@gobing-ai/spur 0.3.83 → 0.3.85

package/.claude-plugin/marketplace.json

@@ -7,7 +7,7 @@
     "plugins": [
         {
             "name": "sp",
-            "version": "0.3.83",
+            "version": "0.3.85",
             "source": "./plugins/sp"
         }
     ]

package/config/config.global.yaml

@@ -95,7 +95,7 @@ agent:
     # claude-*-5 via claude, grok-4.6 via grok, gemini-3.x via agy) —
     # consistently outperforms the same model routed through a third-party CLI,
     # and belongs at the capable rungs. `portable` models (glm-5.x,
-    # deepseek-v4-*, …) are provider-agnostic and best carried by omp or pi at
+    # deepseek-v4-*, …) are provider-agnostic and best carried by pi at
     # the cheap/standard rungs. The ladder below mixes both classes on purpose;
     # measure pairings with the history plane before promoting a portable model
     # into a capable rung. Live tiers: cheap | standard | capable-1 | capable-2
@@ -106,19 +106,24 @@ agent:
     # capable-1.
     #
     # Because this is the global layer, a project that wants one different model
-    # writes only `- name: omp` + `model: …` — the agent and tier come from
-    # here.
+    # writes only `- name: pi-dsv4-flash` + `model: …` — the agent and tier come
+    # from here.
     executors:
         # cheap rung is optional: with none declared, scribe-role work starts
         # one rung up at the cheapest standard executor. Add one when
         # transcript-heavy mechanical commands (changelog/gitmsg/handover) get
         # noisy on cost.
         # - name: minimax
-        #   agent: omp
+        #   agent: pi
         #   model: minimax/MiniMax-M3
         #   tier: cheap
-        - name: omp
-          agent: omp
+        # Standard rung. An `agent:` here is a live DISPATCH target, not a label:
+        # role routing (`agent.default: coder`, `--agent auto`, every workflow step
+        # declaring `agent: auto`) lands on the cheapest eligible standard executor,
+        # so listing an agent makes Spur delegate work to it. Keep this rung on an
+        # agent the supported install targets actually carry.
+        - name: pi-dsv4-flash
+          agent: pi
           model: opencode/deepseek-v4-flash
           tier: standard
         - name: pi

package/config/rules/strict/runtime-boundaries.yaml

@@ -57,6 +57,7 @@ rules:
             - "packages/app/src/services/token-ledger-service.ts" # FD byte-window log tailing
             - "packages/app/src/services/token-ledger-watcher.ts" # node:fs watch() live watcher
             - "packages/app/src/services/project-registry.ts" # atomic projects.json persistence
+            - "packages/app/src/services/slash-commands-service.ts" # synchronous ~/.config/spur/slash_commands.json persistence (mirrors project-registry.ts)
             - "packages/app/src/services/history-service.ts" # versioned analyze artifact + bounded-errors sidecar + latest.json symlink pointer (task 0474); ts-runtime FileSystem seam has no symlink, so the pointer uses node:fs directly (mirrors project-registry.ts persistence exemption)
             - "packages/app/src/observability/workflow-run-log-sink.ts" # sync FD append for mid-run tail-able all-in-one run log (task 0426 / feature D2); append() is sync from the observability bus
             - "apps/cli/src/commands/workflow.ts" # FD byte-window tail of the mid-run run log for `workflow trace --follow` streaming (task 0428 / feature D2); readSync at offset over the observability sink's FDs

package/config/workflows/idea-pipeline.yaml

@@ -33,7 +33,7 @@
 #              lets profile=auto route around the idea-eval taste gate (default "false")
 #   CLI --approve-taste sets both design_approved and idea_approved to true.
 #   spurBin  — PATH-independent spur invocation (overridden by CLI at run start)
-#   agent    — agent for agent.run steps (default: omp)
+#   agent    — agent for agent.run steps (default: auto → `agent.default` in config)
 #
 # Reliability (aligned with task-pipeline / ADR-043):
 #   - Soft agent doctor at start → failed via transitions (not raw lifecycle abort)

package/config/workflows/wrapup-pipeline.yaml

@@ -31,7 +31,7 @@
 # merge    — set --vars '{"merge":"true"}' to run branch cleanup (irreversible
 # HITL)
 #   spurBin  — PATH-independent spur invocation (overridden by CLI at run start)
-#   agent    — agent for agent.run steps (default: omp)
+#   agent    — agent for agent.run steps (default: auto → `agent.default` in config)
 #
 # Reliability (aligned with task-pipeline / ADR-043):
 # - Prefer pure slash commands when a command exists; free-form inputs remain

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@gobing-ai/spur",
-    "version": "0.3.83",
+    "version": "0.3.85",
     "description": "Spur CLI — local-first harness for mainstream coding agents: constraint checking, workflow orchestration, agent health, and history analytics. Bun-native; exposes the `spur` command.",
     "keywords": [
         "spur",

package/plugins/sp/agents/super-planner.md

@@ -97,7 +97,8 @@ You own the spaces **between** task runs:
 You explicitly do **NOT** own step-level execution:
 
 - How an `agent.run` step (implement/test/review/verify) runs is `vars.agent`'s concern - default
-  `omp`, pinned in `task-pipeline.yaml`. `--agent <value>` from the command flows into each
+  `auto` (`agent.default` in config resolves the executor), pinned in `task-pipeline.yaml`.
+  `--agent <value>` from the command flows into each
   per-task `vars.agent`; you forward it, you do not interpret it.
 - You never edit the pipeline YAML, never reach into a step, and never decide how a single
   `agent.run` stage executes. The per-task pipeline is invoked **verbatim**.

