@adhd/agent-plugin-budget 0.0.7 → 0.2.0

package/CHANGELOG.md

@@ -1,3 +1,43 @@
+## 0.2.0 (2026-08-11)
+
+
+### 🚀 Features
+
+- **agent-plugin-budget:** rename tokens cap to context with peak semantics + contextWindowFraction (Packet B, BUG-AGENTMCP-009)
+
+- **agent-plugin-budget:** add errors/consecutiveErrors caps + reject tool-scoped context (Packet C, plan §8.9)
+
+
+### 🩹 Fixes
+
+- **apigen-cli:** restore 2768 files mass-deleted by 0117eb22 (BUG-APIGEN-052)
+
+- **agent-plugin-budget:** cache-weighted cost enforcement (BUG-AGENTMCP-008)
+
+- **agent-plugin-budget:** honor cap.mode on the model path with budget:warning/budget:block events (Packet A)
+
+- **agent-plugin-budget:** emit budget:block on tool path; add typecheck targets (Packet A review F1/F5)
+
+- **agent-plugin-budget:** reject windowed caps at task scope (F2 fold-in — silent no-op class, plan §8.9-precedent)
+
+
+### ❤️  Thank You
+
+- parity-harness-self-test
+- pseudosky
+
+## 0.0.7 (2026-07-25)
+
+
+### 🔥 Performance
+
+- **test:** bound vitest thread pools to curb CPU oversubscription (DEBT-TEST-CPU-OVERSUBSCRIBED-001)
+
+
+### ❤️  Thank You
+
+- pseudosky
+
 ## 0.0.6 (2026-07-24)
 
 This was a version bump only for agent-plugin-budget to align it with other projects, there were no code changes.

package/README.md

@@ -165,12 +165,16 @@ For the `cost` field, set token prices on the defaults dimension:
   "defaults": {
     "costPerInputToken": 0.000003,
     "costPerOutputToken": 0.000015,
+    "costPerCacheReadToken": 0.0000003,
+    "costPerCacheWriteToken": 0.00000375,
     "caps": [{ "field": "cost", "maximum": 0.1 }]
   }
 }
 ```
 
-Cost = `inputTokens × costPerInputToken + outputTokens × costPerOutputToken`.
+Cost = `uncachedInputTokens × costPerInputToken + cacheReadTokens × costPerCacheReadToken + cacheCreationTokens × costPerCacheWriteToken + outputTokens × costPerOutputToken`.
+
+`costPerCacheReadToken` and `costPerCacheWriteToken` default to `costPerInputToken` when unset, so a config that only sets the two flat rates bills every input token at the flat input rate — identical to the pre-cache-aware behavior. To match a model's real rate card (e.g. DeepSeek: input `$0.14/M`, cache-read `$0.0028/M`), copy the per-class rates from the provider rate card; on cache-warm runs most input is cache-read, and billing it at the flat input rate over-counts spend by up to ~50x.
 
 ## Backward compat (flat format)
 

package/index.cjs

@@ -1,26 +1,29 @@
-"use strict";Object.defineProperties(exports,{__esModule:{value:!0},[Symbol.toStringTag]:{value:"Module"}});const p=require("zod");function E(f){const e=/^P(?:(\d+)Y)?(?:(\d+)M)?(?:(\d+)D)?(?:T(?:(\d+)H)?(?:(\d+)M)?(?:(\d+(?:\.\d+)?)S)?)?$/,t=f.match(e);if(!t)throw new Error(`invalid ISO 8601 duration: ${f}`);const[,o,s,i,n,a,c]=t;let r=0;return o&&(r+=parseInt(o)*365.25*864e5),s&&(r+=parseInt(s)*30.44*864e5),i&&(r+=parseInt(i)*864e5),n&&(r+=parseInt(n)*36e5),a&&(r+=parseInt(a)*6e4),c&&(r+=parseFloat(c)*1e3),Math.round(r)}const w=["tokens","inputTokens","outputTokens","calls","wallClock","modelMs","cost","toolCalls","responseSize"],b=p.z.object({field:p.z.enum(w),maximum:p.z.number().min(0),window:p.z.string().optional(),scope:p.z.enum(["task","session","agent","global"]).optional(),mode:p.z.enum(["warning","block"]).optional(),message:p.z.string().optional()}),g=p.z.object({caps:p.z.array(b).optional(),mode:p.z.enum(["warning","block"]).optional(),costPerInputToken:p.z.number().min(0).optional(),costPerOutputToken:p.z.number().min(0).optional(),scope:p.z.enum(["task","session","agent","global"]).optional()}),T=p.z.object({defaults:g.optional(),agent:p.z.object({default:g.optional(),overrides:p.z.record(p.z.string(),g.partial()).optional().default({})}).optional(),provider:p.z.object({default:g.optional(),overrides:p.z.record(p.z.string(),g.partial()).optional().default({})}).optional(),tool:p.z.object({default:g.optional(),overrides:p.z.record(p.z.string(),g.partial()).optional().default({})}).optional()}),x=p.z.object({}).passthrough(),_={maxInputTokens:{field:"inputTokens"},maxOutputTokens:{field:"outputTokens"},maxTotalTokens:{field:"tokens"},maxModelCalls:{field:"calls"},maxWallClockMs:{field:"wallClock"},maxModelMs:{field:"modelMs"},maxCostUSD:{field:"cost"},maxTokensPer24h:{field:"tokens",window:"PT24H"},maxCalls:{field:"toolCalls"}};function v(f){const e=[],t={};for(const[o,s]of Object.entries(f)){const i=_[o];if(i&&typeof s=="number"){const n={field:i.field,maximum:s};i.window&&(n.window=i.window),e.push(n)}else(o==="scope"||o==="mode"||o==="costPerInputToken"||o==="costPerOutputToken"||o==="message")&&(t[o]=s)}return e.length>0&&(t.caps=e),g.parse(t)}function O(f){const e=f;if(e.defaults!==void 0||e.agent!==void 0||e.provider!==void 0||e.tool!==void 0){const t=T.parse(f);return{defaults:t.defaults??g.parse({}),agent:t.agent??{overrides:{}},provider:t.provider??{overrides:{}},tool:t.tool??{overrides:{}}}}return{defaults:v(e),agent:{overrides:{}},provider:{overrides:{}},tool:{overrides:{}}}}function S(f,e,t,o){return{isEnforcementError:!0,code:"BUDGET_EXCEEDED",message:o??