diff --git a/apps/hub/src/skills-mount.ts b/apps/hub/src/skills-mount.ts
index 44793076e..e60fe76ce 100644
--- a/apps/hub/src/skills-mount.ts
+++ b/apps/hub/src/skills-mount.ts
@@ -69,7 +69,7 @@ export function mountSkills(deps: {
         const skill = byName.get(name);
         if (skill === undefined) {
           // Pinning a skill the pusher cannot see would advertise a
-          // skill `load_skill` will refuse to fetch at run time. Reject
+          // skill `skills_load` will refuse to fetch at run time. Reject
           // the push instead of shipping an index that lies.
           throw new SkillRegistryError(
             "not_found",
diff --git a/packages/agent-directory/src/agent-workflow.ts b/packages/agent-directory/src/agent-workflow.ts
index d744358ef..5837d13c5 100644
--- a/packages/agent-directory/src/agent-workflow.ts
+++ b/packages/agent-directory/src/agent-workflow.ts
@@ -39,11 +39,11 @@ export const AGENT_DEFINITION_STEP_ID = "agent";
  * The tool package that turns a name in the `<available_skills>` index
  * into an actual skill body at run time. A definition that pins skills
  * must pin this too, or its prompt would tell the model to call a
- * `load_skill` tool that does not exist.
+ * `skills_load` tool that does not exist.
  */
 export const SKILLS_TOOL_PACKAGE_PIN = {
   name: "@corbits/tools-skills",
-  version: "0.0.1",
+  version: "0.0.2",
 } as const;
 
 /**
diff --git a/packages/agent-directory/test/routes.test.ts b/packages/agent-directory/test/routes.test.ts
index 48c0f8608..70c7e0bc0 100644
--- a/packages/agent-directory/test/routes.test.ts
+++ b/packages/agent-directory/test/routes.test.ts
@@ -1402,11 +1402,11 @@ test("a create request indexes its pinned skills into the stored system prompt",
   const prompt = promptFrom(workflowJson);
   expect(prompt.startsWith("You are a careful research assistant.")).toBe(true);
   expect(prompt).toContain("- web-research: What web-research does.");
-  expect(prompt).toContain("load_skill");
-  // The prompt tells the model to call `load_skill`, so the bundle that
+  expect(prompt).toContain("skills_load");
+  // The prompt tells the model to call `skills_load`, so the bundle that
   // provides it must be pinned on the same push.
   expect(pinsFrom(workflowJson)).toEqual([
-    { name: "@corbits/tools-skills", version: "0.0.1" },
+    { name: "@corbits/tools-skills", version: "0.0.2" },
   ]);
 });
 
diff --git a/packages/mocks/README.md b/packages/mocks/README.md
index 0c7ff4138..53ad1e830 100644
--- a/packages/mocks/README.md
+++ b/packages/mocks/README.md
@@ -142,7 +142,7 @@ ollama.onChat(() => ollama.reply.malformedToolName());
 | `.truncatedToolArgs(name, rawArgs)` | Arguments that are NOT valid JSON — a mid-stream cutoff, reaching the wire byte-for-byte.                  |
 | `.refusal(text?)`                   | The model declines to answer.                                                                              |
 | `.oversized(approxChars?)`          | A large enough text blob to exercise a truncation or blob-spill path.                                      |
-| `.hallucinatedToolName()`           | A plausible-but-nonexistent tool name (`skills_load` instead of the real `load_skill`).                    |
+| `.hallucinatedToolName()`           | A model calling `load_skill` instead of the canonical `skills_load`.                                       |
 
 `sequence([...])` scripts one reply per turn — turn 1 malformed, turn 2
 normal — so CL-6478's "does the room survive?" contract is directly
diff --git a/packages/mocks/src/ollama/scenarios.test.ts b/packages/mocks/src/ollama/scenarios.test.ts
index c37345745..53c21ec3c 100644
--- a/packages/mocks/src/ollama/scenarios.test.ts
+++ b/packages/mocks/src/ollama/scenarios.test.ts
@@ -146,20 +146,20 @@ describe("ollama.reply adversarial scenarios", () => {
     expect(body.choices[0]?.finish_reason).toBe("length");
   });
 