package/plugins/sp/lib/idea-handoff.generated.mjs

@@ -1325,5 +1325,5 @@ ${G}
 `);this._frontmatter={raw:Z,data:{...this._frontmatter.data,[$]:U}};let W=this._frontmatterBlock.indexOf(X);this._frontmatterBlock=W===-1?`---
 ${Z}
 ---
-`:this._frontmatterBlock.slice(0,W)+Z+this._frontmatterBlock.slice(W+X.length)}serialize(){let $=this._frontmatterBlock;$+=this._preamble;for(let U of this._sections)$+=U.modifiedText??U.originalText;return $}}function Do($){return`"${$.replace(/\\/g,"\\\\").replace(/"/g,"\\\"")}"`}function Ko($){if($.startsWith('"')&&$.endsWith('"'))return $;if(/[:{}[\],&*?|>!%@`#"'\\\n\r]/.test($))return`"${$.replace(/\\/g,"\\\\").replace(/"/g,"\\\"")}"`;if(/^(null|true|false|yes|no|on|off)$/i.test($))return`"${$}"`;if(/^\d/.test($)&&!/^\d{4}-\d{2}-\d{2}/.test($))return`"${$}"`;return $}G6();var LA=["backlog","todo","wip","testing","blocked","done","cancelled"],Fo=["backlog","active","verifying","blocked","done","cancelled"];var QV=["P0","P1","P2","P3"],qo=["simple","standard","complex","research","refine","plan","unit","review","docs"],Io=["task","issue","review","meta","brainstorm"],OA=["standard","feature-impl","issue","review","meta","brainstorm"];var PA=/^[A-Z][1-9]*$/;var Bo={completed:"done",complete:"done","in-progress":"wip","in progress":"wip",in_progress:"wip",dropped:"cancelled",cancel:"cancelled",canceled:"cancelled",new:"backlog",pending:"backlog",blocked:"blocked",wip:"wip",testing:"testing",done:"done",cancelled:"cancelled",backlog:"backlog",todo:"todo"};function Lo($){let U=$.trim().toLowerCase(),J=Bo[U];if(J===void 0)throw Error(`Unknown task status: ${JSON.stringify($)} (allowed: ${LA.join(", ")})`);return J}var w7=F.string().min(1).refine(($)=>!Number.isNaN(Date.parse($)),{message:"must be an ISO 8601 timestamp"}),jA=F.string().regex(PA,{message:"feature id must match ^[A-Z][1-9]*$ (DD-14)"}).nullable().optional(),RA=F.string().regex(/^\d{4}$/,{message:"parent_wbs must be a 4-digit WBS string"}).nullable().optional(),yF$=F.object({schema_version:F.literal(1),name:F.string().min(1),description:F.string().optional(),status:F.preprocess(($)=>{if(typeof $==="string")try{return Lo($)}catch{}return $},F.enum(LA)),type:F.enum(Io).optional().default("task"),template:F.enum(OA).optional(),profile:F.enum(qo).optional(),feature_id:jA,parent_wbs:RA,priority:F.enum(QV).optional(),tags:F.array(F.string()).optional(),dependencies:F.array(F.string()).optional(),ac_numbering:F.literal("task-local").optional(),ac_altitude:F.enum(["graduating","task-local"]).optional(),done_forced:F.preprocess(($)=>typeof $==="string"?$==="true":$,F.boolean().optional()),done_reason:F.string().optional(),feature_link_declined:F.preprocess(($)=>typeof $==="string"?$==="true":$,F.boolean().optional()),created_at:w7,updated_at:w7}),fF$=F.object({schema_version:F.literal(1),id:F.string().regex(PA,{message:"feature id must match ^[A-Z][1-9]*$ (DD-14)"}),name:F.string().min(1),status:F.enum(Fo),priority:F.enum(QV).optional(),tags:F.array(F.string()).optional(),created_at:w7,updated_at:w7});var Oo=F.object({name:F.string().min(1,"name is required"),background:F.string().optional(),requirements:F.string().optional(),design:F.string().optional(),plan:F.string().optional(),acceptance_criteria:F.string().optional(),feature_id:jA,parent_wbs:RA,priority:F.enum(QV).optional(),tags:F.array(F.string()).optional(),template:F.enum(OA).optional()}).strict(),Po=F.array(Oo).min(1,"batch must contain at least one task");u0();G6();var DV=F.enum(Q0),jo=F.object({required:F.array(DV).optional(),optional:F.array(DV).optional(),forbidden:F.array(DV).optional(),gate:F.boolean().optional()}).strict(),Ro=F.enum(["backlog","todo","wip","testing","blocked","done","cancelled"]),cF$=F.object({variants:F.record(F.string(),F.partialRecord(Ro,jo))}).strict();var dF$=new Map(Q0.map(($,U)=>[$,U]));var No=["requirements","design","plan","ac","decisions","dependencies","premises"];function MA($){return`/sp:dev-refine ${$} --auto --depth ready`}var Ao={Solution:!0,Testing:!0,Review:!0,History:!0},vo=Q0.filter(($)=>!($ in Ao));function NA($){let U=H2.parse($,"task"),J=U.frontmatterData??{},G={sections:vo.map((X)=>{let Y=U.getSection(X);return{name:X,body:Y===null?null:Y.trim()}}),featureId:typeof J.feature_id==="string"?J.feature_id:null,template:typeof J.template==="string"?J.template:null,dependencies:Array.isArray(J.dependencies)?[...J.dependencies].map(String).sort():null};return Mo("sha256").update(JSON.stringify(G)).digest("hex")}function AA($){if($===void 0||$.length===0)return{ok:!1,reason:"no ready-checklist evidence"};for(let U of No){let J=$.find((G)=>G.id===U);if(J===void 0)return{ok:!1,reason:`checklist row "${U}" missing`};if(J.pass!==!0)return{ok:!1,reason:`checklist row "${U}" not passing`};if(typeof J.evidence!=="string"||J.evidence.trim()==="")return{ok:!1,reason:`checklist row "${U}" has empty evidence`}}return{ok:!0}}var To=/[;&|<>$`(){}[\]!*?~#\n\r"']/;function vA($,U){if(To.test($))return{error:`${U} must not contain shell metacharacters (got ${$})`};let J=$.trim().split(/\s+/).filter((X)=>X.length>0),G=J[0];if(G===void 0)return{error:`Action option ${U.slice(U.indexOf('"'))} must be a non-empty string`};return{command:G,leadingArgs:J.slice(1)}}function K0($){return`exit=${$.exitCode??"null"}${$.signal?