follow the returned new-session restoration guidance, and never claim why it is absent or silently replace it.\n\nIf asked to import or execute user-authored records, state that this MCP uses only its installed redistributable SRD catalogs. Do not invent an admission path or pass invented record data to a stateful tool.\n"
}