-  test("hallucinatedToolName calls a plausible but nonexistent tool", async () => {
+  test("hallucinatedToolName calls load_skill instead of skills_load", async () => {
     const ollama = createOllamaMock();
     ollama.onChat(() => ollama.reply.hallucinatedToolName());
 
     const response = await chat(ollama.fetch, {
       model: "qwen3.8:27b",
       messages: [{ role: "user", content: "load the git skill" }],
-      tools: [{ type: "function", function: { name: "load_skill" } }],
+      tools: [{ type: "function", function: { name: "skills_load" } }],
     });
     const body = (await response.json()) as ChatCompletionBody;
 
     const name = body.choices[0]?.message.tool_calls?.[0]?.function.name;
-    expect(name).toBe("skills_load");
-    expect(name).not.toBe("load_skill");
+    expect(name).toBe("load_skill");
+    expect(name).not.toBe("skills_load");
   });
 });
 
diff --git a/packages/mocks/src/ollama/scenarios.ts b/packages/mocks/src/ollama/scenarios.ts
index 40d0fd90c..e2012f7a4 100644
--- a/packages/mocks/src/ollama/scenarios.ts
+++ b/packages/mocks/src/ollama/scenarios.ts
@@ -12,9 +12,11 @@ import type { OllamaChatReply } from "./types";
 const CL_6478_MALFORMED_TOOL_NAME =
   "@intx/tools-posix/sidecar-bundle:run_shell\n</parameter";
 
-// Round 1: a model called this instead of the real tool, `load_skill`
-// (packages/tools-skills/src/tool.ts). Plausible-sounding, never declared.
-const HALLUCINATED_TOOL_NAME = "skills_load";
+// A model calling the tool by its old, retired name instead of the
+// canonical `skills_load` (packages/tools-skills/src/tool.ts). `load_skill`
+// no longer exists as a tool; this fixture reproduces a model still
+// guessing it.
+const HALLUCINATED_TOOL_NAME = "load_skill";
 
 function nameOfExactLength(length: number): string {
   const prefix = "namespaced_tool_probe_";
@@ -51,8 +53,8 @@ export type AdversarialReplies = {
   /** A large enough text blob to exercise a truncation or blob-spill path
    * (`finishReason: "length"` -- the model hit its token cap). */
   oversized(approxChars?: number): OllamaChatReply;
-  /** A plausible-but-nonexistent tool name -- round 1 saw a model call
-   * `skills_load` when the real tool is `load_skill`. */
+  /** A plausible-but-nonexistent tool name -- a model calling `load_skill`,
+   * the tool's retired name, when the canonical tool is `skills_load`. */
   hallucinatedToolName(): OllamaChatReply;
 };
 
diff --git a/packages/skills/src/index.ts b/packages/skills/src/index.ts
index fcbe70fd6..4a36fa071 100644
--- a/packages/skills/src/index.ts
+++ b/packages/skills/src/index.ts
@@ -22,7 +22,7 @@ export {
 export {
   AVAILABLE_SKILLS_CLOSE_TAG,
   AVAILABLE_SKILLS_OPEN_TAG,
-  LOAD_SKILL_TOOL,
+  SKILLS_LOAD_TOOL,
   buildAvailableSkillsStanza,
   stripAvailableSkillsStanza,
   withAvailableSkills,
diff --git a/packages/skills/src/prompt.test.ts b/packages/skills/src/prompt.test.ts
index 4f4a48d49..898402c76 100644
--- a/packages/skills/src/prompt.test.ts
+++ b/packages/skills/src/prompt.test.ts
@@ -20,8 +20,8 @@ describe("buildAvailableSkillsStanza", () => {
     expect(stanza).toContain("- summarize: Condenses a long thread.");
   });
 