` signal=${$.signal}`:""}: ${$.stderr.trim()||"no stderr"}`}async function TA($){let U=$.projectRoot??process.cwd(),J=$.fileSystem??HZ(),G=$.processExecutor??new Z7,X=$.spurBin??"spur",{runId:Y,featureId:H}=$,_=D0(U,".spur","run"),Z=D0(_,`${Y}-idea-task-batch.json`),W=D0(_,`${Y}-idea-batch-create-result.json`),z=D0(_,`${Y}-idea-task-order.json`),V=D0(_,`${Y}-idea-ready.json`),Q=D0(_,`${Y}-idea-handoff.md`),D=vA(X,'idea-handoff "spurBin"');if("error"in D)return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:D.error};let{command:q,leadingArgs:I}=D;if(!await J.exists(Z)||!await J.exists(W)||!await J.exists(z))return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:"Required batch, result, or order files missing in .spur/run/"};try{let L=JSON.parse(await J.readFile(Z)),P=JSON.parse(await J.readFile(W)),R=JSON.parse(await J.readFile(z)),v=await J.exists(V),M=await(async()=>{if(!v)return[];try{let f=JSON.parse(await J.readFile(V)),V$=typeof f==="object"&&f!==null&&"tasks"in f?f.tasks:void 0;return Array.isArray(V$)?V$:[]}catch{return[]}})(),E=new Map;for(let f of M)if(typeof f?.wbs==="string")E.set(f.wbs,f);if(!Array.isArray(L)||!Array.isArray(P?.wbs)||!Array.isArray(R))return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:"Malformed batch, result, or order JSON structure"};if(L.length!==P.wbs.length)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Batch size mismatch: ${L.length} items declared but ${P.wbs.length} WBS created`};let x=L.map((f)=>f.name);if(new Set(x).size!==x.length)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:"Duplicate task names found in batch declaration"};for(let f of R){let V$=x.indexOf(f.name);if(V$===-1||!P.wbs[V$])return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Task name "${f.name}" from order could not be mapped to created WBS`};let K$=P.wbs[V$],H4=[];for(let G$ of f.depends_on_names??[]){let F$=x.indexOf(G$);if(F$===-1||!P.wbs[F$])return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Dependency task name "${G$}" could not be mapped to created WBS`};H4.push(P.wbs[F$])}if(H4.length>0){let G$=await G.run({command:q,args:[...I,"task","deps",K$,"set",...H4,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(G$.exitCode!==0)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Failed to set dependencies for task ${K$}: ${K0(G$)}`}}}let p=await G.run({command:q,args:[...I,"feature","refresh","--feature",H,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(p.exitCode!==0)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Feature refresh for ${H} failed: ${K0(p)}`};let w=[];for(let f of P.wbs){let V$=await G.run({command:q,args:[...I,"task","path",f,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(V$.exitCode===null)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Task path for ${f} could not be spawned (${q}): ${K0(V$)}`};let K$;if(V$.exitCode!==0)K$=`task path resolution failed: ${K0(V$)}`;else{let G$;try{let F$=JSON.parse(V$.stdout);if(typeof F$==="object"&&F$!==null&&"filePath"in F$){let j$=F$.filePath;if(typeof j$==="string")G$=j$}}catch{G$=void 0}if(G$===void 0)K$="task path output carried no filePath";else{let F$;try{F$=NA(await J.readFile(G$))}catch{F$=void 0}if(F$===void 0)K$=`task file unreadable at ${G$}`;else{let j$=E.get(f);if(j$===void 0)K$="no ready evidence recorded for this run";else if(j$.status!=="ready")K$=`ready evidence status is ${String(j$.status)}`;else if(typeof j$.planningDigest!=="string"||j$.planningDigest!==F$)K$="planning digest stale — task content changed after preparation";else{let F0=AA(Array.isArray(j$.checks)?j$.checks:void 0);if(!F0.ok)K$=F0.reason}}}}if(K$===void 0){let G$=await G.run({command:q,args:[...I,"task","check",f,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(G$.exitCode===null)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Task check for ${f} could not be spawned (${q}): ${K0(G$)}`};if(G$.exitCode!==0)K$=`deterministic task check failed (${K0(G$)})`}let H4=K$===void 0;w.push({wbs:f,pass:H4,status:H4?"ready":"unready",...H4?{}:{reason:K$,action:MA(f)}})}let r=w.some((f)=>!f.pass),Z$=r?`/sp:dev-refineall --feature ${H} --auto --depth ready`:`/sp:dev-runall --feature ${H} --auto`,z$=["# Idea pipeline handoff report","",`Feature: ${H}`,`Run ID: ${Y}`,"","## Created tasks",...P.wbs.map((f)=>`  - ${f}`),"","## Per-task readiness (evidence + digest + task check)","","| WBS | Outcome | Reason |","|-----|---------|--------|",...w.map((f)=>`| ${f.wbs} | ${f.pass?"READY":"UNREADY"} | ${f.reason??"-"} |`),"",...r?["## Preparation actions","",...w.filter((f)=>!f.pass).map((f)=>`  - ${f.wbs}: ${f.action}`),""]:[],"## Next command","",Z$,""];return await J.ensureDir(_),await J.writeFile(Q,z$.join(`