`${f} limit is ${e}, current value is ${Math.round(t)}`}}function A(f,e,t){return{isToolWarning:!0,toolName:f,callId:e,message:t}}class y{constructor(e,t,o=0,s=0){this.db=e,this.cfg=t,this.costPerInput=o,this.costPerOutput=s,this.name="agent-mcp-budget",this.accumulators=new Map}install(e){e.register("task:start",t=>{try{this.onTaskStart(t)}catch{}}),e.register("pre:model_request",t=>{try{this.onPreModelRequest(t)}catch{}}),e.register("post:model_response",t=>{try{this.onPostModelResponse(t)}catch{}}),e.register("task:completed",t=>{try{this.onTerminal(t.executionContext.taskId)}catch{}}),e.register("task:failed",t=>{try{this.onTerminal(t.executionContext.taskId)}catch{}}),e.register("task:cancelled",t=>{try{this.onTerminal(t.executionContext.taskId)}catch{}}),e.registerEnforcement("pre:model_request",t=>this.enforcePreModel(t)),e.registerEnforcement("pre:tool_call",t=>this.enforcePreTool(t)),e.register("transform:tool_result",t=>{try{this.enforceResponseSize(t)}catch{}})}onTaskStart(e){var n,a;const{taskId:t,sessionId:o,agentName:s}=e.executionContext,i=((a=(n=e.executionContext.agentDefinition)==null?void 0:n.provider)==null?void 0:a.type)??"unknown";this.accumulators.set(t,{taskId:t,sessionId:o??void 0,agentName:s,providerType:i,startedAtMs:Date.now(),inputTokens:0,outputTokens:0,modelCalls:0,totalModelMs:0,toolCalls:new Map})}onPreModelRequest(e){const t=this.accumulators.get(e.executionContext.taskId);t&&(t.modelCallStartMs=Date.now())}onPostModelResponse(e){const t=this.accumulators.get(e.executionContext.taskId);if(!t)return;const o=e.tokenUsage;o&&(t.inputTokens+=o.inputTokens??0,t.outputTokens+=o.outputTokens??0),t.modelCalls+=1,t.modelCallStartMs!==void 0&&(t.totalModelMs+=Date.now()-t.modelCallStartMs,t.modelCallStartMs=void 0)}onTerminal(e){this.accumulators.delete(e)}mergeDim(e){let t={caps:[]};for(const o of e)o&&(t={caps:[...t.caps??[],...o.caps??[]],mode:o.mode??t.mode,costPerInputToken:o.costPerInputToken??t.costPerInputToken,costPerOutputToken:o.costPerOutputToken??t.costPerOutputToken,scope:o.scope??t.scope});return t}resolveCaps(e,t,o){var u,m,d;const s=this.cfg.defaults,i=this.cfg.agent,n=this.cfg.provider,a=this.cfg.tool;if(o){const k=(u=a==null?void 0:a.overrides)==null?void 0:u[o],h=this.mergeDim([s,a==null?void 0:a.default,k]);return{caps:h.caps??[],mode:h.mode,scope:h.scope}}const c=(m=i==null?void 0:i.overrides)==null?void 0:m[e],r=(d=n==null?void 0:n.overrides)==null?void 0:d[t],l=this.mergeDim([s,i==null?void 0:i.default,c,n==null?void 0:n.default,r]);return{caps:l.caps??[],mode:l.mode,scope:l.scope}}queryScopeTotals(e,t,o,s){const i=this.accumulators.get(e),n=i?{inputTokens:i.inputTokens,outputTokens:i.outputTokens,modelCalls:i.modelCalls}:{inputTokens:0,outputTokens:0,modelCalls:0};if(s==="task"||!this.db)return n;try{const a=this.db;let c;if(s==="session"&&t?c=a.prepare(`SELECT
+"use strict";Object.defineProperties(exports,{__esModule:{value:!0},[Symbol.toStringTag]:{value:"Module"}});const u=require("zod"),M=require("@adhd/agent-base-types");function b(c){const t=/^P(?:(\d+)Y)?(?:(\d+)M)?(?:(\d+)D)?(?:T(?:(\d+)H)?(?:(\d+)M)?(?:(\d+(?:\.\d+)?)S)?)?$/,e=c.match(t);if(!e)throw new Error(`invalid ISO 8601 duration: ${c}`);const[,o,s,n,i,r,a]=e;let l=0;return o&&(l+=parseInt(o)*365.25*864e5),s&&(l+=parseInt(s)*30.44*864e5),n&&(l+=parseInt(n)*864e5),i&&(l+=parseInt(i)*36e5),r&&(l+=parseInt(r)*6e4),a&&(l+=parseFloat(a)*1e3),Math.round(l)}const y=["context","inputTokens","outputTokens","calls","wallClock","modelMs","cost","toolCalls","errors","consecutiveErrors","responseSize"],_=new Set(["toolCalls","errors","consecutiveErrors"]),A=u.z.object({field:u.z.enum(y),maximum:u.z.number().min(0).optional(),contextWindowFraction:u.z.number().min(0).max(1).optional(),window:u.z.string().optional(),scope:u.z.enum(["task","session","agent","global"]).optional(),mode:u.z.enum(["warning","block"]).optional(),message:u.z.string().optional()}).superRefine((c,t)=>{if(c.field==="context"){const e=c.maximum!==void 0,o=c.contextWindowFraction!==void 0;e===o&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["contextWindowFraction"],message:"cap field 'context' requires exactly one of 'maximum' or 'contextWindowFraction'"}),c.window!==void 0&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["window"],message:"cap field 'context' rejects 'window' — a windowed peak is meaningless; windowed cumulative volume is expressible via 'inputTokens'/'outputTokens' + 'window'"})}else(c.field==="errors"||c.field==="consecutiveErrors")&&(c.window!==void 0&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["window"],message:`cap field '${c.field}' rejects 'window' — windowed error budgets are not expressible this wave`}),c.scope!==void 0&&c.scope!=="task"&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["scope"],message:`cap field '${c.field}' is counted per-task in memory (no task_usage column) — only 'task' scope is expressible this wave`})),c.window!==void 0&&(c.scope===void 0||c.scope==="task")&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["scope"],message:`cap field '${c.field}' carries 'window' but resolves to 'task' scope, where windowed caps are a silent no-op (no task-level window query); set an explicit 'scope': 'session' | 'agent' | 'global'`}),c.contextWindowFraction!