@battlegrid/mcp-server 31.2.39 → 31.2.41

package/README.md

@@ -24,6 +24,57 @@ Seeing package `31.x` alongside handshake `battlegrid@33.x` — the package **be
 
 **What this changes for you:** nothing about how you call anything. Upgrading the package no longer waits on a server deploy, and a server deploy no longer strands you on a package that names the wrong contract — reconnect and the announcement follows. **Contract breaking-change notes are no longer keyed to package versions**, since a contract move is no longer a release here; the v11-and-earlier notes below are kept as history, and the live vocabulary is always discovery.
 
+## Contract history — v64.1
+
+Purely additive in schema, with **one refusal behind an unchanged one.** An agent's unsaved
+configuration now has a server home — the same owner-scoped *draft* strategies gained in v63.1 — so
+the agent form, Telegram and a connected client all see one set of unsaved values. These tools are
+how a conversation reaches it, and the two agent committers now step aside while it is open.
+
+### Wider surface — four tools added
+
+- **`stage_agent_draft({ request: { agentId?, axes } })`** writes proposed axes into the player's
+  draft for an agent and commits **nothing**: the values become the agent's configuration only when
+  the player saves. Unlike `stage_strategy_plan` it takes the axis *values* — `IDENTITY`,
+  `BEHAVIOR`, `MODEL`, `TRADING_CONFIG`, and for an agent not yet created `STRATEGY_BINDING` — each
+  written **whole**, so `BEHAVIOR` carries all three of risk, outlook and conviction and
+  `TRADING_CONFIG` the complete agent-owned configuration. Axes you omit keep their values.
+  Structure is checked on the way in; ranges are checked only when the player saves.
+
+- **Omit `agentId` to open a create draft.** The server mints the id and the response carries it;
+  name that id on every later call so one draft accumulates rather than a second opening beside it.
+
+- **`get_agent_draft({ request: { agentId } })`** reads one draft, or answers `{ draft: null }` —
+  not part-way through that agent is a value, not an error. `baseRevision` is `null` for an agent not
+  yet created, and `baseMoved` reports an agent committed past the revision the draft was written
+  against.
+
+- **`list_agent_drafts({ request: { cursor? } })`** answers "what am I part-way through?" for a
+  conversation holding no agent id, newest first. Offer to continue one of these before starting a
+  second.
+
+- **`discard_agent_draft({ request: { agentId, confirm } })`** destroys one draft on the player's
+  word. Called with `confirm: false` it destroys nothing and is refused with when the draft was last
+  written and which surface wrote it — tell the player that, then ask.
+
+### Refusals worth knowing before you stage or save
+
+- **A staged proposal refuses the axes the player typed after your call read the draft.** The
+  refusal names the axes; read the draft again and propose against what they now have. Another
+  surface's write to a *different* axis is retried once for you, then reported as a conflict.
+
+- **`STRATEGY_BINDING` is refused on an agent that exists.** Rebinding replaces an agent's
+  configuration and stays its own confirmed call, `rebind_intelligence_agent`.
+
+- **`update_intelligence_agent` and `rebind_intelligence_agent` answer `CONFLICT` while the player
+  holds a draft for that agent**, naming staging as the act available. Nothing about their input or
+  output changed, and a caller whose player holds no draft is unaffected — this is the one change
+  existing code can observe.
+
+### Vocabulary
+
+`toolCount` 122 → 126. No input or output schema of an existing tool moves.
+
 ## Contract history — v64
 
 **Breaking, and it is a removal you will feel on two tools.** A vocabulary that never depended on

package/dist/index.d.ts

@@ -49,7 +49,7 @@ import { type Implementation, type Prompt, type Resource } from '@modelcontextpr
  * being asked. Move it for a change to THIS package — a proxy fix, a dependency bump, a docs
  * correction. Never move it to track the server.
  */
-export declare const PACKAGE_VERSION = "31.2.39";
+export declare const PACKAGE_VERSION = "31.2.41";
 export declare const DEFAULT_URL = "https://mcp.battlegrid.trade/mcp";
 export interface EnvConfig {
     apiKeys: string[];