+`:this._frontmatterBlock.slice(0,W)+Z+this._frontmatterBlock.slice(W+X.length)}serialize(){let $=this._frontmatterBlock;$+=this._preamble;for(let U of this._sections)$+=U.modifiedText??U.originalText;return $}}function Do($){return`"${$.replace(/\\/g,"\\\\").replace(/"/g,"\\\"")}"`}function Ko($){if($.startsWith('"')&&$.endsWith('"'))return $;if(/[:{}[\],&*?|>!%@`#"'\\\n\r]/.test($))return`"${$.replace(/\\/g,"\\\\").replace(/"/g,"\\\"")}"`;if(/^(null|true|false|yes|no|on|off)$/i.test($))return`"${$}"`;if(/^\d/.test($)&&!/^\d{4}-\d{2}-\d{2}/.test($))return`"${$}"`;return $}G6();var LA=["backlog","todo","wip","testing","blocked","done","cancelled"],Fo=["backlog","active","verifying","blocked","done","cancelled"];var QV=["P0","P1","P2","P3"],qo=["simple","standard","complex","research","refine","plan","unit","review","docs"],Io=["task","issue","review","meta","brainstorm"],OA=["standard","feature-impl","issue","review","meta","brainstorm"];var PA=/^[A-Z][1-9]*$/;var Bo={completed:"done",complete:"done","in-progress":"wip","in progress":"wip",in_progress:"wip",dropped:"cancelled",cancel:"cancelled",canceled:"cancelled",new:"backlog",pending:"backlog",blocked:"blocked",wip:"wip",testing:"testing",done:"done",cancelled:"cancelled",backlog:"backlog",todo:"todo"};function Lo($){let U=$.trim().toLowerCase(),J=Bo[U];if(J===void 0)throw Error(`Unknown task status: ${JSON.stringify($)} (allowed: ${LA.join(", ")})`);return J}var w7=F.string().min(1).refine(($)=>!Number.isNaN(Date.parse($)),{message:"must be an ISO 8601 timestamp"}),jA=F.string().regex(PA,{message:"feature id must match ^[A-Z][1-9]*$ (DD-14)"}).nullable().optional(),RA=F.string().regex(/^\d{4}$/,{message:"parent_wbs must be a 4-digit WBS string"}).nullable().optional(),yF$=F.object({schema_version:F.literal(1),name:F.string().min(1),description:F.string().optional(),status:F.preprocess(($)=>{if(typeof $==="string")try{return Lo($)}catch{}return $},F.enum(LA)),type:F.enum(Io).optional().default("task"),template:F.enum(OA).optional(),profile:F.enum(qo).optional(),feature_id:jA,parent_wbs:RA,priority:F.enum(QV).optional(),tags:F.array(F.string()).optional(),dependencies:F.array(F.string()).optional(),ac_numbering:F.literal("task-local").optional(),ac_altitude:F.enum(["graduating","task-local"]).optional(),done_forced:F.preprocess(($)=>typeof $==="string"?$==="true":$,F.boolean().optional()),done_reason:F.string().optional(),feature_link_declined:F.preprocess(($)=>typeof $==="string"?$==="true":$,F.boolean().optional()),estimate_hours:F.preprocess(($)=>typeof $==="string"&&$.trim()!==""?Number($):$,F.number().positive().optional()),created_at:w7,updated_at:w7}),fF$=F.object({schema_version:F.literal(1),id:F.string().regex(PA,{message:"feature id must match ^[A-Z][1-9]*$ (DD-14)"}),name:F.string().min(1),status:F.enum(Fo),priority:F.enum(QV).optional(),tags:F.array(F.string()).optional(),created_at:w7,updated_at:w7});var Oo=F.object({name:F.string().min(1,"name is required"),background:F.string().optional(),requirements:F.string().optional(),design:F.string().optional(),plan:F.string().optional(),acceptance_criteria:F.string().optional(),feature_id:jA,parent_wbs:RA,priority:F.enum(QV).optional(),tags:F.array(F.string()).optional(),template:F.enum(OA).optional(),estimate_hours:F.number().positive().optional()}).strict(),Po=F.array(Oo).min(1,"batch must contain at least one task");u0();G6();var DV=F.enum(Q0),jo=F.object({required:F.array(DV).optional(),optional:F.array(DV).optional(),forbidden:F.array(DV).optional(),gate:F.boolean().optional()}).strict(),Ro=F.enum(["backlog","todo","wip","testing","blocked","done","cancelled"]),cF$=F.object({variants:F.record(F.string(),F.partialRecord(Ro,jo))}).strict();var dF$=new Map(Q0.map(($,U)=>[$,U]));var No=["requirements","design","plan","ac","decisions","dependencies","premises"];function MA($){return`/sp:dev-refine ${$} --auto --depth ready`}var Ao={Solution:!0,Testing:!0,Review:!0,History:!0},vo=Q0.filter(($)=>!($ in Ao));function NA($){let U=H2.parse($,"task"),J=U.frontmatterData??{},G={sections:vo.map((X)=>{let Y=U.getSection(X);return{name:X,body:Y===null?null:Y.trim()}}),featureId:typeof J.feature_id==="string"?J.feature_id:null,template:typeof J.template==="string"?J.template:null,dependencies:Array.isArray(J.dependencies)?[...J.dependencies].map(String).sort():null};return Mo("sha256").update(JSON.stringify(G)).digest("hex")}function AA($){if($===void 0||$.length===0)return{ok:!1,reason:"no ready-checklist evidence"};for(let U of No){let J=$.find((G)=>G.id===U);if(J===void 0)return{ok:!1,reason:`checklist row "${U}" missing`};if(J.pass!==!0)return{ok:!1,reason:`checklist row "${U}" not passing`};if(typeof J.evidence!=="string"||J.evidence.trim()==="")return{ok:!1,reason:`checklist row "${U}" has empty evidence`}}return{ok:!0}}var To=/[;&|<>$`(){}[\]!*?~#\n\r"']/;function vA($,U){if(To.test($))return{error:`${U} must not contain shell metacharacters (got ${$})`};let J=$.trim().split(/\s+/).filter((X)=>X.length>0),G=J[0];if(G===void 0)return{error:`Action option ${U.slice(U.indexOf('"'))} must be a non-empty string`};return{command:G,leadingArgs:J.slice(1)}}function K0($){return`exit=${$.exitCode??"null"}${$.signal?