diff --git a/packages/gittensory-miner/README.md b/packages/gittensory-miner/README.md
index 218c4392bb..e7538a43c1 100644
--- a/packages/gittensory-miner/README.md
+++ b/packages/gittensory-miner/README.md
@@ -47,7 +47,9 @@ package manager, and the build / test / lint / format commands — for Node (npm
 (pip/poetry/pipenv/uv), Rust, Go, Maven, and Gradle. It is pure (injectable `existsSync` / `readFileSync`), never
 throws, and per its acceptance criteria **fails closed** — a repo with no recognized manifest returns
 `{ detected: false, reason }` and a command that can't be inferred without guessing stays `null`, rather than being
-assumed. Detection only; wiring the description into the attempt prompt is the follow-up ([#4786](https://github.com/JSONbored/gittensory/issues/4786)). (#4785)
+assumed. The attempt path consumes it: `buildCodingTaskSpec` appends the real stack summary (and any
+confidently-inferred build/test/lint/format commands) to the coding-agent instructions so validation uses
+the target repo's own tooling rather than assuming LoopOver/gittensory CI ([#4786](https://github.com/JSONbored/gittensory/issues/4786)). (#4785)
 
 The package also includes an append-only governor decision ledger: `initGovernorLedger` / `appendGovernorEvent`
 persist structured allow/deny/throttle/kill-switch outcomes in local SQLite for contributor audit. Insert-only —
diff --git a/packages/gittensory-miner/lib/coding-task-spec.d.ts b/packages/gittensory-miner/lib/coding-task-spec.d.ts
index 711851d2c8..3398e8696d 100644
--- a/packages/gittensory-miner/lib/coding-task-spec.d.ts
+++ b/packages/gittensory-miner/lib/coding-task-spec.d.ts
@@ -1,4 +1,5 @@
 import type { AcceptanceCriteria, FeasibilityGateResult, FeasibilityVerdict, IssueRecord, PullRequestRecord } from "@loopover/engine";
+import type { RepoStackResult } from "./stack-detection.js";
 
 export type CodingTaskIssue = { number: number; title: string; body?: string | null | undefined; labels?: string[] | undefined };
 
@@ -25,6 +26,8 @@ export type CodingTaskSpecInput = {
   context: CodingTaskContext;
   claimLedger: CodingTaskClaimLedger;
   workingDirectory: string;
+  /** Injectable stack detector (#4786); omitted falls back to stack-detection.js's real `detectRepoStack`. */
+  detectRepoStack?: (repoPath: string) => RepoStackResult;
 };
 
 export type CodingTaskSpecResult =
diff --git a/packages/gittensory-miner/lib/coding-task-spec.js b/packages/gittensory-miner/lib/coding-task-spec.js
index 8fd2864d53..e5d2697f88 100644
--- a/packages/gittensory-miner/lib/coding-task-spec.js
+++ b/packages/gittensory-miner/lib/coding-task-spec.js
@@ -10,6 +10,7 @@ import {
   serializeAcceptanceCriteria,
   shouldWriteAcceptanceCriteria,
 } from "@loopover/engine";
+import { detectRepoStack, renderStackSummary } from "./stack-detection.js";
 
 // Coding-task-spec builder (#5132, Wave 3.5 follow-up). The second gap discovered alongside #5132's CLI
 // wiring: `IterateLoopInput.title`/`instructions`/`acceptanceCriteriaPath` had no builder anywhere in this
@@ -24,6 +25,12 @@ import {
 // @loopover/engine (same gap #5145's own header documents for `issueQuality`). This is not a
 // fabrication -- feasibilityInputFromPreStartCheck's OWN documented default for a missing
 // issueQualityStatus/lifecycle is "ready", the same honest-default precedent already established.
+//
+// Target-repo stack detection (#4786 / #4785 follow-up): `detectRepoStack` already returned a structured
+// language/package-manager/command description, but nothing in the attempt path consumed it -- instructions
+// were issue text + an acceptance-criteria path only. This module now appends that real stack summary (and
+// any confidently-inferred validation commands) to the coding-agent prompt so the agent validates against
+// THIS repository's tooling rather than assuming LoopOver/gittensory CI, Codecov, or `npm run test:ci`.
 