==void 0&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["contextWindowFraction"],message:`'contextWindowFraction' is only valid on cap field 'context' (got '${c.field}')`}),c.maximum===void 0&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["maximum"],message:`cap field '${c.field}' requires a 'maximum'`})}),g=u.z.object({caps:u.z.array(A).optional(),mode:u.z.enum(["warning","block"]).optional(),costPerInputToken:u.z.number().min(0).optional(),costPerOutputToken:u.z.number().min(0).optional(),costPerCacheReadToken:u.z.number().min(0).optional(),costPerCacheWriteToken:u.z.number().min(0).optional(),scope:u.z.enum(["task","session","agent","global"]).optional()}),x=g.partial().superRefine((c,t)=>{const e=c.caps??[];for(const[o,s]of e.entries())s.field==="context"&&t.addIssue({code:u.z.ZodIssueCode.custom,path:["caps",o,"field"],message:"cap field 'context' is only valid at model scope — a tool-scoped 'context' cap is a silent no-op (enforced on the model path, invisible to tool overrides); place it in 'defaults'/'agent'/'provider' instead"})}),S=u.z.object({defaults:g.optional(),agent:u.z.object({default:g.optional(),overrides:u.z.record(u.z.string(),g.partial()).optional().default({})}).optional(),provider:u.z.object({default:g.optional(),overrides:u.z.record(u.z.string(),g.partial()).optional().default({})}).optional(),tool:u.z.object({default:x.optional(),overrides:u.z.record(u.z.string(),x).optional().default({})}).optional()}),O=u.z.object({}).passthrough(),P={maxInputTokens:{field:"inputTokens"},maxOutputTokens:{field:"outputTokens"},maxModelCalls:{field:"calls"},maxWallClockMs:{field:"wallClock"},maxModelMs:{field:"modelMs"},maxCostUSD:{field:"cost"},maxTokensPer24h:{field:"inputTokens",window:"PT24H"},maxCalls:{field:"toolCalls"}};function I(c){const t=[],e={};for(const[n,i]of Object.entries(c)){const r=P[n];if(r&&typeof i=="number"){const a={field:r.field,maximum:i};r.window&&(a.window=r.window),t.push(a)}else(n==="scope"||n==="mode"||n==="costPerInputToken"||n==="costPerOutputToken"||n==="costPerCacheReadToken"||n==="costPerCacheWriteToken"||n==="message")&&(e[n]=i)}t.length>0&&(e.caps=t);const o=e.scope,s=o==="task"||o==="session"||o==="agent"||o==="global"?o:void 0;for(const n of t)n.window!==void 0&&n.scope===void 0&&(n.scope=s??"global");return g.parse(e)}const w=c=>`legacy 'tokens' budget cap detected (${c}): cap field 'tokens' is removed; use 'context' (peak request input) or 'inputTokens'/'outputTokens' (windowed volume)`;function z(c){const t=c??{};if(typeof t.maxTotalTokens=="number")throw new Error(w("flat 'maxTotalTokens'"));const e=s=>{if(typeof s!="object"||s===null)return;const n=s;if(typeof n.maxTotalTokens=="number")throw new Error(w("structured 'maxTotalTokens'"));const i=n.caps;if(Array.isArray(i)){for(const r of i)if(typeof r=="object"&&r!==null&&r.field==="tokens")throw new Error(w("caps[].field === 'tokens'"))}},o=s=>{if(typeof s!="object"||s===null)return;const n=s;e(n),"default"in n&&e(n.default);const i=n.overrides;if(typeof i=="object"&&i!==null)for(const r of Object.values(i))e(r)};e(t.defaults),o(t.agent),o(t.provider),o(t.tool)}function $(c){z(c);const t=c;if(t.defaults!==void 0||t.agent!==void 0||t.provider!==void 0||t.tool!==void 0){const e=S.parse(c);return{defaults:e.defaults??g.parse({}),agent:e.agent??{overrides:{}},provider:e.provider??{overrides:{}},tool:e.tool??{overrides:{}}}}return{defaults:I(t),agent:{overrides:{}},provider:{overrides:{}},tool:{overrides:{}}}}function v(c,t,e,o){return{isEnforcementError:!0,code:"BUDGET_EXCEEDED",message:o??`${c} limit is ${t}, current value is ${Math.round(e)}`}}function R(c,t,e){return{isToolWarning:!0,toolName:c,callId:t,message:e}}function L(c,t){var o;let e=0;for(const s of c){e+=((o=s.content)==null?void 0:o.length)??0;for(const n of s.toolCalls??[])e+=JSON.stringify(n.arguments??{}).length;for(const n of s.toolResults??[])e+=JSON.stringify(n.result??null).length}for(const s of t)e+=JSON.stringify({name:s.name,description:s.description,inputSchema:s.inputSchema}).length;return Math.ceil(e/4)}class W{constructor(t,e,o=0,s=0,n=0,i=0){this.db=t,this.cfg=e,this.costPerInput=o,this.costPerOutput=s,this.costPerCacheRead=n,this.costPerCacheWrite=i,this.name="agent-mcp-budget",this.accumulators=new Map}install(t){this.hooks=t,t.register("task:start",e=>{try{this.onTaskStart(e)}catch{}}),t.register("pre:model_request",e=>{try{this.onPreModelRequest(e)}catch{}}),t.register("post:model_response",e=>{try{this.onPostModelResponse(e)}catch{}}),t.register("post:tool_call",e=>{try{this.onPostToolCall(e)}catch{}}),t.register("task:completed",e=>{try{this.onTerminal(e.executionContext.taskId)}catch{}}),t.register("task:failed",e=>{try{this.onTerminal(e.executionContext.taskId)}catch{}}),t.register("task:cancelled",e=>{try{this.onTerminal(e.executionContext.taskId)}catch{}}),t.registerEnforcement("pre:model_request",e=>this.enforcePreModel(e)),t.registerEnforcement("pre:tool_call",e=>this.enforcePreTool(e)),t.register("transform:tool_result",e=>{try{this.enforceResponseSize(e)}catch{}})}onTaskStart(t){var i,r;const{taskId:e,sessionId:o,agentName:s}=t.executionContext,n=((r=(i=t.executionContext.agentDefinition)==null?void 0:i.provider)==null?void 0:r.type)??"unknown";this.accumulators.set(e,{taskId:e,sessionId:o??void 0,agentName:s,providerType:n,startedAtMs:Date.now(),uncachedInputTokens:0,cacheReadTokens:0,cacheCreationTokens:0,inputTokens:0,outputTokens:0,peakContextTokens:0,modelCalls:0,totalModelMs:0,toolCalls:new Map,errors:0,consecutiveErrors:0})}onPreModelRequest(t){const e=this.accumulators.get(t.executionContext.taskId);e&&(e.modelCallStartMs=Date.now())}onPostModelResponse(t){const e=this.accumulators.get(t.executionContext.taskId);if(!e)return;const o=t.tokenUsage;if(o){const s=o.inputTokens??0,n=o.cacheReadTokens??0,i=o.cacheCreationTokens??0,a=o.uncachedInputTokens!