` signal=${$.signal}`:""}: ${$.stderr.trim()||"no stderr"}`}async function TA($){let U=$.projectRoot??process.cwd(),J=$.fileSystem??HZ(),G=$.processExecutor??new Z7,X=$.spurBin??"spur",{runId:Y,featureId:H}=$,_=D0(U,".spur","run"),Z=D0(_,`${Y}-idea-task-batch.json`),W=D0(_,`${Y}-idea-batch-create-result.json`),z=D0(_,`${Y}-idea-task-order.json`),V=D0(_,`${Y}-idea-ready.json`),Q=D0(_,`${Y}-idea-handoff.md`),D=vA(X,'idea-handoff "spurBin"');if("error"in D)return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:D.error};let{command:q,leadingArgs:I}=D;if(!await J.exists(Z)||!await J.exists(W)||!await J.exists(z))return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:"Required batch, result, or order files missing in .spur/run/"};try{let L=JSON.parse(await J.readFile(Z)),P=JSON.parse(await J.readFile(W)),R=JSON.parse(await J.readFile(z)),v=await J.exists(V),M=await(async()=>{if(!v)return[];try{let f=JSON.parse(await J.readFile(V)),V$=typeof f==="object"&&f!==null&&"tasks"in f?f.tasks:void 0;return Array.isArray(V$)?V$:[]}catch{return[]}})(),E=new Map;for(let f of M)if(typeof f?.wbs==="string")E.set(f.wbs,f);if(!Array.isArray(L)||!Array.isArray(P?.wbs)||!Array.isArray(R))return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:"Malformed batch, result, or order JSON structure"};if(L.length!==P.wbs.length)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Batch size mismatch: ${L.length} items declared but ${P.wbs.length} WBS created`};let x=L.map((f)=>f.name);if(new Set(x).size!==x.length)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:"Duplicate task names found in batch declaration"};for(let f of R){let V$=x.indexOf(f.name);if(V$===-1||!P.wbs[V$])return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Task name "${f.name}" from order could not be mapped to created WBS`};let K$=P.wbs[V$],H4=[];for(let G$ of f.depends_on_names??[]){let F$=x.indexOf(G$);if(F$===-1||!P.wbs[F$])return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Dependency task name "${G$}" could not be mapped to created WBS`};H4.push(P.wbs[F$])}if(H4.length>0){let G$=await G.run({command:q,args:[...I,"task","deps",K$,"set",...H4,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(G$.exitCode!==0)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Failed to set dependencies for task ${K$}: ${K0(G$)}`}}}let p=await G.run({command:q,args:[...I,"feature","refresh","--feature",H,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(p.exitCode!==0)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Feature refresh for ${H} failed: ${K0(p)}`};let w=[];for(let f of P.wbs){let V$=await G.run({command:q,args:[...I,"task","path",f,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(V$.exitCode===null)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Task path for ${f} could not be spawned (${q}): ${K0(V$)}`};let K$;if(V$.exitCode!==0)K$=`task path resolution failed: ${K0(V$)}`;else{let G$;try{let F$=JSON.parse(V$.stdout);if(typeof F$==="object"&&F$!==null&&"filePath"in F$){let j$=F$.filePath;if(typeof j$==="string")G$=j$}}catch{G$=void 0}if(G$===void 0)K$="task path output carried no filePath";else{let F$;try{F$=NA(await J.readFile(G$))}catch{F$=void 0}if(F$===void 0)K$=`task file unreadable at ${G$}`;else{let j$=E.get(f);if(j$===void 0)K$="no ready evidence recorded for this run";else if(j$.status!=="ready")K$=`ready evidence status is ${String(j$.status)}`;else if(typeof j$.planningDigest!=="string"||j$.planningDigest!==F$)K$="planning digest stale — task content changed after preparation";else{let F0=AA(Array.isArray(j$.checks)?j$.checks:void 0);if(!F0.ok)K$=F0.reason}}}}if(K$===void 0){let G$=await G.run({command:q,args:[...I,"task","check",f,"--json"],cwd:U,forceBuffered:!0,rejectOnError:!1});if(G$.exitCode===null)return{ok:!1,wbsList:P.wbs,nextCommand:"",reportPath:Q,error:`Task check for ${f} could not be spawned (${q}): ${K0(G$)}`};if(G$.exitCode!==0)K$=`deterministic task check failed (${K0(G$)})`}let H4=K$===void 0;w.push({wbs:f,pass:H4,status:H4?"ready":"unready",...H4?{}:{reason:K$,action:MA(f)}})}let r=w.some((f)=>!f.pass),Z$=r?`/sp:dev-refineall --feature ${H} --auto --depth ready`:`/sp:dev-runall --feature ${H} --auto`,z$=["# Idea pipeline handoff report","",`Feature: ${H}`,`Run ID: ${Y}`,"","## Created tasks",...P.wbs.map((f)=>`  - ${f}`),"","## Per-task readiness (evidence + digest + task check)","","| WBS | Outcome | Reason |","|-----|---------|--------|",...w.map((f)=>`| ${f.wbs} | ${f.pass?"READY":"UNREADY"} | ${f.reason??"-"} |`),"",...r?["## Preparation actions","",...w.filter((f)=>!f.pass).map((f)=>`  - ${f.wbs}: ${f.action}`),""]:[],"## Next command","",Z$,""];return await J.ensureDir(_),await J.writeFile(Q,z$.join(`
 `)),{ok:!0,wbsList:P.wbs,nextCommand:Z$,reportPath:Q}}catch(L){return{ok:!1,wbsList:[],nextCommand:"",reportPath:Q,error:L instanceof Error?L.message:String(L)}}}async function lq$($,U=TA){let J=$.__runId??"",G=$.featureId??"";if(J===""||G==="")return y7("idea-handoff: __runId and featureId env vars are required"),{exitCode:1};let X=await U({runId:J,featureId:G,spurBin:$.spurBin??"spur"});if(!X.ok)return y7(`idea-handoff: ${X.error??"failed"}`),{exitCode:1,result:X};return b7(`idea-handoff: wrote ${X.reportPath} (${X.wbsList.length} task(s))`),b7(`idea-handoff: next -> ${X.nextCommand}`),{exitCode:0,result:X}}export{lq$ as runIdeaHandoffCli};