 function buildTaskBrief(issue) {
   const body = (issue.body ?? "").trim();
@@ -131,32 +138,78 @@ export function writeAcceptanceCriteriaFile(workingDirectory, acceptanceCriteria
   return { written: true, path };
 }
 
+/**
+ * Prompt guidance derived from a real `detectRepoStack` result (#4786). Lists only commands the detector
+ * confidently inferred -- a `null` command stays omitted rather than guessed -- and always tells the agent
+ * not to assume LoopOver/gittensory's own CI/coverage conventions.
+ *
+ * @param {import("./stack-detection.js").RepoStackResult} stack
+ * @returns {string}
+ */
+function buildValidationGuidance(stack) {
+  const lines = [
+    `Detected target-repo stack: ${renderStackSummary(stack)}`,
+    "",
+    "Validate your change with THIS repository's own build/test/lint tooling from the stack summary above.",
+    "Do not assume LoopOver/gittensory CI conventions, Codecov patch coverage, or `npm run test:ci` unless those commands appear in the detected stack.",
+  ];
+  if (stack?.detected === true) {
+    const commands = [
+      stack.testCommand ? `- test: \`${stack.testCommand}\`` : null,
+      stack.lintCommand ? `- lint: \`${stack.lintCommand}\`` : null,
+      stack.buildCommand ? `- build: \`${stack.buildCommand}\`` : null,
+      stack.formatCommand ? `- format: \`${stack.formatCommand}\`` : null,
+    ].filter((entry) => entry !== null);
+    if (commands.length > 0) {
+      lines.push("", "Run these commands before finishing:", ...commands);
+    } else {
+      lines.push(
+        "",
+        "No build/test/lint/format commands were confidently inferred — discover and use this repo's own tooling rather than guessing.",
+      );
+    }
+  }
+  return lines.join("\n");
+}
+
 /**
  * The coding-agent driver's own prompt text (agent-sdk-driver.ts's header: "forwarded verbatim as the
  * prompt -- the acceptance-criteria document already lives inside the worktree", so this points to it
- * rather than repeating its content).
+ * rather than repeating its content). Also carries the target repo's detected stack + validation commands
+ * (#4786) so the agent does not default to gittensory-specific CI assumptions.
+ *
+ * @param {{ number: number, title: string, body?: string | null }} issue
+ * @param {string} acceptanceCriteriaPath
+ * @param {import("./stack-detection.js").RepoStackResult} stack
  */
-function buildInstructions(issue, acceptanceCriteriaPath) {
+function buildInstructions(issue, acceptanceCriteriaPath, stack) {
   return [
     `Resolve the following GitHub issue in this repository: #${issue.number} -- ${issue.title}`,
     "",
     (issue.body ?? "").trim(),
     "",
     `A structured acceptance-criteria document describing what "done" means for this attempt is at ${acceptanceCriteriaPath} -- read it and ensure your change satisfies every criterion before finishing.`,
+    "",
+    buildValidationGuidance(stack),
   ].join("\n");
 }
 
 /**
- * Full composition: feasibility -> acceptance criteria -> (if authorized) write the file -> instructions.
- * Returns `ready: false` (with the computed feasibility verdict, for the caller to report) when the
- * verdict is `raise`/`avoid` -- the caller should abandon the attempt rather than proceed with no real
- * acceptance-criteria file on disk.
+ * Full composition: feasibility -> acceptance criteria -> (if authorized) write the file -> detect the
+ * target-repo stack (#4786) -> instructions. Returns `ready: false` (with the computed feasibility verdict,
+ * for the caller to report) when the verdict is `raise`/`avoid` -- the caller should abandon the attempt
+ * rather than proceed with no real acceptance-criteria file on disk.
+ *
+ * `detectRepoStack` is injectable so tests can assert both the detected and fail-closed undiscovered stack
+ * branches without depending on real filesystem probes; omitted falls back to stack-detection.js's real
+ * `detectRepoStack` (the production default).
  *
  * @param {{
  *   repoFullName: string, issue: { number: number, title: string, body?: string | null, labels?: string[] },
  *   context: { issues: Array<{ number: number }>, pullRequests: unknown[] },
  *   claimLedger: { listClaims: (filter: { repoFullName: string, status: string }) => Array<{ issueNumber: number }> },
  *   workingDirectory: string,
+ *   detectRepoStack?: (repoPath: string) => import("./stack-detection.js").RepoStackResult,
  * }} input
  * @returns {import("./coding-task-spec.js").CodingTaskSpecResult}
  */
@@ -169,12 +222,18 @@ export function buildCodingTaskSpec(input) {
     return { ready: false, verdict: feasibility.verdict, feasibility };
   }
 