package/dist/index.js

@@ -52,7 +52,7 @@ import { ListToolsRequestSchema, CallToolRequestSchema, ListPromptsRequestSchema
  * being asked. Move it for a change to THIS package — a proxy fix, a dependency bump, a docs
  * correction. Never move it to track the server.
  */
-export const PACKAGE_VERSION = '31.2.39';
+export const PACKAGE_VERSION = '31.2.41';
 export const DEFAULT_URL = 'https://mcp.battlegrid.trade/mcp';
 const MAX_RETRIES = 3;
 const RETRY_DELAYS_MS = [2000, 4000, 8000];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@battlegrid/mcp-server",
-  "version": "31.2.39",
+  "version": "31.2.41",
   "description": "BattleGrid MCP server — play crypto prediction games from AI agents",
   "type": "module",
   "main": "dist/index.js",

package/skills/EXPORT.json

@@ -1,8 +1,8 @@
 {
   "generator": "battlegrid-app/server/scripts/export-mcp-skills.mjs",
-  "contractVersion": "64.0.0",
+  "contractVersion": "64.1.0",
   "files": {
-    "battlegrid-agent-management/SKILL.md": "d56834fa2016ec9371994999fa13a3dbb69906efc54db522e10fd2eb0ebb6044",
+    "battlegrid-agent-management/SKILL.md": "f43f3ac852e032c320d43f599d3e683203c68daf42637a9dbf292b380f17c635",
     "battlegrid-arena-play/SKILL.md": "4cfaa92a8d0dcaf477b227abd774e7aa8642c185d72108dfd401be0cf8aecd7f",
     "battlegrid-market-analysis/SKILL.md": "22fde9c4eac0261c89a8056ab8b20b66fcbe1fd58447fa0ac84fd410b4d134d7",
     "battlegrid-radar-deployment/SKILL.md": "d2a012b154e1e529db19529bd7c3d75d069f4e410d5804add51165f0f0902b2e",

package/skills/battlegrid-agent-management/SKILL.md

@@ -21,10 +21,12 @@ step named beside it before doing anything else.
 2. **Acting without blast radius.** *Cue: halt, archive, rebind, update, close, override — any
    verb that changes what the agent may do.* Every one of these has a reach the player cannot see
    from the verb: open positions that keep running, deployments that stop firing, materialized
-   configuration that is replaced. State the reach from the reads BEFORE the confirm form. → step 3.
+   configuration that is replaced. State the reach from the reads BEFORE the confirm form. And a
+   configuration change to an agent the player is mid-edit on is staged, not committed. → steps 3
+   and 4.
 3. **Mis-lever'd halt recovery.** *Cue: "why is it stopped", "start it again", any halted agent.*
    There are three halt reasons and they do not share a lever. Offering a drawdown baseline reset
-   to a daily-loss halt is offering something that cannot work. → step 4.
+   to a daily-loss halt is offering something that cannot work. → step 5.
 4. **Driving strategy writes from this arc.** *Cue: the answer to an agent problem turns out to be
    "change the strategy".* You can SEE strategies here because you must name what you bind. That is
    not permission to author them. → the cross-skill rule below.
@@ -54,6 +56,9 @@ discipline governs the change, not about what the server would accept.)
   never its `id`. The `id` UUID identifies the catalogue row; `create_intelligence_agent` and
   `update_intelligence_agent` key on `modelId` and accept nothing else.
 - `list_strategies` — the binding candidates.
+- `list_agent_drafts` — what the player is already part-way through. **Mention an existing draft
+  before starting new work** — "you have an unsaved <agent / new agent>, last touched from
+  <surface> at <time>" — and offer to continue it rather than opening a second one beside it.
 
 Read each thing once. A roster or agent you already fetched in this conversation is still in front
 of you; re-fetching costs the player the same payload twice and it rides every later step.
@@ -73,8 +78,16 @@ When the picks come back, restate in one line: *"Commissioning: <name>, bound to
 budget posture** — the capital ceiling and stops the trading configuration will carry, or that it
 will take platform defaults.
 
