feat: use CLI builtin review skill instead of fetching from GitHub

factory-nizar · factory-droid[bot] · factory-nizar · commit 61e6660330b0 · 2026-03-27T15:00:11.000-07:00
The Droid CLI bundles the review skill as a builtin, so the agent can
invoke it via the Skill tool at runtime. Remove the load-skill.ts
GitHub fetch and instead instruct the agent to invoke the 'review'
skill directly.

This eliminates the network dependency during the prepare step and
uses the CLI as the single source of truth for the review methodology.

Co-authored-by: factory-droid[bot] &lt;138933559+factory-droid[bot]@users.noreply.github.com&gt;
diff --git a/src/create-prompt/index.ts b/src/create-prompt/index.ts
@@ -305,7 +305,6 @@ export type PromptCreationOptions = {
   reviewArtifacts?: ReviewArtifacts;
   outputFilePath?: string;
   includeSuggestions?: boolean;
-  reviewSkillContent?: string;
 };
 
 export async function createPrompt({
@@ -321,7 +320,6 @@ export async function createPrompt({
   reviewArtifacts,
   outputFilePath,
   includeSuggestions,
-  reviewSkillContent,
 }: PromptCreationOptions) {
   try {
     const droidCommentId = commentId?.toString();
@@ -342,10 +340,6 @@ export async function createPrompt({
       preparedContext.includeSuggestions = includeSuggestions;
     }
 
-    if (reviewSkillContent) {
-      preparedContext.reviewSkillContent = reviewSkillContent;
-    }
-
     await mkdir(`${process.env.RUNNER_TEMP || "/tmp"}/droid-prompts`, {
       recursive: true,
     });
diff --git a/src/create-prompt/templates/review-candidates-prompt.ts b/src/create-prompt/templates/review-candidates-prompt.ts
@@ -1,4 +1,3 @@
-import { formatSkillSection } from "../../utils/load-skill";
 import type { PreparedContext } from "../types";
 
 export function generateReviewCandidatesPrompt(
@@ -53,7 +52,9 @@ export function generateReviewCandidatesPrompt(
   return `You are a senior staff software engineer and expert code reviewer.
 
 Your task: Review PR #${prNumber} in ${repoFullName} and generate a JSON file with **high-confidence, actionable** review comments that pinpoint genuine issues.
-${formatSkillSection(context.reviewSkillContent)}
+
+Before starting, invoke the 'review' skill to load the review methodology. Follow its bug patterns, analysis discipline, reporting gate, and priority levels throughout your review.
+
 <context>
 Repo: ${repoFullName}
 PR Number: ${prNumber}
diff --git a/src/create-prompt/templates/review-validator-prompt.ts b/src/create-prompt/templates/review-validator-prompt.ts
@@ -1,4 +1,3 @@
-import { formatSkillSection } from "../../utils/load-skill";
 import type { PreparedContext } from "../types";
 
 export function generateReviewValidatorPrompt(
@@ -44,7 +43,9 @@ export function generateReviewValidatorPrompt(
   return `You are validating candidate review comments for PR #${prNumber} in ${repoFullName}.
 
 IMPORTANT: This is Phase 2 (validator) of a two-pass review pipeline.
-${formatSkillSection(context.reviewSkillContent)}
+
+Before starting, invoke the 'review' skill to load the review methodology. Apply its reporting gate, confidence calibration, and deduplication rules.
+
 ### Context
 
 * Repo: ${repoFullName}
diff --git a/src/create-prompt/types.ts b/src/create-prompt/types.ts
@@ -119,5 +119,4 @@ export type PreparedContext = CommonFields & {
   reviewArtifacts?: ReviewArtifacts;
   outputFilePath?: string;
   includeSuggestions?: boolean;
-  reviewSkillContent?: string;
 };
diff --git a/src/entrypoints/generate-review-prompt.ts b/src/entrypoints/generate-review-prompt.ts
@@ -15,7 +15,6 @@ import { prepareMcpTools } from "../mcp/install-mcp-server";
 import { generateReviewCandidatesPrompt } from "../create-prompt/templates/review-candidates-prompt";
 import { generateSecurityReviewPrompt } from "../create-prompt/templates/security-review-prompt";
 import { normalizeDroidArgs, parseAllowedTools } from "../utils/parse-tools";
-import { loadSkill } from "../utils/load-skill";
 
 async function run() {
   try {
@@ -99,9 +98,6 @@ async function run() {
     const outputFilePath = process.env.DROID_OUTPUT_FILE || undefined;
     const includeSuggestions = process.env.INCLUDE_SUGGESTIONS !== "false";
 
-    const reviewSkillContent =
-      reviewType === "code" ? await loadSkill("review") : undefined;
-
     await createPrompt({
       githubContext: context,
       commentId,
@@ -114,7 +110,6 @@ async function run() {
       reviewArtifacts,
       outputFilePath,
       includeSuggestions,
-      reviewSkillContent,
     });
 