package/plugins/sp/plugin.json

@@ -1,6 +1,6 @@
 {
     "name": "sp",
-    "version": "0.3.83",
+    "version": "0.3.85",
     "description": "Spur — a local-first harness engineering toolkit that wraps mainstream coding agents with constraint checking, workflow orchestration, and history analytics.",
     "extensions": {
         "pi": ["./hooks/pi/guard-extension.ts"]

package/plugins/sp/scripts/surface-drift-inventory.ts

@@ -636,7 +636,7 @@ const JSON_PROBES: string[][] = [
     ['feature', 'show', 'I3', '--json'],
     ['rule', 'list', '--json'],
     ['agent', 'list', '--json'],
-    ['agent', 'doctor', 'omp', '--json'],
+    ['agent', 'doctor', 'pi', '--json'],
     ['workflow', 'list', '--json'],
     ['workflow', 'validate', 'basic.yaml', '--json'],
     ['status', '--json'],

package/plugins/sp/skills/parallel-execution/references/dispatch-surface.md

@@ -65,6 +65,19 @@ and put prompt-layer routing policy into a domain-layer registry. This reference
 surface carries the work*; ADR-033 decides *which tier runs on it*. Read both; apply each to its
 own axis.
 
+## Cheap-model hint for read-only fan-out (2026-09-15 subagent-dispatch evaluation)
+
+Surface choice and model choice stay separate axes, but the invocation surface is where the
+cheap-model hint is applied. On hosts whose native subagent invocation accepts a per-call model
+override (Claude Code: the Agent tool's `model` parameter), dispatch read-only fan-out workers —
+`dev-parallel --mode investigation` and competency-lens review's read-only lenses — with the host's
+cheapest capable model. The rule is owned by
+[fan-out-patterns.md § Cheap-model hint for read-only fan-out](fan-out-patterns.md#cheap-model-hint-for-read-only-fan-out)
+and carries two prohibitions and one exemption: invocation-side only, **never** a subagent
+definition-file edit (superskill owns those), **never** a hard pin (the host may ignore it), and
+`sp:spur-dev` pipeline stage dispatch is exempt — ADR-033's `model_policy` and ADR-078's role
+config own that axis.
+
 ## The sandbox reliability tax on `spur agent run`
 
 `spur agent run` runs the target agent as an external process. Under a sandboxed Bash session it
@@ -98,6 +111,30 @@ native subagent with shared-worktree capability); the inline driver remains the
 provenance, artifact validation, and no-replay guarantees. This reference stays the authority for
 the native-subagent versus `spur agent run` choice everywhere else.
 
+## Fork dispatch on hosts that support it (2026-09-15 subagent-dispatch evaluation)
+
+A fresh native subagent starts with an empty context: it re-loads every skill file, task document,
+and convention the host already holds — the dominant cost of dispatching a context-heavy stage.
+Claude Code offers a **fork** subagent that inherits the parent's full conversation and shares its
+prompt cache, so the worker starts warm instead of re-ingesting. When a dispatch's value depends on
+the host's accumulated context — an implement stage following a long precheck/design discussion, or
+an investigation fan-out whose question only makes sense against what the session already learned —
+prefer the fork surface over a cold native subagent.
+
+Three constraints keep fork honest:
+
+- **Host-specific.** Fork is a Claude Code capability. Hosts without it use the standard native
+  subagent; this paragraph changes nothing there.
+- **Fork always runs the parent's model.** It is never a cost-routing move — cheap-tier work
+  (scribe-role fan-out, read-only exploration) still goes to a cold cheap worker.
+- **Fork inherits the host's framing.** Stages whose value is independent eyes — review, verify,
+  adversarial panels — must NOT fork; a worker that inherits the author's reasoning grades its own
+  homework. Fork is for continuity, cold dispatch is for independence.
+
+The inline driver's resume-over-re-dispatch rule
+([inline-pipeline-driver.md](../../spur-dev/references/inline-pipeline-driver.md)) is the same
+instinct one level down: reuse a warm worker for continuation, dispatch fresh for judgment.
+
 ## Role propagation across fan-out (task 0551, feature I4)
 
 When a run dispatches subagents, the **effective role** each subagent resolves through follows one

package/plugins/sp/skills/parallel-execution/references/fan-out-patterns.md

@@ -42,7 +42,7 @@ Four proven patterns for parallel subagent execution. Each pattern has a distinc
 
 **Example:** Code review of a PR → correctness lens, security lens, efficiency lens, maintainability lens. Each subagent produces per-lens findings; synthesis merges into a unified P1–P4 table.
 
-**Anti-pattern:** Using competency-lens review when a single reviewer would catch everything. For a 20-line change, one thorough review beats 3 shallow ones.
+**Anti-pattern:** Using competency-lens review when a single reviewer would catch everything. A sub-floor scope (see [Size floor](#size-floor)) is cheaper as one thorough pass than as N shallow ones — do not fan it out at all.
 
 ### 3. Independent-Task Batch
 
@@ -89,6 +89,7 @@ Four proven patterns for parallel subagent execution. Each pattern has a distinc
 | Sequential dependency chain | **Do not fan out** | — |
 | Single file touched by multiple tasks | **Do not fan out** | — |
 | Token budget < 20k remaining | **Do not fan out** | — |
+| Scope below the [Size floor](#size-floor) | **Do not fan out — execute inline** | — |
 