==void 0||o.cacheReadTokens!==void 0||o.cacheCreationTokens!==void 0?o.uncachedInputTokens??Math.max(0,s-n-i):s;e.uncachedInputTokens+=a,e.cacheReadTokens+=n,e.cacheCreationTokens+=i,e.inputTokens+=a+n+i,e.outputTokens+=o.outputTokens??0,e.peakContextTokens=Math.max(e.peakContextTokens,s)}e.modelCalls+=1,e.modelCallStartMs!==void 0&&(e.totalModelMs+=Date.now()-e.modelCallStartMs,e.modelCallStartMs=void 0)}onTerminal(t){this.accumulators.delete(t)}onPostToolCall(t){const e=this.accumulators.get(t.executionContext.taskId);e&&(t.isError?(e.errors+=1,e.consecutiveErrors+=1):e.consecutiveErrors=0)}mergeDim(t){let e={caps:[]};for(const o of t)o&&(e={caps:[...e.caps??[],...o.caps??[]],mode:o.mode??e.mode,costPerInputToken:o.costPerInputToken??e.costPerInputToken,costPerOutputToken:o.costPerOutputToken??e.costPerOutputToken,costPerCacheReadToken:o.costPerCacheReadToken??e.costPerCacheReadToken,costPerCacheWriteToken:o.costPerCacheWriteToken??e.costPerCacheWriteToken,scope:o.scope??e.scope});return e}resolveCaps(t,e,o){var f,d,k;const s=this.cfg.defaults,n=this.cfg.agent,i=this.cfg.provider,r=this.cfg.tool;if(o){const m=(f=r==null?void 0:r.overrides)==null?void 0:f[o],h=this.mergeDim([s,r==null?void 0:r.default,m]);return{caps:h.caps??[],mode:h.mode,scope:h.scope}}const a=(d=n==null?void 0:n.overrides)==null?void 0:d[t],l=(k=i==null?void 0:i.overrides)==null?void 0:k[e],p=this.mergeDim([s,n==null?void 0:n.default,a,i==null?void 0:i.default,l]);return{caps:p.caps??[],mode:p.mode,scope:p.scope}}queryScopeTotals(t,e,o,s){const n=this.accumulators.get(t),i=n?{inputTokens:n.inputTokens,outputTokens:n.outputTokens,modelCalls:n.modelCalls,peakContextTokens:n.peakContextTokens}:{inputTokens:0,outputTokens:0,modelCalls:0,peakContextTokens:0};if(s==="task"||!this.db)return i;try{const r=this.db;let a;if(s==="session"&&e?a=r.prepare(`SELECT
                COALESCE(SUM(tu.input_tokens), 0) AS input,
                COALESCE(SUM(tu.output_tokens), 0) AS output,
-               COALESCE(SUM(tu.model_calls), 0) AS calls
+               COALESCE(SUM(tu.model_calls), 0) AS calls,
+               COALESCE(MAX(tu.peak_context_tokens), 0) AS peak
              FROM task_usage tu
              JOIN tasks t ON tu.task_id = t.id
-             WHERE t.session_id = ? AND tu.task_id != ?`).get(t,e):s==="agent"?c=a.prepare(`SELECT
+             WHERE t.session_id = ? AND tu.task_id != ?`).get(e,t):s==="agent"?a=r.prepare(`SELECT
                COALESCE(SUM(input_tokens), 0) AS input,
                COALESCE(SUM(output_tokens), 0) AS output,
-               COALESCE(SUM(model_calls), 0) AS calls
+               COALESCE(SUM(model_calls), 0) AS calls,
+               COALESCE(MAX(peak_context_tokens), 0) AS peak
              FROM task_usage
-             WHERE agent_name = ? AND task_id != ?`).get(o,e):s==="global"&&(c=a.prepare(`SELECT
+             WHERE agent_name = ? AND task_id != ?`).get(o,t):s==="global"&&(a=r.prepare(`SELECT
                COALESCE(SUM(input_tokens), 0) AS input,
                COALESCE(SUM(output_tokens), 0) AS output,
-               COALESCE(SUM(model_calls), 0) AS calls
+               COALESCE(SUM(model_calls), 0) AS calls,
+               COALESCE(MAX(peak_context_tokens), 0) AS peak
              FROM task_usage
-             WHERE task_id != ?`).get(e)),c)return{inputTokens:(c.input??0)+n.inputTokens,outputTokens:(c.output??0)+n.outputTokens,modelCalls:(c.calls??0)+n.modelCalls}}catch{}return n}queryWindowTokens(e,t,o,s){if(!this.db)return 0;try{const i=this.db,n=new Date(Date.now()-o).toISOString();let a;if(e==="session"){const c=s?" AND tu.task_id != ?":"",r=[t,n];s&&r.push(s),a=i.prepare(`SELECT COALESCE(SUM(tu.input_tokens + tu.output_tokens), 0) AS total
+             WHERE task_id != ?`).get(t)),a)return{inputTokens:(a.input??0)+i.inputTokens,outputTokens:(a.output??0)+i.outputTokens,modelCalls:(a.calls??0)+i.modelCalls,peakContextTokens:Math.max(a.peak??0,i.peakContextTokens)}}catch{}return i}queryWindowTokens(t,e,o,s){if(!this.db)return 0;try{const n=this.db,i=new Date(Date.now()-o).toISOString();let r;if(t==="session"){const a=s?" AND tu.task_id != ?":"",l=[e,i];s&&l.push(s),r=n.prepare(`SELECT COALESCE(SUM(tu.input_tokens + tu.output_tokens), 0) AS total
              FROM task_usage tu
              JOIN tasks t ON tu.task_id = t.id
-             WHERE t.session_id = ? AND tu.created_at >= ?${c}`).get(...r)}else if(e==="agent"){const c=s?" AND task_id != ?":"",r=[t,n];s&&r.push(s),a=i.prepare(`SELECT COALESCE(SUM(input_tokens + output_tokens), 0) AS total