-  test("names load_skill as the way to read a body", () => {
-    expect(buildAvailableSkillsStanza(PINNED)).toContain("load_skill");
+  test("names skills_load as the way to read a body", () => {
+    expect(buildAvailableSkillsStanza(PINNED)).toContain("skills_load");
   });
 
   test("is empty when nothing is pinned", () => {
diff --git a/packages/skills/src/prompt.ts b/packages/skills/src/prompt.ts
index f875bb38c..4338f8556 100644
--- a/packages/skills/src/prompt.ts
+++ b/packages/skills/src/prompt.ts
@@ -1,7 +1,7 @@
 // The pinned-skills prompt index. A definition that pins skills gets an
 // `<available_skills>` stanza appended to its system prompt at push
 // time — name and description only, plus the instruction to call
-// `load_skill` for the body.
+// `skills_load` for the body.
 //
 // Index-only is the whole point: a skill body is arbitrarily long and
 // most turns need none of it, so inlining every pinned body would burn
@@ -10,7 +10,7 @@
 export const AVAILABLE_SKILLS_OPEN_TAG = "<available_skills>";
 export const AVAILABLE_SKILLS_CLOSE_TAG = "</available_skills>";
 
-export const LOAD_SKILL_TOOL = "load_skill";
+export const SKILLS_LOAD_TOOL = "skills_load";
 
 export type PinnedSkillIndexEntry = {
   readonly name: string;
@@ -31,7 +31,7 @@ export function buildAvailableSkillsStanza(
     AVAILABLE_SKILLS_OPEN_TAG,
     "These skills are available to you. Only their names and descriptions",
     `are listed here. To read a skill's full instructions, call the`,
-    `\`${LOAD_SKILL_TOOL}\` tool with that skill's name — never assume its`,
+    `\`${SKILLS_LOAD_TOOL}\` tool with that skill's name — never assume its`,
     "contents from the description alone.",
     "",
     ...lines,
diff --git a/packages/tool-registry-publish/src/describe.test.ts b/packages/tool-registry-publish/src/describe.test.ts
index 324a9b622..e4e4aafa5 100644
--- a/packages/tool-registry-publish/src/describe.test.ts
+++ b/packages/tool-registry-publish/src/describe.test.ts
@@ -33,4 +33,17 @@ describe("describeCorbitsToolPackages", () => {
     const second = await describeCorbitsToolPackages();
     expect(second).toBe(first);
   });
+
+  test("tools-skills grants exactly one qualifiedId for skills_load, not the old load_skill name", async () => {
+    const descriptions = await describeCorbitsToolPackages();
+    const skills = descriptions.find(
+      (description) => description.name === "@corbits/tools-skills",
+    );
+    expect(skills).toBeDefined();
+    const qualifiedIds = skills?.tools.map((tool) => tool.qualifiedId) ?? [];
+    expect(qualifiedIds).toContain("@corbits/tools-skills/skills:skills_load");
+    expect(qualifiedIds).not.toContain(
+      "@corbits/tools-skills/skills:load_skill",
+    );
+  });
 });
diff --git a/packages/tools-skills/README.md b/packages/tools-skills/README.md
index 1e94ce15c..06dd44567 100644
--- a/packages/tools-skills/README.md
+++ b/packages/tools-skills/README.md
@@ -1,12 +1,12 @@
 # @corbits/tools-skills
 
-`skills_list`, `skills_search`, and `load_skill` as an `@intx/agent` tool
+`skills_list`, `skills_search`, and `skills_load` as an `@intx/agent` tool
 bundle — the agent-facing half of the workbench's skill registry
 (`@corbits/skills`).
 