 ## Token-budget guard
 
@@ -99,3 +100,56 @@ remaining_budget >= (N × per_subagent_estimate) + synthesis_estimate
 ```
 
 If not: reduce N, or serialize. A fan-out that exhausts the budget mid-run leaves partial results that are worse than sequential. The driver is responsible for this check — the skill provides the estimates; the orchestrator applies them.
+
+## Size floor
+
+Fanning out below a minimum work size wastes the dispatch: a ~20-line change is cheaper as one
+thorough inline pass than as 3 shallow parallel ones. The floor is met — and the driver MUST
+**not** fan out, executing inline instead — when **any** of these constants holds:
+
+- the scope is a **single file**; or
+- the estimated diff is **under ~50 lines**; or
+- the work is **one focused question**.
+
+The thresholds are constants: they are measured from observable scope (files touched, diff lines,
+a single question), never from a model estimate, and never exposed as a config knob — a value that
+never changes is a constant. The floor decides *whether to fan out at all*; once it is met, the
+[When-to-use decision table](#when-to-use-decision-table) and the
+[Token-budget guard](#token-budget-guard) decide *which pattern* and *how many*. This is the
+pre-dispatch gate, not an anti-pattern remark.
+
+## Pre-dispatch permission check
+
+Claude Code exposes no dry-run permission API, so the contract is **name-capabilities + fail-fast
+blocker** — never a permission-probing subsystem:
+
+1. Before dispatching a worker, the orchestrator names the required capabilities in the dispatch
+   prompt: the read/search tools it needs, every shell action it will run, and any write surface it
+   holds.
+2. The worker is instructed to **return a blocker immediately on the first missing capability**
+   instead of waiting at a permission prompt the host cannot see — a worker parked at that prompt is
+   invisible until join.
+3. Record one run-log line per dispatch: `stage <id> permission precheck: ok | <missing capability>`
+   (for ad-hoc fan-out, `<id>` is the dispatched worker's ledger id from
+   [Keep a durable progress ledger](../SKILL.md#keep-a-durable-progress-ledger)).
+
+Read-only investigation fan-out avoids the write-permission class by construction:
+`dev-parallel --mode investigation` and competency-lens review's read-only lenses dispatch
+**read-only worker shapes** — search/read tools only, with no write surface declared.
+
+## Cheap-model hint for read-only fan-out
+
+On hosts whose native subagent invocation accepts a per-call model override (Claude Code: the Agent
+tool's `model` parameter), dispatch read-only fan-out workers — `dev-parallel --mode investigation`
+and competency-lens review's read-only lenses — with the host's cheapest capable model. The rule is
+invocation-side, with two prohibitions and one exemption:
+
+- **Never a definition-file edit.** Subagent definition files (`plugins/sp/agents/*.md`) belong to
+  superskill's lifecycle; this plugin never edits them to route cost.
+- **Never a hard pin.** The host may ignore the override; the hint names a preference, not a
+  contract.
+- **`sp:spur-dev` pipeline stage dispatch is exempt.** Tier routing there is ADR-033/ADR-078's job
+  via [`plugins/sp/references/roles.md`](../../../references/roles.md); this hint does not touch it.
+
+Surface choice is [dispatch-surface.md](dispatch-surface.md)'s axis; this hint applies on the
+surface it selects, not instead of it.

package/plugins/sp/skills/spec-decomposition/references/decomposition.md

@@ -46,6 +46,18 @@ Apply the two in order:
 Without the second clause, cohesion reads as "never split" — the opposite failure. The knobs are
 the escape hatch; cohesion is the default.
 
+### Persist the estimate: `estimate_hours`
+
+Every batch item's hour estimate MUST also travel into the batch JSON as the item's
+`estimate_hours` field (a positive number) — `spur task batch-create` persists it to the task's
+frontmatter, and it can be corrected later with `spur task update <wbs> --estimate-hours <n>`. The
+estimate already exists at this point (the hour knobs were just applied above); writing it down
+costs nothing and makes the size decision visible to later stages instead of re-derived. Its
+primary consumer is the inline pipeline driver's dispatch floor
+([inline-pipeline-driver.md](../../spur-dev/references/inline-pipeline-driver.md)): a task at or
+below 1 hour executes its stages host-inline rather than paying a subagent dispatch. An estimate
+left unwritten forgoes that gate — the task dispatches like any large one.
+
 ### Worked example: H8's own first decomposition
 
 Feature H8 ("sp command surface coherence") decomposed into five tasks, each 3–8h — fully inside

package/plugins/sp/skills/spur-dev/references/cross-cutting.md

@@ -114,7 +114,10 @@ inline` interpret the existing `task-pipeline.yaml` in the host session; they do
 workflow run` and never redirect silently to `agent.default`. Interactive **omit** is
 **host-controlled and non-subprocess**, but no longer guarantees host-context execution for every
 model stage (task 0508): an eligible `agent.run` stage — pure-slash input, non-interactive state,
-native subagent with shared-worktree read/write/shell capability — dispatches **once** to that
+native subagent with shared-worktree read/write/shell capability, and a task above the
+`estimate_hours` dispatch floor (2026-09-15 subagent-dispatch evaluation; the floor and the
+resume-over-re-dispatch rule for worker-role continuation stages are owned by
+[inline-pipeline-driver.md](inline-pipeline-driver.md)) — dispatches **once** to that
 native subagent and joins before the driver continues; any pre-dispatch eligibility failure falls
 back to one host execution, and a failure after dispatch follows the stage's error policy with no
 automatic host replay. Inline resolution (omitted or explicit, 0687 R1/R2) keeps the native-subagent

package/plugins/sp/skills/spur-dev/references/inline-pipeline-driver.md

@@ -236,13 +236,38 @@ Action semantics come from the YAML and the workflow action contract:
    decision, or other operator prompt.
 4. The platform exposes a native subagent that shares the working tree and has read, write, shell,
    and Spur task/run-artifact access.
-
-All four pass → dispatch. Any pre-dispatch failure → execute the stage **once** in the host session.
+5. **Size floor (2026-09-15 subagent-dispatch evaluation).** The task file's frontmatter
+   `estimate_hours` — when present — is **above** the dispatch floor of **1 hour**. A task at or
+   below the floor is cheaper to execute than to delegate: the stage runs host-inline and the run
+   log carries `stage <id> executed inline in session <session-id> (below dispatch floor:
