refactor: remove redundant context from prompts, delegate to nerve-dev skill

- Remove nerveYaml injection from planner (skill has it)
- Remove sensesDir/nerveRoot from coder and tester (skill has conventions)
- Prompts now just say 'read the skill' instead of inlining knowledge
- BuildSenseGeneratorDeps reduced to { provider, cwd }
- index.ts drops getNerveYaml(), SENSES_DIR, readFileSync

workflows/sense-generator/build.ts

@@ -8,23 +8,19 @@ import type { SenseMeta } from "./moderator.js";
 
 export type BuildSenseGeneratorDeps = {
   provider: LlmProvider;
-  nerveRoot: string;
+  cwd: string;
-  sensesDir: string;
-  nerveYaml: string;
 };
 
 export function buildSenseGenerator({
   provider,
-  nerveRoot,
+  cwd,
-  sensesDir,
-  nerveYaml,
 }: BuildSenseGeneratorDeps): WorkflowDefinition<SenseMeta> {
   return {
     name: "sense-generator",
     roles: {
-      planner: buildPlannerRole({ provider, cwd: nerveRoot, nerveYaml }),
+      planner: buildPlannerRole({ provider, cwd }),
-      coder: buildCoderRole({ provider, cwd: nerveRoot, sensesDir, nerveRoot }),
+      coder: buildCoderRole({ provider, cwd }),
-      tester: buildTesterRole({ provider, sensesDir, nerveRoot }),
+      tester: buildTesterRole({ provider }),
     },
     moderator,
   };

workflows/sense-generator/index.ts

@@ -1,4 +1,3 @@
-import { readFileSync } from "node:fs";
 import { join } from "node:path";
 import { spawnSafe } from "@uncaged/nerve-workflow-utils";
 import { buildSenseGenerator } from "./build.js";
@@ -7,7 +6,6 @@ import { buildSenseGenerator } from "./build.js";
 
 const HOME = process.env.HOME ?? "/home/azureuser";
 const NERVE_ROOT = join(HOME, ".uncaged-nerve");
-const SENSES_DIR = join(NERVE_ROOT, "senses");
 
 // --- Resolve provider ---
 
@@ -28,23 +26,11 @@ if (!apiKey || !baseUrl) {
   throw new Error("Set DASHSCOPE_API_KEY and DASHSCOPE_BASE_URL");
 }
 
-// --- Build context ---
-
-function getNerveYaml(): string {
-  try {
-    return readFileSync(join(NERVE_ROOT, "nerve.yaml"), "utf-8");
-  } catch {
-    return "# nerve.yaml unavailable";
-  }
-}
-
 // --- Wire up ---
 
 const workflow = buildSenseGenerator({
   provider: { apiKey, baseUrl, model },
-  nerveRoot: NERVE_ROOT,
+  cwd: NERVE_ROOT,
-  sensesDir: SENSES_DIR,
-  nerveYaml: getNerveYaml(),
 });
 
 export default workflow;

workflows/sense-generator/roles/coder/index.ts

@@ -11,17 +11,13 @@ export const coderMetaSchema = z.object({
 export type BuildCoderDeps = {
   provider: LlmProvider;
   cwd: string;
-  sensesDir: string;
-  nerveRoot: string;
 };
 
-export function buildCoderRole({ provider, cwd, sensesDir, nerveRoot }: BuildCoderDeps) {
+export function buildCoderRole({ provider, cwd }: BuildCoderDeps) {
   return createCursorRole<CoderMeta>({
     cwd,
     mode: "default",
-    prompt: async (threadId) => coderPrompt({ threadId, sensesDir, nerveRoot }),
+    prompt: async (threadId) => coderPrompt({ threadId }),
     extract: { provider, schema: coderMetaSchema },
   });
 }
-
-

workflows/sense-generator/roles/coder/prompt.ts

@@ -1,16 +1,7 @@
-export function coderPrompt(vars: {
+export function coderPrompt({ threadId }: { threadId: string }): string {
-  threadId: string;
+  return `Read the workflow thread for the planner's sense design: \`nerve thread ${threadId}\`
-  sensesDir: string;
+Read the nerve-dev skill for sense file structure and conventions: \`cat node_modules/@uncaged/nerve-skills/nerve-dev/SKILL.md\`
-  nerveRoot: string;
-}): string {
-  return `Read the workflow thread for the planner's sense design: \`nerve thread ${vars.threadId}\`
 
-Implement the sense. Create exactly:
+Implement the sense following the patterns from existing senses and the skill guide.
-1. The sense directory under ${vars.sensesDir}/<sense-name>/
+Create all required files and update nerve.yaml.`;
-2. index.js — export async function compute(db, _peers), import schema from "./schema.ts"
-3. schema.ts — drizzle-orm/sqlite-core
-4. migrations/0001_init.sql — must match schema.ts
-5. Update ${vars.nerveRoot}/nerve.yaml — add sense config + reflex entry
-
-Follow the patterns from existing senses. Create all files now.`;
 }

workflows/sense-generator/roles/planner/index.ts

@@ -11,16 +11,13 @@ export const plannerMetaSchema = z.object({
 export type BuildPlannerDeps = {
   provider: LlmProvider;
   cwd: string;
-  nerveYaml: string;
 };
 