-Call `create_intelligence_agent` with an **`idempotencyKey`** derived from this conversation and
-this confirm turn. A create spends an agent slot against the player's rank quota; a retry after a
+**Build it in a create draft.** As the picks come back, stage them with `stage_agent_draft` —
+omit `agentId` on the first call and reuse the id it returns — each axis WHOLE: IDENTITY
+`{ displayName }`, BEHAVIOR `{ behavior: { risk, outlook, conviction } }`, MODEL `{ modelId }`,
+TRADING_CONFIG with every agent-owned field, and STRATEGY_BINDING `{ strategyId }`. The build then
+survives a refresh, lists in the Agents Hub as *not yet created*, and opens on the create screen,
+where **the player's Save creates it**. Staging creates nothing and spends no slot.
+
+Call `create_intelligence_agent` yourself only when the player asks you to create it from this
+conversation — on a surface with no create screen to save from — with an **`idempotencyKey`**
+derived from this conversation and this confirm turn. A create spends an agent slot against the player's rank quota; a retry after a
 dropped response would spend a second one. With the key, an ambiguous retry replays the original
 result instead.
 
@@ -86,7 +99,24 @@ Profile → Wallet tab path instead, because no control renders there. **Create
 way**: the consent gates entering a game, never commissioning one. A `true` flag is consent alone
 and not readiness — the pipeline's own refusal at the fee is the authority.
 
-### 3. Lifecycle verbs: read first, state the radius, then confirm
+### 3. Stage, review, commit — never commit around an open draft
+
+Before proposing any change to an existing agent, `get_agent_draft`. **A draft means the player is
+mid-edit in their form.** Then:
+
+- **Stage** the change with `stage_agent_draft`: it lands in their open form, labelled as yours, and
+  **their Save commits it** together with their own edits. `update_intelligence_agent` and
+  `rebind_intelligence_agent` are REFUSED while the draft exists — that refusal is the server's, so
+  do not retry them.
+- A staging call **refused naming an axis** means the player changed it after you read the draft:
+  read again and propose against what they now have — never re-send the same values.
+- **Rebinding is never staged.** While a draft is open, ask the player to save or discard it first.
+- `discard_agent_draft` only on the player's word: call it with `confirm:false` first, tell them
+  when the draft was last written and from which surface, ask, then confirm.
+
+With **no draft open**, the update verb below stands, confirm and all.
+
+### 4. Lifecycle verbs: read first, state the radius, then confirm
 
 Every one of update, rebind, halt, resume, activate and archive runs this shape. The reads
 (`get_agent_budget`, `get_agent_fund_allocation`, `get_agent_open_positions`,
@@ -118,7 +148,7 @@ the reason, and name what would clear it (un-deploy via the radar/deployment too
 the positions resolve, wait for the session to settle). Never paraphrase the refusal into "it
 didn't work", and never retry it unchanged.
 
-### 4. Halt recovery: branch on the served halt reason
+### 5. Halt recovery: branch on the served halt reason
 
 `get_agent_budget` serves `haltReason`. There are exactly three, and the lever differs:
 
@@ -136,7 +166,7 @@ figure against the limit. Surface those served figures and the applicable lever
 above. Do not retry the resume, and do not reach for the reset to "get past" a refusal that names
 a different stop.
 
-### 5. Risk limits are a whole object
+### 6. Risk limits are a whole object
 
 `update_intelligence_agent`'s `tradingConfig` is a **complete** configuration: what you send
 replaces what is stored, and every field in that schema is required when the object is present.
@@ -146,7 +176,7 @@ name **every changed value** in the confirm — from what, to what. Never assemb
 and never echo a read config back unchanged: the read shape is wider than the write shape
 (`strategyTimeframe` and `regimeTimeframe` are strategy-derived and rejected as unknown keys).
 
-### 6. Live positions: present the served state, name the bypass
+### 7. Live positions: present the served state, name the bypass
 
 - `get_agent_open_positions` / `list_user_active_positions` / `get_position_audit_history` first —
   the `decisionId` these two tools need is discoverable only through those reads.