+             WHERE t.session_id = ? AND tu.created_at >= ?${a}`).get(...l)}else if(t==="agent"){const a=s?" AND task_id != ?":"",l=[e,i];s&&l.push(s),r=n.prepare(`SELECT COALESCE(SUM(input_tokens + output_tokens), 0) AS total
              FROM task_usage
-             WHERE agent_name = ? AND created_at >= ?${c}`).get(...r)}else if(e==="global"){const c=s?" AND task_id != ?":"",r=[n];s&&r.push(s),a=i.prepare(`SELECT COALESCE(SUM(input_tokens + output_tokens), 0) AS total
+             WHERE agent_name = ? AND created_at >= ?${a}`).get(...l)}else if(t==="global"){const a=s?" AND task_id != ?":"",l=[i];s&&l.push(s),r=n.prepare(`SELECT COALESCE(SUM(input_tokens + output_tokens), 0) AS total
              FROM task_usage
-             WHERE created_at >= ?${c}`).get(...r)}return(a==null?void 0:a.total)??0}catch{return 0}}buildSnapshot(e,t,o,s,i,n){const a={};a.inputTokens=t.inputTokens,a.outputTokens=t.outputTokens,a.calls=t.modelCalls,a.wallClock=Date.now()-t.startedAtMs,a.modelMs=t.totalModelMs,a.cost=t.inputTokens*this.costPerInput+t.outputTokens*this.costPerOutput;const c=new Set,r=new Map;for(const l of e){const u=l.scope??n??"task";if(u!=="task"&&c.add(u),l.window){const m=`${u}:${l.window}`;r.has(m)||r.set(m,{scope:u,windowMs:E(l.window)})}}for(const l of c){const u=this.queryScopeTotals(o,s,i,l);a[`${l}:inputTokens`]=u.inputTokens,a[`${l}:outputTokens`]=u.outputTokens,a[`${l}:calls`]=u.modelCalls}for(const[l,{scope:u,windowMs:m}]of r){let d="";u==="session"?d=s??"":u==="agent"&&(d=i??""),a[l]=this.queryWindowTokens(u,d,m,o)}return a}getSnapshotValue(e,t,o){const s=t.scope??o??"task";let i;t.window?i="":i=s!=="task"?`${s}:`:"";let n;switch(t.field){case"inputTokens":n=e[`${i}inputTokens`]??e.inputTokens;break;case"outputTokens":n=e[`${i}outputTokens`]??e.outputTokens;break;case"tokens":n=(e[`${i}inputTokens`]??e.inputTokens)+(e[`${i}outputTokens`]??e.outputTokens);break;case"calls":n=e[`${i}calls`]??e.calls;break;case"wallClock":n=e.wallClock;break;case"modelMs":n=e.modelMs;break;case"cost":n=e.cost;break;case"toolCalls":n=0;break;default:n=0}return t.window&&(n+=e[`${s}:${t.window}`]??0),n}evaluateCap(e,t,o){const s=this.getSnapshotValue(t,e,o);if(s>=e.maximum)throw S(e.field,e.maximum,s,e.message)}enforcePreModel(e){var u,m;const{taskId:t,sessionId:o,agentName:s}=e.executionContext,i=((m=(u=e.executionContext.agentDefinition)==null?void 0:u.provider)==null?void 0:m.type)??"unknown",n=this.accumulators.get(t);if(!n)return;const{caps:a,scope:c}=this.resolveCaps(s,i),r=a.filter(d=>d.field!=="toolCalls");if(r.length===0)return;const l=this.buildSnapshot(r,n,t,o,s,c);for(const d of r)this.evaluateCap(d,l)}enforcePreTool(e){const{toolName:t,callId:o,executionContext:s}=e,{caps:i,mode:n,scope:a}=this.resolveCaps(s.agentName,"",t),c=this.accumulators.get(s.taskId);if(!c||i.length===0)return;const r=c.toolCalls.get(t)??0,l=this.buildSnapshot(i,c,s.taskId,s.sessionId,s.agentName,a);for(const u of i){const m=u.field==="toolCalls"?r:this.getSnapshotValue(l,u,a);if(m>=u.maximum){const d=u.message??`tool "${t}": ${u.field} limit is ${u.maximum}, current value is ${Math.round(m)}`;throw(u.mode??n??"warning")==="warning"?A(t,o,d):S(`tool:${t}:${u.field}`,u.maximum,m,u.message)}}c.toolCalls.set(t,r+1)}enforceResponseSize(e){const{toolName:t,result:o}=e;if(typeof o!="object"||o===null)return;const{caps:s,mode:i}=this.resolveCaps("","",t),n=s.filter(l=>l.field==="responseSize");if(n.length===0)return;const a=o,c=a.content;if(!Array.isArray(c))return;let r=0;for(const l of c)if(typeof l=="object"&&l!==null){const u=l;u.type==="text"&&(r+=(u.text??"").length)}for(const l of n){if(r<=l.maximum)continue;if((l.mode??i??"warning")==="block")a.content=[{type:"text",text:l.message??`Response size (${r} chars) exceeds limit of ${l.maximum}. Use offset/limit or shell paging tools instead.`}],e.isError=!0;else{let m=l.maximum;const d=[];for(const k of c){if(typeof k!="object"||k===null){d.push(k);continue}const h=k;if(h.type!=="text"){d.push(k);continue}const C=h.text??"";if(C.length<=m)d.push(k),m-=C.length;else{d.push({type:"text",text:C.slice(0,m)});break}}d.push({type:"text",text:`
+             WHERE created_at >= ?${a}`).get(...l)}return(r==null?void 0:r.total)??0}catch{return 0}}buildSnapshot(t,e,o,s,n,i,r=0){const a={};a.inputTokens=e.inputTokens,a.outputTokens=e.outputTokens,a.calls=e.modelCalls,a.wallClock=Date.now()-e.startedAtMs,a.modelMs=e.totalModelMs,a.context=Math.max(e.peakContextTokens,r),a.errors=e.errors,a.consecutiveErrors=e.consecutiveErrors,a.cost=e.uncachedInputTokens*this.costPerInput+e.cacheReadTokens*this.costPerCacheRead+e.cacheCreationTokens*this.costPerCacheWrite+e.outputTokens*this.costPerOutput;const l=new Set,p=new Map;for(const f of t){const d=f.scope??i??"task";if(d!=="task"&&l.add(d),f.window){const k=`${d}:${f.window}`;p.has(k)||p.set(k,{scope:d,windowMs:b(f.window)})}}for(const f of l){const d=this.queryScopeTotals(o,s,n,f);a[`${f}:inputTokens`]=d.inputTokens,a[`${f}:outputTokens`]=d.outputTokens,a[`${f}:calls`]=d.modelCalls,a[`${f}:context`]=Math.max(d.peakContextTokens,r)}for(const[f,{scope:d,windowMs:k}]of p){let m="";d==="session"?m=s??"":d==="agent"&&(m=n??""),a[f]=this.queryWindowTokens(d,m,k,o)}return a}getSnapshotValue(t,e,o){const s=e.scope??o??"task";let n;e.window?n="":n=s!