-export function buildPlannerRole({ provider, cwd, nerveYaml }: BuildPlannerDeps) {
+export function buildPlannerRole({ provider, cwd }: BuildPlannerDeps) {
   return createCursorRole<PlannerMeta>({
     cwd,
     mode: "ask",
-    prompt: async (threadId) => plannerPrompt({ threadId, nerveYaml }),
+    prompt: async (threadId) => plannerPrompt({ threadId }),
     extract: { provider, schema: plannerMetaSchema },
   });
 }
-
-

workflows/sense-generator/roles/planner/prompt.ts

@@ -1,10 +1,9 @@
-export function plannerPrompt({ threadId, nerveYaml }: {
+export function plannerPrompt({ threadId }: { threadId: string }): string {
-  threadId: string;
-  nerveYaml: string;
-}): string {
   return `You are planning a new Nerve sense.
 
 Read the workflow thread for the user's request: \`nerve thread ${threadId}\`
+Read the nerve-dev skill for sense conventions: \`cat node_modules/@uncaged/nerve-skills/nerve-dev/SKILL.md\`
+Also look at existing senses in the \`senses/\` directory for patterns.
 
 Pick a good kebab-case name for this sense. Produce a PLAN (not code) in markdown:
 
@@ -14,14 +13,5 @@ Pick a good kebab-case name for this sense. Produce a PLAN (not code) in markdow
 ### Compute Logic — step-by-step, specific Node.js APIs or shell commands
 ### Trigger Config — group, interval, throttle, timeout
 
-For reference examples of existing senses (schema, compute logic, migrations), read the nerve-dev skill:
-\`cat node_modules/@uncaged/nerve-skills/nerve-dev/SKILL.md\`
-Also look at existing senses in the \`senses/\` directory for patterns.
-
-Current nerve.yaml:
-\`\`\`yaml
-${nerveYaml}
-\`\`\`
-
 Output ONLY the plan. Be precise and implementation-ready.`;
 }

workflows/sense-generator/roles/tester/index.ts

@@ -10,15 +10,11 @@ export const testerMetaSchema = z.object({
 
 export type BuildTesterDeps = {
   provider: LlmProvider;
-  sensesDir: string;
-  nerveRoot: string;
 };
 
-export function buildTesterRole({ provider, sensesDir, nerveRoot }: BuildTesterDeps) {
+export function buildTesterRole({ provider }: BuildTesterDeps) {
   return createHermesRole<TesterMeta>({
-    prompt: async (threadId) => testerPrompt({ threadId, sensesDir, nerveRoot }),
+    prompt: async (threadId) => testerPrompt({ threadId }),
     extract: { provider, schema: testerMetaSchema },
   });
 }
-
-

workflows/sense-generator/roles/tester/prompt.ts

@@ -1,32 +1,16 @@
-export function testerPrompt(vars: {
+export function testerPrompt({ threadId }: { threadId: string }): string {
-  threadId: string;
-  sensesDir: string;
-  nerveRoot: string;
-}): string {
   return `You are testing a newly created Nerve sense end-to-end.
 
-Read the workflow thread for context: \`nerve thread ${vars.threadId}\`
+Read the workflow thread for context: \`nerve thread ${threadId}\`
-The planner named the sense and the coder created the files.
+Read the nerve-dev skill for expected file structure: \`cat node_modules/@uncaged/nerve-skills/nerve-dev/SKILL.md\`
 
 Verify the full lifecycle:
-
+1. Check all required sense files exist
-1. Check files exist under ${vars.sensesDir}/<sense-name>/
+2. Check nerve.yaml has the sense config
-   - index.js, schema.ts, migrations/0001_init.sql
+3. Run \`nerve sense list\` — confirm the sense appears
-   All three must exist.
-
-2. Check ${vars.nerveRoot}/nerve.yaml has the sense config and reflex entry
-   The sense name should appear under \`senses:\` with group, throttle, etc.
-
-3. Run \`nerve sense list\` — confirm the sense appears in the output
-
 4. Run \`nerve sense trigger <sense-name>\` — should complete without error
-
+5. Run \`nerve sense query <sense-name>\` — retry up to 20s until rows appear
-5. Wait a few seconds, then run \`nerve sense query <sense-name>\`
+6. If any step fails, run \`nerve logs\` and include relevant errors
-   Keep retrying (up to ~20 seconds) until it returns at least one row.
-   If it still says "0 rows", that's a failure.
-
-6. If any step fails, run \`nerve logs\` to check for errors and include
-   relevant log lines in your report.
 
 Output a clear summary: what you checked, what passed, what failed, and why.`;
 }