 A definition's pinned skills already appear in its system prompt's
 `<available_skills>` index, with names and descriptions only.
-`load_skill` is how a model turns one of those names into the actual
+`skills_load` is how a model turns one of those names into the actual
 instructions, so the index stays cheap and only the skills a turn truly
 needs are paid for in context.
 
diff --git a/packages/tools-skills/package.json b/packages/tools-skills/package.json
index 22153c5b0..e2f42ba37 100644
--- a/packages/tools-skills/package.json
+++ b/packages/tools-skills/package.json
@@ -1,8 +1,8 @@
 {
   "name": "@corbits/tools-skills",
   "private": true,
-  "description": "skills_list / skills_search / load_skill as an @intx/agent tool bundle: calls the sanctioned workflow-skills HTTP surface (@corbits/skills) with the sidecar's own bearer token and the run's own address, and fails closed rather than returning an empty registry",
-  "version": "0.0.1",
+  "description": "skills_list / skills_search / skills_load as an @intx/agent tool bundle: calls the sanctioned workflow-skills HTTP surface (@corbits/skills) with the sidecar's own bearer token and the run's own address, and fails closed rather than returning an empty registry",
+  "version": "0.0.2",
   "license": "LGPL-2.1-or-later",
   "type": "module",
   "exports": {
diff --git a/packages/tools-skills/src/index.ts b/packages/tools-skills/src/index.ts
index f96bc5391..b10bb2827 100644
--- a/packages/tools-skills/src/index.ts
+++ b/packages/tools-skills/src/index.ts
@@ -9,8 +9,8 @@ export {
 } from "./client";
 export {
   skillsTools,
-  LOAD_SKILL_TOOL,
   SKILLS_LIST_TOOL,
+  SKILLS_LOAD_TOOL,
   SKILLS_SEARCH_TOOL,
   type WorkflowSkillsToolEnv,
 } from "./tool";
diff --git a/packages/tools-skills/src/tool.test.ts b/packages/tools-skills/src/tool.test.ts
index d67e5e1c0..9c15ec756 100644
--- a/packages/tools-skills/src/tool.test.ts
+++ b/packages/tools-skills/src/tool.test.ts
@@ -2,8 +2,8 @@ import { afterEach, expect, test } from "bun:test";
 import type { ToolCall } from "@intx/types/runtime";
 
 import {
-  LOAD_SKILL_TOOL,
   SKILLS_LIST_TOOL,
+  SKILLS_LOAD_TOOL,
   SKILLS_SEARCH_TOOL,
   skillsTools,
 } from "./tool";
@@ -31,12 +31,12 @@ function stubFetch(respond: () => Response | Promise<Response>): void {
   globalThis.fetch = (async () => respond()) as unknown as typeof fetch;
 }
 
-test("declares skills_list, skills_search, and load_skill", () => {
+test("declares skills_list, skills_search, and skills_load", () => {
   const bundle = skillsTools(testEnv());
   expect(bundle.definitions.map((d) => d.name)).toEqual([
     SKILLS_LIST_TOOL,
     SKILLS_SEARCH_TOOL,
-    LOAD_SKILL_TOOL,
+    SKILLS_LOAD_TOOL,
   ]);
 });
 
@@ -100,7 +100,7 @@ test("skills_search requires a query", async () => {
   expect(String(result.content)).toContain("requires a query");
 });
 
-test("load_skill returns the skill body", async () => {
+test("skills_load returns the skill body", async () => {
   stubFetch(
     () =>
       new Response(
@@ -114,32 +114,41 @@ test("load_skill returns the skill body", async () => {
       ),
   );
   const result = await skillsTools(testEnv()).run(
-    callFor(LOAD_SKILL_TOOL, { name: "triage" }),
+    callFor(SKILLS_LOAD_TOOL, { name: "triage" }),
     new AbortController().signal,
   );
   expect(result.isError).toBeFalsy();
   expect(String(result.content)).toContain("Pick one label.");
 });
 
-test("load_skill surfaces a missing skill as an error rather than inventing one", async () => {
+test("skills_load surfaces a missing skill as an error rather than inventing one", async () => {
   stubFetch(() => new Response("nope", { status: 404 }));
   const result = await skillsTools(testEnv()).run(
-    callFor(LOAD_SKILL_TOOL, { name: "triage" }),
+    callFor(SKILLS_LOAD_TOOL, { name: "triage" }),
     new AbortController().signal,
   );
   expect(result.isError).toBe(true);
   expect(String(result.content)).toContain("404");
 });
 