=="task"?`${s}:`:"";let i;switch(e.field){case"inputTokens":i=t[`${n}inputTokens`]??t.inputTokens;break;case"outputTokens":i=t[`${n}outputTokens`]??t.outputTokens;break;case"context":i=t[`${n}context`]??t.context;break;case"calls":i=t[`${n}calls`]??t.calls;break;case"wallClock":i=t.wallClock;break;case"modelMs":i=t.modelMs;break;case"cost":i=t.cost;break;case"toolCalls":i=0;break;case"errors":i=t.errors;break;case"consecutiveErrors":i=t.consecutiveErrors;break;default:i=0}return e.window&&(i+=t[`${s}:${e.window}`]??0),i}async emitBudgetEvent(t,e,o,s,n,i){this.hooks&&await this.hooks.emit(t,{executionContext:e,field:o.field,maximum:n,current:s,message:i??o.message??`${o.field} limit is ${n}, current value is ${Math.round(s)}`})}resolveCapLimit(t,e){if(t.maximum!==void 0)return t.maximum;const o=e.agentDefinition.provider;return Math.floor(M.contextWindowFor(o.model)*(t.contextWindowFraction??0))}async evaluateCap(t,e,o,s,n){const i=this.resolveCapLimit(t,o),r=this.getSnapshotValue(e,t,n);if(r<i)return;if((t.mode??s??"warning")==="warning"){await this.emitBudgetEvent("budget:warning",o,t,r,i);return}throw await this.emitBudgetEvent("budget:block",o,t,r,i),v(t.field,i,r,t.message)}async enforcePreModel(t){var d,k;const{taskId:e,sessionId:o,agentName:s}=t.executionContext,n=((k=(d=t.executionContext.agentDefinition)==null?void 0:d.provider)==null?void 0:k.type)??"unknown",i=this.accumulators.get(e);if(!i)return;const{caps:r,mode:a,scope:l}=this.resolveCaps(s,n),p=r.filter(m=>!_.has(m.field));if(p.length===0)return;const f=this.buildSnapshot(p,i,e,o,s,l,L(t.messages,t.tools));for(const m of p)await this.evaluateCap(m,f,t.executionContext,a,l)}async enforcePreTool(t){const{toolName:e,callId:o,executionContext:s}=t,{caps:n,mode:i,scope:r}=this.resolveCaps(s.agentName,"",e),a=this.accumulators.get(s.taskId);if(!a)return;const l=n.filter(d=>d.field!=="context");if(l.length===0)return;const p=a.toolCalls.get(e)??0,f=this.buildSnapshot(l,a,s.taskId,s.sessionId,s.agentName,r);for(const d of l){const k=this.resolveCapLimit(d,s),m=d.field==="toolCalls"?p:this.getSnapshotValue(f,d,r);if(m>=k){const h=d.message??`tool "${e}": ${d.field} limit is ${k}, current value is ${Math.round(m)}`;throw(d.mode??i??"warning")==="warning"?(await this.emitBudgetEvent("budget:warning",s,d,m,k,h),R(e,o,h)):(await this.emitBudgetEvent("budget:block",s,d,m,k,h),v(`tool:${e}:${d.field}`,k,m,d.message))}}a.toolCalls.set(e,p+1)}enforceResponseSize(t){const{toolName:e,result:o}=t;if(typeof o!="object"||o===null)return;const{caps:s,mode:n}=this.resolveCaps("","",e),i=s.filter(p=>p.field==="responseSize");if(i.length===0)return;const r=o,a=r.content;if(!Array.isArray(a))return;let l=0;for(const p of a)if(typeof p=="object"&&p!==null){const f=p;f.type==="text"&&(l+=(f.text??"").length)}for(const p of i){const f=this.resolveCapLimit(p,t.executionContext);if(l<=f)continue;if((p.mode??n??"warning")==="block")r.content=[{type:"text",text:p.message??`Response size (${l} chars) exceeds limit of ${f}. Use offset/limit or shell paging tools instead.`}],t.isError=!0;else{let k=f;const m=[];for(const h of a){if(typeof h!="object"||h===null){m.push(h);continue}const C=h;if(C.type!=="text"){m.push(h);continue}const T=C.text??"";if(T.length<=k)m.push(h),k-=T.length;else{m.push({type:"text",text:T.slice(0,k)});break}}m.push({type:"text",text:`
 
-[truncated: response was ${r} chars, limited to ${l.maximum}. ${l.message??"Use offset/limit or shell paging tools for full content."}]`}),a.content=d}break}}}const M=({db:f,config:e})=>{var i,n;const t=O(e),o=((i=t.defaults)==null?void 0:i.costPerInputToken)??0,s=((n=t.defaults)==null?void 0:n.costPerOutputToken)??0;return new y(f,t,o,s)};exports.configSchema=x;exports.createPlugin=M;exports.default=M;exports.pluginConfigSchema=T;
+[truncated: response was ${l} chars, limited to ${p.maximum}. ${p.message??"Use offset/limit or shell paging tools for full content."}]`}),r.content=m}break}}}const E=({db:c,config:t})=>{var r,a,l,p;const e=$(t),o=((r=e.defaults)==null?void 0:r.costPerInputToken)??0,s=((a=e.defaults)==null?void 0:a.costPerOutputToken)??0,n=((l=e.defaults)==null?void 0:l.costPerCacheReadToken)??o,i=((p=e.defaults)==null?void 0:p.costPerCacheWriteToken)??o;return new W(c,e,o,s,n,i)};exports.configSchema=O;exports.createPlugin=E;exports.default=E;exports.pluginConfigSchema=S;

package/index.d.ts

@@ -1,9 +1,20 @@
 import { PluginFactory } from '@adhd/agent-base-types';
 import { z } from 'zod';
 
+/**
+ * Cap schema. `maximum` is optional at the base level because a `context` cap may
+ * instead use `contextWindowFraction` (0..1 of the model's true window); the
+ * superRefine below enforces the cross-field invariants:
+ *   - `context` requires EXACTLY ONE of `maximum` / `contextWindowFraction`;
+ *   - `context` REJECTS `window` (a windowed PEAK is meaningless; windowed
+ *     cumulative volume is expressible via `inputTokens`/`outputTokens` + `window`);
+ *   - non-`context` fields REJECT `contextWindowFraction` and REQUIRE `maximum`;
+ *   - `errors`/`consecutiveErrors` REJECT `window` and non-task `scope`
+ *     (Packet C — windowed/scoped error budgets are not expressible this wave).