     // Set run type
diff --git a/src/tag/commands/review-validator.ts b/src/tag/commands/review-validator.ts
@@ -9,7 +9,6 @@ import { prepareMcpTools } from "../../mcp/install-mcp-server";
 import { normalizeDroidArgs, parseAllowedTools } from "../../utils/parse-tools";
 import type { PrepareResult } from "../../prepare/types";
 import { generateReviewValidatorPrompt } from "../../create-prompt/templates/review-validator-prompt";
-import { loadSkill } from "../../utils/load-skill";
 
 export async function prepareReviewValidatorMode({
   context,
@@ -47,7 +46,6 @@ export async function prepareReviewValidatorMode({
   };
 
   const includeSuggestions = process.env.INCLUDE_SUGGESTIONS !== "false";
-  const reviewSkillContent = await loadSkill("review");
 
   await createPrompt({
     githubContext: context,
@@ -61,7 +59,6 @@ export async function prepareReviewValidatorMode({
     generatePrompt: generateReviewValidatorPrompt,
     reviewArtifacts,
     includeSuggestions,
-    reviewSkillContent,
   });
 
   core.exportVariable("DROID_EXEC_RUN_TYPE", "droid-review");
diff --git a/src/tag/commands/review.ts b/src/tag/commands/review.ts
@@ -9,7 +9,6 @@ import { createInitialComment } from "../../github/operations/comments/create-in
 import { normalizeDroidArgs, parseAllowedTools } from "../../utils/parse-tools";
 import { isEntityContext } from "../../github/context";
 import { generateReviewCandidatesPrompt } from "../../create-prompt/templates/review-candidates-prompt";
-import { loadSkill } from "../../utils/load-skill";
 import type { Octokits } from "../../github/api/client";
 import type { PrepareResult } from "../../prepare/types";
 
@@ -86,7 +85,6 @@ export async function prepareReviewMode({
   });
 
   const includeSuggestions = process.env.INCLUDE_SUGGESTIONS !== "false";
-  const reviewSkillContent = await loadSkill("review");
 
   await createPrompt({
     githubContext: context,
@@ -100,7 +98,6 @@ export async function prepareReviewMode({
     generatePrompt: generateReviewCandidatesPrompt,
     reviewArtifacts,
     includeSuggestions,
-    reviewSkillContent,
   });
   core.exportVariable("DROID_EXEC_RUN_TYPE", "droid-review");
 
diff --git a/src/utils/load-skill.ts b/src/utils/load-skill.ts
diff --git a/test/integration/review-flow.test.ts b/test/integration/review-flow.test.ts
@@ -8,7 +8,6 @@ import * as createInitial from "../../src/github/operations/comments/create-init
 import * as mcpInstaller from "../../src/mcp/install-mcp-server";
 import * as actorValidation from "../../src/github/validation/actor";
 import * as promptModule from "../../src/create-prompt";
-import * as skillLoader from "../../src/utils/load-skill";
 import * as core from "@actions/core";
 import * as childProcess from "node:child_process";
 
@@ -24,7 +23,6 @@ describe("review command integration", () => {
   let exportVarSpy: ReturnType<typeof spyOn>;
   let promptSpy: ReturnType<typeof spyOn>;
   let execSyncSpy: ReturnType<typeof spyOn>;
-  let loadSkillSpy: ReturnType<typeof spyOn>;
 
   beforeEach(async () => {
     tmpDir = await mkdtemp(path.join(os.tmpdir(), "review-int-"));
@@ -41,9 +39,6 @@ describe("review command integration", () => {
     promptSpy = spyOn(promptModule, "createPrompt").mockResolvedValue();
     setOutputSpy = spyOn(core, "setOutput").mockImplementation(() => {});
     exportVarSpy = spyOn(core, "exportVariable").mockImplementation(() => {});
-    loadSkillSpy = spyOn(skillLoader, "loadSkill").mockResolvedValue(
-      "mock skill content",
-    );
 
     execSyncSpy = spyOn(childProcess, "execSync").mockImplementation(((
       cmd: string,
@@ -65,7 +60,6 @@ describe("review command integration", () => {
     setOutputSpy.mockRestore();
     exportVarSpy.mockRestore();
     execSyncSpy.mockRestore();
-    loadSkillSpy.mockRestore();
 
     if (process.env.RUNNER_TEMP) {
       await rm(process.env.RUNNER_TEMP, { recursive: true, force: true });
diff --git a/test/modes/tag/review-command.test.ts b/test/modes/tag/review-command.test.ts
@@ -6,7 +6,6 @@ import { createMockContext } from "../../mockContext";
 import * as promptModule from "../../../src/create-prompt";
 import * as mcpInstaller from "../../../src/mcp/install-mcp-server";
 import * as comments from "../../../src/github/operations/comments/create-initial";
-import * as skillLoader from "../../../src/utils/load-skill";
 import * as childProcess from "child_process";
 import * as fsPromises from "fs/promises";
 
@@ -40,16 +39,12 @@ describe("prepareReviewMode", () => {
   let execSyncSpy: ReturnType<typeof spyOn>;
   let writeFileSpy: ReturnType<typeof spyOn>;
   let mkdirSpy: ReturnType<typeof spyOn>;
-  let loadSkillSpy: ReturnType<typeof spyOn>;
 
   beforeEach(() => {
     process.env.DROID_ARGS = "";
     delete process.env.REVIEW_MODEL;
     process.env.RUNNER_TEMP = "/tmp/test-runner";
 
-    loadSkillSpy = spyOn(skillLoader, "loadSkill").mockResolvedValue(
-      "mock skill content",
-    );
     promptSpy = spyOn(promptModule, "createPrompt").mockResolvedValue();
     mcpSpy = spyOn(mcpInstaller, "prepareMcpTools").mockResolvedValue(
       "mock-config",
@@ -89,7 +84,6 @@ describe("prepareReviewMode", () => {
     execSyncSpy.mockRestore();
     writeFileSpy.mockRestore();
     mkdirSpy.mockRestore();
-    loadSkillSpy.mockRestore();
     process.env.DROID_ARGS = originalArgs;
     if (originalReviewModel !== undefined) {
       process.env.REVIEW_MODEL = originalReviewModel;
diff --git a/test/utils/load-skill.test.ts b/test/utils/load-skill.test.ts