+   estimate_hours <n> <= 1)`. The field is authored at decomposition time (batch item
+   `estimate_hours` → frontmatter) or set via `spur task update <wbs> --estimate-hours <n>`; a task
+   with no `estimate_hours` passes this condition unchanged. The driver reads the frontmatter value
+   directly — never estimates size itself.
+
+All five pass → dispatch. Any pre-dispatch failure → execute the stage **once** in the host session.
 An `agent.run` whose `input` is free-form prose rather than a pure slash command fails condition 2
 and is never dispatch-eligible: the driver executes it in the host session and logs it with the
 existing host-fallback line `stage <id> executed inline in session <session-id>` — it does not
 reformulate the prose into a command, spawn a subagent for it, or silently promote it to dispatch.
-No token estimate, stage-size threshold, model heuristic, or configuration switch is added.
+Beyond the deterministic `estimate_hours` floor in condition 5, no token estimate, model heuristic,
+or configuration switch is added.
+
+**Pre-dispatch permission check (2026-09-15 subagent-dispatch evaluation).** Claude Code exposes no
+dry-run permission API, so the contract is **name-capabilities + fail-fast blocker** — never a
+permission-probing subsystem. Before dispatching a stage, the driver names the stage's required
+capabilities in the dispatch payload: the slash command (field 2), the resolved Spur invocation
+(field 4), and any shell actions the YAML action declares. The delegate is instructed to **return a
+blocker immediately on the first missing permission** rather than stalling — a worker parked at a
+prompt the host cannot see is invisible until join. Record one run-log line per dispatch:
+
+```text
+stage <id> permission precheck: ok | <missing capability>
+```
+
+Read-only investigation fan-out (`/sp:dev-parallel --mode investigation`) dispatches read-only
+worker shapes by construction — see the [pre-dispatch permission
+rule](../../parallel-execution/references/fan-out-patterns.md).
 
 **Dispatch and join:** before dispatch, capture the same pre-action git snapshot used by
 `requireDiff` enforcement, and resolve `answerFile`/`expectFile` against the worktree root — the
@@ -251,7 +276,7 @@ to write and what post-join validation reads. Resolving once at the dispatch bou
 surface at once; a relative path would resolve against whatever cwd the writer process happens to
 have.
 
-**Dispatch payload (task 0818 R2).** Send exactly these five fields. The earlier "send only the
+**Dispatch payload (task 0818 R2).** Send exactly these six fields. The earlier "send only the
 stage id, the slash command, and the no-recursion notice" restriction is **deliberately replaced**:
 the execution-tree cwd, the Spur invocation, and the output path are all already resolved at this
 boundary, and a delegate left to re-derive them re-derives them against its own cwd and PATH.
@@ -268,6 +293,17 @@ boundary, and a delegate left to re-derive them re-derives them against its own
    `--spur-bin` flag rather than a new mechanism.
 5. The **resolved absolute output path** (`answerFile`/`expectFile`, resolved as above) and the
    **owning stage's artifact contract** — for a verify stage, the compact contract below.
+6. **The implement-stage acceptance-evidence requirement** — for the implement stage and any stage
+   whose YAML action declares `requireDiff`. A `requireDiff` stage is the one whose delegate writes
+   the deliverable, so its handoff carries three components:
+   (a) **the task's AC identities verbatim** — the exact `Scenario:` titles / checklist text read by
+   the driver from the task file, never paraphrased;
+   (b) **required evidence** — the pasted output of the narrow targeted tests the delegate ran, and
+   a `file:line` change map written into the task's `## Solution` section;
+   (c) the reminder that **a delegate success message is not evidence** — post-join validation reads
+   the artifacts and the diff, not the delegate's claims.
+   Component (a) is read from the task by the driver; this extends the payload contract and is NOT a
+   new YAML key.
 
 Nothing else: no task/session transcripts, no machine-specific session paths. The WBS/path already
 carried by the slash command remains the task handoff.
@@ -313,6 +349,26 @@ before the subagent starts, log the reason and use host fallback. If a started s
 leaves invalid artifacts, do **not** replay the stage in the host — follow the YAML error policy so
 partial mutations are not duplicated.
 
+**Resume over re-dispatch for worker-role continuation (2026-09-15 subagent-dispatch evaluation).**
+Some stages continue the *same* task's work product after a finding: `test-fix` after a failing
+`test`, an implement rework after review findings. A cold re-dispatch makes the next worker
+re-ingest the task, the diff, and every skill file it already loaded. When the host platform
+supports addressing a completed subagent again (Claude Code: send a follow-up message to the same
+agent — its context survives completion), the driver SHOULD resume the prior same-task worker
+subagent for that continuation stage instead of dispatching a fresh one, carrying the same
+six-field payload plus the finding that triggered the continuation. Provenance uses the resumed
+form:
+
+```text
+stage <id> executed via subagent <agent-id> (resumed; host session <session-id>)
+```
+
+The resume rule applies to **worker-role** stages only (implement, test-fix). Reviewer and verify
+stages always dispatch **fresh**: their value is independent eyes on the diff, and a resumed worker
+grading its own work defeats the stage. A continuation stage also dispatches fresh when no prior
+same-task subagent exists (the earlier stage ran host-inline or below the dispatch floor), when a
+host-owned gate sat between the stages, or when the platform cannot address completed subagents.
+
 **Timeout boundary (task 0727):** a dispatched subagent is governed by
 **the host platform's subagent limit, not the YAML timeoutMs** — `timeoutMs` stays not-applicable
 for host execution only — and before dispatch the driver must

package/schemas/task-batch.schema.json

@@ -58,6 +58,11 @@
                 "type": "string",
                 "enum": ["feature-impl", "issue", "review", "meta"],
                 "description": "Template variant for section structure (default: the standard template)."
+            },
+            "estimate_hours": {
+                "type": "number",
+                "exclusiveMinimum": 0,
+                "description": "Size estimate in hours — persisted to frontmatter; the inline pipeline driver's dispatch floor reads it (at/below floor → host-inline, no subagent dispatch)."
             }
         },
         "additionalProperties": false