+ */
 declare const capSchema: z.ZodObject<{
     field: z.ZodEnum<{
-        tokens: "tokens";
+        context: "context";
         inputTokens: "inputTokens";
         outputTokens: "outputTokens";
         calls: "calls";
@@ -11,9 +22,12 @@ declare const capSchema: z.ZodObject<{
         modelMs: "modelMs";
         cost: "cost";
         toolCalls: "toolCalls";
+        errors: "errors";
+        consecutiveErrors: "consecutiveErrors";
         responseSize: "responseSize";
     }>;
-    maximum: z.ZodNumber;
+    maximum: z.ZodOptional<z.ZodNumber>;
+    contextWindowFraction: z.ZodOptional<z.ZodNumber>;
     window: z.ZodOptional<z.ZodString>;
     scope: z.ZodOptional<z.ZodEnum<{
         task: "task";
@@ -32,7 +46,7 @@ export declare const pluginConfigSchema: z.ZodObject<{
     defaults: z.ZodOptional<z.ZodObject<{
         caps: z.ZodOptional<z.ZodArray<z.ZodObject<{
             field: z.ZodEnum<{
-                tokens: "tokens";
+                context: "context";
                 inputTokens: "inputTokens";
                 outputTokens: "outputTokens";
                 calls: "calls";
@@ -40,9 +54,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                 modelMs: "modelMs";
                 cost: "cost";
                 toolCalls: "toolCalls";
+                errors: "errors";
+                consecutiveErrors: "consecutiveErrors";
                 responseSize: "responseSize";
             }>;
-            maximum: z.ZodNumber;
+            maximum: z.ZodOptional<z.ZodNumber>;
+            contextWindowFraction: z.ZodOptional<z.ZodNumber>;
             window: z.ZodOptional<z.ZodString>;
             scope: z.ZodOptional<z.ZodEnum<{
                 task: "task";
@@ -62,6 +79,8 @@ export declare const pluginConfigSchema: z.ZodObject<{
         }>>;
         costPerInputToken: z.ZodOptional<z.ZodNumber>;
         costPerOutputToken: z.ZodOptional<z.ZodNumber>;
+        costPerCacheReadToken: z.ZodOptional<z.ZodNumber>;
+        costPerCacheWriteToken: z.ZodOptional<z.ZodNumber>;
         scope: z.ZodOptional<z.ZodEnum<{
             task: "task";
             session: "session";
@@ -73,7 +92,7 @@ export declare const pluginConfigSchema: z.ZodObject<{
         default: z.ZodOptional<z.ZodObject<{
             caps: z.ZodOptional<z.ZodArray<z.ZodObject<{
                 field: z.ZodEnum<{
-                    tokens: "tokens";
+                    context: "context";
                     inputTokens: "inputTokens";
                     outputTokens: "outputTokens";
                     calls: "calls";
@@ -81,9 +100,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     modelMs: "modelMs";
                     cost: "cost";
                     toolCalls: "toolCalls";
+                    errors: "errors";
+                    consecutiveErrors: "consecutiveErrors";
                     responseSize: "responseSize";
                 }>;
-                maximum: z.ZodNumber;
+                maximum: z.ZodOptional<z.ZodNumber>;
+                contextWindowFraction: z.ZodOptional<z.ZodNumber>;
                 window: z.ZodOptional<z.ZodString>;
                 scope: z.ZodOptional<z.ZodEnum<{
                     task: "task";
@@ -103,6 +125,8 @@ export declare const pluginConfigSchema: z.ZodObject<{
             }>>;
             costPerInputToken: z.ZodOptional<z.ZodNumber>;
             costPerOutputToken: z.ZodOptional<z.ZodNumber>;
+            costPerCacheReadToken: z.ZodOptional<z.ZodNumber>;
+            costPerCacheWriteToken: z.ZodOptional<z.ZodNumber>;
             scope: z.ZodOptional<z.ZodEnum<{
                 task: "task";
                 session: "session";
@@ -113,7 +137,7 @@ export declare const pluginConfigSchema: z.ZodObject<{
         overrides: z.ZodDefault<z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodObject<{
             caps: z.ZodOptional<z.ZodOptional<z.ZodArray<z.ZodObject<{
                 field: z.ZodEnum<{
-                    tokens: "tokens";
+                    context: "context";
                     inputTokens: "inputTokens";
                     outputTokens: "outputTokens";
                     calls: "calls";
@@ -121,9 +145,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     modelMs: "modelMs";
                     cost: "cost";
                     toolCalls: "toolCalls";
+                    errors: "errors";
+                    consecutiveErrors: "consecutiveErrors";
                     responseSize: "responseSize";
                 }>;
-                maximum: z.ZodNumber;
+                maximum: z.ZodOptional<z.ZodNumber>;
+                contextWindowFraction: z.ZodOptional<z.ZodNumber>;
                 window: z.ZodOptional<z.ZodString>;
                 scope: z.ZodOptional<z.ZodEnum<{
                     task: "task";
@@ -143,6 +170,8 @@ export declare const pluginConfigSchema: z.ZodObject<{
             }>>>;
             costPerInputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
             costPerOutputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheReadToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheWriteToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
             scope: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
                 task: "task";
                 session: "session";
@@ -155,7 +184,7 @@ export declare const pluginConfigSchema: z.ZodObject<{
         default: z.ZodOptional<z.ZodObject<{
             caps: z.ZodOptional<z.ZodArray<z.ZodObject<{
                 field: z.ZodEnum<{
-                    tokens: "tokens";
+                    context: "context";
                     inputTokens: "inputTokens";
                     outputTokens: "outputTokens";
                     calls: "calls";
@@ -163,9 +192,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     modelMs: "modelMs";
                     cost: "cost";
                     toolCalls: "toolCalls";
+                    errors: "errors";
+                    consecutiveErrors: "consecutiveErrors";
                     responseSize: "responseSize";
                 }>;
-                maximum: z.ZodNumber;
+                maximum: z.ZodOptional<z.ZodNumber>;
+                contextWindowFraction: z.ZodOptional<z.ZodNumber>;