-test("load_skill requires a name", async () => {
+test("skills_load requires a name", async () => {
   const result = await skillsTools(testEnv()).run(
-    callFor(LOAD_SKILL_TOOL),
+    callFor(SKILLS_LOAD_TOOL),
     new AbortController().signal,
   );
   expect(result.isError).toBe(true);
   expect(String(result.content)).toContain("requires a skill name");
 });
 
+test("load_skill is no longer a recognized tool name", async () => {
+  const result = await skillsTools(testEnv()).run(
+    callFor("load_skill", { name: "triage" }),
+    new AbortController().signal,
+  );
+  expect(result.isError).toBe(true);
+  expect(String(result.content)).toContain("unknown tool");
+});
+
 test("an unknown tool name is an error result, not a throw", async () => {
   const result = await skillsTools(testEnv()).run(
     callFor("skills_delete"),
diff --git a/packages/tools-skills/src/tool.ts b/packages/tools-skills/src/tool.ts
index 339d2c16d..6c94ad5e1 100644
--- a/packages/tools-skills/src/tool.ts
+++ b/packages/tools-skills/src/tool.ts
@@ -1,8 +1,8 @@
 // The `@corbits/tools-skills` bundle: `skills_list`, `skills_search`,
-// and `load_skill` — the agent-facing half of the skill registry. A
+// and `skills_load` — the agent-facing half of the skill registry. A
 // definition's pinned skills are already named in its system prompt's
 // `<available_skills>` index (`@corbits/skills`' `withAvailableSkills`);
-// `load_skill` is how the model turns one of those names into the actual
+// `skills_load` is how the model turns one of those names into the actual
 // instructions, so the index stays cheap and only the skills a turn
 // truly needs are paid for in context.
 //
@@ -19,7 +19,7 @@ import { listSkills, loadSkill, searchSkills } from "./client";
 
 export const SKILLS_LIST_TOOL = "skills_list";
 export const SKILLS_SEARCH_TOOL = "skills_search";
-export const LOAD_SKILL_TOOL = "load_skill";
+export const SKILLS_LOAD_TOOL = "skills_load";
 
 /** Env this bundle needs beyond `BaseEnv`: the run's hub-reach credential. */
 export interface WorkflowSkillsToolEnv extends BaseEnv {
@@ -80,13 +80,13 @@ async function runSkillsSearch(
   }
 }
 
-async function runLoadSkill(
+async function runSkillsLoad(
   env: WorkflowSkillsToolEnv,
   call: ToolCall,
 ): Promise<ToolResult> {
   const name = call.arguments["name"];
   if (typeof name !== "string" || name === "") {
-    return errorResult(call.id, new Error("load_skill requires a skill name"));
+    return errorResult(call.id, new Error("skills_load requires a skill name"));
   }
   try {
     const skill = await loadSkill(clientConfig(env), name);
@@ -102,7 +102,7 @@ export const skillsTools = defineTool<WorkflowSkillsToolEnv>({
   definitions: [
     { name: SKILLS_LIST_TOOL },
     { name: SKILLS_SEARCH_TOOL },
-    { name: LOAD_SKILL_TOOL },
+    { name: SKILLS_LOAD_TOOL },
   ],
   factory: (env) => ({
     definitions: [
@@ -130,7 +130,7 @@ export const skillsTools = defineTool<WorkflowSkillsToolEnv>({
         },
       },
       {
-        name: LOAD_SKILL_TOOL,
+        name: SKILLS_LOAD_TOOL,
         description:
           "Reads one skill's full instructions by name. Call this " +
           "before following a skill listed in <available_skills> — the " +
@@ -154,8 +154,8 @@ export const skillsTools = defineTool<WorkflowSkillsToolEnv>({
           return runSkillsList(env, call);
         case SKILLS_SEARCH_TOOL:
           return runSkillsSearch(env, call);
-        case LOAD_SKILL_TOOL:
-          return runLoadSkill(env, call);
+        case SKILLS_LOAD_TOOL:
+          return runSkillsLoad(env, call);
         default:
           return Promise.resolve(
             errorResult(