+  // Real target-repo stack (#4786): detected from the prepared worktree's own manifests, not guessed from
+  // gittensory conventions. Fail-closed `{ detected: false }` results still reach the prompt (via
+  // renderStackSummary) so the agent is told detection failed rather than silently defaulting to npm/Codecov.
+  const detect = input.detectRepoStack ?? detectRepoStack;
+  const stack = detect(input.workingDirectory);
+
   return {
     ready: true,
     verdict: feasibility.verdict,
     feasibility,
     acceptanceCriteriaPath: writeResult.path,
-    instructions: buildInstructions(input.issue, writeResult.path),
+    instructions: buildInstructions(input.issue, writeResult.path, stack),
     title: input.issue.title,
     body: input.issue.body ?? undefined,
     labels: input.issue.labels,
diff --git a/test/unit/miner-coding-task-spec.test.ts b/test/unit/miner-coding-task-spec.test.ts
index 6c0db2025c..3f86961152 100644
--- a/test/unit/miner-coding-task-spec.test.ts
+++ b/test/unit/miner-coding-task-spec.test.ts
@@ -258,4 +258,152 @@ describe("buildCodingTaskSpec (#5132)", () => {
     if (!result.ready) throw new Error("expected ready");
     expect(result.body).toBeUndefined();
   });
+
+  it("REGRESSION (#4786): embeds a detected Node stack's real test/build/lint commands in the coding-agent instructions", () => {
+    const dir = tempDir();
+    writeFileSync(
+      join(dir, "package.json"),
+      JSON.stringify({
+        name: "widgets",
+        scripts: { test: "vitest run", build: "tsc -b", lint: "eslint .", format: "prettier --write ." },
+      }),
+    );
+    const target = issue();
+    const result = buildCodingTaskSpec({
+      repoFullName: "acme/widgets",
+      issue: target,
+      context: { issues: [target], pullRequests: [] },
+      claimLedger: claimLedger(),
+      workingDirectory: dir,
+    });
+
+    expect(result.ready).toBe(true);
+    if (!result.ready) throw new Error("expected ready");
+    expect(result.instructions).toContain("Detected target-repo stack:");
+    expect(result.instructions).toMatch(/javascript|typescript/i);
+    expect(result.instructions).toContain("npm");
+    expect(result.instructions).toContain("Do not assume LoopOver/gittensory CI conventions");
+    expect(result.instructions).toContain("- test: `");
+    expect(result.instructions).toContain("- build: `");
+    expect(result.instructions).toContain("- lint: `");
+    expect(result.instructions).toContain("- format: `");
+    expect(result.instructions).not.toContain("No build/test/lint/format commands were confidently inferred");
+  });
+
+  it("REGRESSION (#4786): a fail-closed undetected stack still reaches the prompt (no silent gittensory default)", () => {
+    const dir = tempDir();
+    const target = issue();
+    const result = buildCodingTaskSpec({
+      repoFullName: "acme/widgets",
+      issue: target,
+      context: { issues: [target], pullRequests: [] },
+      claimLedger: claimLedger(),
+      workingDirectory: dir,
+    });
+
+    expect(result.ready).toBe(true);
+    if (!result.ready) throw new Error("expected ready");
+    expect(result.instructions).toContain("Detected target-repo stack: stack not detected:");
+    expect(result.instructions).toContain("Do not assume LoopOver/gittensory CI conventions");
+    expect(result.instructions).not.toContain("Run these commands before finishing:");
+  });
+
+  it("REGRESSION (#4786): a detected stack with no confidently-inferred commands tells the agent not to guess", () => {
+    const dir = tempDir();
+    const target = issue();
+    const result = buildCodingTaskSpec({
+      repoFullName: "acme/widgets",
+      issue: target,
+      context: { issues: [target], pullRequests: [] },
+      claimLedger: claimLedger(),
+      workingDirectory: dir,