                 window: z.ZodOptional<z.ZodString>;
                 scope: z.ZodOptional<z.ZodEnum<{
                     task: "task";
@@ -185,6 +217,8 @@ export declare const pluginConfigSchema: z.ZodObject<{
             }>>;
             costPerInputToken: z.ZodOptional<z.ZodNumber>;
             costPerOutputToken: z.ZodOptional<z.ZodNumber>;
+            costPerCacheReadToken: z.ZodOptional<z.ZodNumber>;
+            costPerCacheWriteToken: z.ZodOptional<z.ZodNumber>;
             scope: z.ZodOptional<z.ZodEnum<{
                 task: "task";
                 session: "session";
@@ -195,7 +229,7 @@ export declare const pluginConfigSchema: z.ZodObject<{
         overrides: z.ZodDefault<z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodObject<{
             caps: z.ZodOptional<z.ZodOptional<z.ZodArray<z.ZodObject<{
                 field: z.ZodEnum<{
-                    tokens: "tokens";
+                    context: "context";
                     inputTokens: "inputTokens";
                     outputTokens: "outputTokens";
                     calls: "calls";
@@ -203,9 +237,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     modelMs: "modelMs";
                     cost: "cost";
                     toolCalls: "toolCalls";
+                    errors: "errors";
+                    consecutiveErrors: "consecutiveErrors";
                     responseSize: "responseSize";
                 }>;
-                maximum: z.ZodNumber;
+                maximum: z.ZodOptional<z.ZodNumber>;
+                contextWindowFraction: z.ZodOptional<z.ZodNumber>;
                 window: z.ZodOptional<z.ZodString>;
                 scope: z.ZodOptional<z.ZodEnum<{
                     task: "task";
@@ -225,6 +262,8 @@ export declare const pluginConfigSchema: z.ZodObject<{
             }>>>;
             costPerInputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
             costPerOutputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheReadToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheWriteToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
             scope: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
                 task: "task";
                 session: "session";
@@ -235,9 +274,9 @@ export declare const pluginConfigSchema: z.ZodObject<{
     }, z.core.$strip>>;
     tool: z.ZodOptional<z.ZodObject<{
         default: z.ZodOptional<z.ZodObject<{
-            caps: z.ZodOptional<z.ZodArray<z.ZodObject<{
+            caps: z.ZodOptional<z.ZodOptional<z.ZodArray<z.ZodObject<{
                 field: z.ZodEnum<{
-                    tokens: "tokens";
+                    context: "context";
                     inputTokens: "inputTokens";
                     outputTokens: "outputTokens";
                     calls: "calls";
@@ -245,9 +284,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     modelMs: "modelMs";
                     cost: "cost";
                     toolCalls: "toolCalls";
+                    errors: "errors";
+                    consecutiveErrors: "consecutiveErrors";
                     responseSize: "responseSize";
                 }>;
-                maximum: z.ZodNumber;
+                maximum: z.ZodOptional<z.ZodNumber>;
+                contextWindowFraction: z.ZodOptional<z.ZodNumber>;
                 window: z.ZodOptional<z.ZodString>;
                 scope: z.ZodOptional<z.ZodEnum<{
                     task: "task";
@@ -260,24 +302,26 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     block: "block";
                 }>>;
                 message: z.ZodOptional<z.ZodString>;
-            }, z.core.$strip>>>;
-            mode: z.ZodOptional<z.ZodEnum<{
+            }, z.core.$strip>>>>;
+            mode: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
                 warning: "warning";
                 block: "block";
-            }>>;
-            costPerInputToken: z.ZodOptional<z.ZodNumber>;
-            costPerOutputToken: z.ZodOptional<z.ZodNumber>;
-            scope: z.ZodOptional<z.ZodEnum<{
+            }>>>;
+            costPerInputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerOutputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheReadToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheWriteToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            scope: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
                 task: "task";
                 session: "session";
                 agent: "agent";
                 global: "global";
-            }>>;
+            }>>>;
         }, z.core.$strip>>;
         overrides: z.ZodDefault<z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodObject<{
             caps: z.ZodOptional<z.ZodOptional<z.ZodArray<z.ZodObject<{
                 field: z.ZodEnum<{
-                    tokens: "tokens";
+                    context: "context";
                     inputTokens: "inputTokens";
                     outputTokens: "outputTokens";
                     calls: "calls";
@@ -285,9 +329,12 @@ export declare const pluginConfigSchema: z.ZodObject<{
                     modelMs: "modelMs";
                     cost: "cost";
                     toolCalls: "toolCalls";
+                    errors: "errors";
+                    consecutiveErrors: "consecutiveErrors";
                     responseSize: "responseSize";
                 }>;
-                maximum: z.ZodNumber;
+                maximum: z.ZodOptional<z.ZodNumber>;
+                contextWindowFraction: z.ZodOptional<z.ZodNumber>;
                 window: z.ZodOptional<z.ZodString>;
                 scope: z.ZodOptional<z.ZodEnum<{
                     task: "task";
@@ -307,6 +354,8 @@ export declare const pluginConfigSchema: z.ZodObject<{
             }>>>;
             costPerInputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
             costPerOutputToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheReadToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
+            costPerCacheWriteToken: z.ZodOptional<z.ZodOptional<z.ZodNumber>>;
             scope: z.ZodOptional<z.ZodOptional<z.ZodEnum<{
                 task: "task";
                 session: "session";