+      detectRepoStack: () => ({
+        detected: true,
+        language: "python",
+        packageManager: "pip",
+        buildCommand: null,
+        testCommand: null,
+        lintCommand: null,
+        formatCommand: null,
+        evidence: { manifest: "requirements.txt", lockfile: null },
+      }),
+    });
+
+    expect(result.ready).toBe(true);
+    if (!result.ready) throw new Error("expected ready");
+    expect(result.instructions).toContain("python via pip");
+    expect(result.instructions).toContain("no validation commands detected");
+    expect(result.instructions).toContain("No build/test/lint/format commands were confidently inferred");
+    expect(result.instructions).not.toContain("Run these commands before finishing:");
+  });
+
+  it("REGRESSION (#4786): when input.detectRepoStack is omitted, uses the REAL stack-detection.js default against the worktree", () => {
+    const dir = tempDir();
+    writeFileSync(join(dir, "Cargo.toml"), '[package]\nname = "widgets"\nversion = "0.1.0"\n');
+    const target = issue();
+    const result = buildCodingTaskSpec({
+      repoFullName: "acme/widgets",
+      issue: target,
+      context: { issues: [target], pullRequests: [] },
+      claimLedger: claimLedger(),
+      workingDirectory: dir,
+    });
+
+    expect(result.ready).toBe(true);
+    if (!result.ready) throw new Error("expected ready");
+    expect(result.instructions).toContain("rust via cargo");
+    expect(result.instructions).toContain("- test: `cargo test`");
+    expect(result.instructions).toContain("- build: `cargo build`");
+  });
+
+  it("REGRESSION (#4786): includes only the non-null commands from a partial injected stack (both sides of each command ternary)", () => {
+    const target = issue();
+    const withBuildAndTest = buildCodingTaskSpec({
+      repoFullName: "acme/widgets",
+      issue: target,
+      context: { issues: [target], pullRequests: [] },
+      claimLedger: claimLedger(),
+      workingDirectory: tempDir(),
+      detectRepoStack: () => ({
+        detected: true,
+        language: "go",
+        packageManager: "go",
+        buildCommand: "go build ./...",
+        testCommand: "go test ./...",
+        lintCommand: null,
+        formatCommand: null,
+        evidence: { manifest: "go.mod", lockfile: null },
+      }),
+    });
+    expect(withBuildAndTest.ready).toBe(true);
+    if (!withBuildAndTest.ready) throw new Error("expected ready");
+    expect(withBuildAndTest.instructions).toContain("- test: `go test ./...`");
+    expect(withBuildAndTest.instructions).toContain("- build: `go build ./...`");
+    expect(withBuildAndTest.instructions).not.toContain("- lint:");
+    expect(withBuildAndTest.instructions).not.toContain("- format:");
+
+    const withLintAndFormat = buildCodingTaskSpec({
+      repoFullName: "acme/widgets",
+      issue: target,
+      context: { issues: [target], pullRequests: [] },
+      claimLedger: claimLedger(),
+      workingDirectory: tempDir(),
+      detectRepoStack: () => ({
+        detected: true,
+        language: "javascript",
+        packageManager: "npm",
+        buildCommand: null,
+        testCommand: null,
+        lintCommand: "npm run lint",
+        formatCommand: "npm run format",
+        evidence: { manifest: "package.json", lockfile: null },
+      }),
+    });
+    expect(withLintAndFormat.ready).toBe(true);
+    if (!withLintAndFormat.ready) throw new Error("expected ready");
+    expect(withLintAndFormat.instructions).toContain("- lint: `npm run lint`");
+    expect(withLintAndFormat.instructions).toContain("- format: `npm run format`");
+    expect(withLintAndFormat.instructions).not.toContain("- test:");
+    expect(withLintAndFormat.instructions).not.toContain("- build:");
+  });
 });