chore: prepare packages for npm registry publishing

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -16,9 +16,9 @@ body:
       options:
         - CLI (`opfor run` / `opfor setup`)
         - CLI — Autonomous mode (`opfor hunt`)
-        - MCP Server (`@agent-opfor/mcp`)
+        - MCP Server (`@keyvaluesystems/agent-opfor-mcp`)
         - Browser Extension
-        - SDK (`@agent-opfor/sdk`)
+        - SDK (`@keyvaluesystems/agent-opfor-sdk`)
         - Evaluators / Suites
         - Judge (LLM-as-judge)
         - Report (HTML / JSON output)

.github/workflows/ci.yml

@@ -89,7 +89,7 @@ jobs:
       - run: npm run build
 
       - name: Core tests
-        run: npm test --workspace=@agent-opfor/core
+        run: npm test --workspace=@keyvaluesystems/agent-opfor-core
 
       - name: SDK tests
-        run: npm test --workspace=@agent-opfor/sdk
+        run: npm test --workspace=@keyvaluesystems/agent-opfor-sdk

.github/workflows/release.yml

@@ -18,3 +18,5 @@ jobs:
       - uses: googleapis/release-please-action@v4
         with:
           token: ${{ secrets.GITHUB_TOKEN }}
+          config-file: release-please-config.json
+          manifest-file: .release-please-manifest.json

.gitignore

@@ -23,6 +23,14 @@ cli/node_modules/
 .opfor
 tracedata.json
 
+# Vendored into core/ at pack time by its prepack script (see core/package.json).
+# Cleaned up by postpack; ignored here so an interrupted pack can't commit copies.
+core/evaluators/
+core/suites/
+core/skills/
+core/data/
+core/atlas-data/
+
 # Pre-refactor local trees / accidental rebuilds (canonical code is cli/, core/, mcp/)
 /src/
 src/agent/**/dist/

AGENTS.md

@@ -18,15 +18,15 @@ Opfor is an open-source red-teaming toolkit for AI agents and MCP servers. It ge
 | Browser extension | Click the toolbar icon on any chat UI                                                          | Non-developers — QA, PMs, security analysts  |
 | MCP server        | `opfor_setup`, `opfor_run` tools                                                               | MCP-compatible host (Cursor, Claude Desktop) |
 | Skills            | `/opfor-setup`, `/opfor-run` slash commands                                                    | AI coding agent reads markdown skill files   |
-| SDK               | `import { run, hunt } from "@agent-opfor/sdk"`                                                 | Developers embedding opfor in their own code |
+| SDK               | `import { run, hunt } from "@keyvaluesystems/agent-opfor-sdk"`                                 | Developers embedding opfor in their own code |
 
 ---
 
 ## Monorepo structure
 
-```
+```text
 opfor/
-├── core/                          # @agent-opfor/core — shared engine (npm workspace, compiled to core/dist/)
+├── core/                          # @keyvaluesystems/agent-opfor-core — shared engine (npm workspace, compiled to core/dist/)
 │   └── src/
 │       ├── autonomous/            # Autonomous red-teaming orchestration (orchestrator, prompts, tools, state, report, knowledge)
 │       ├── catalog/               # discoverEvaluators.ts, loadCatalog.ts — YAML evaluator/suite discovery
@@ -46,7 +46,7 @@ opfor/
 │       ├── prompts/               # Inlined system prompts (attacker, judge) used by core
 │       └── util/                  # yamlFrontmatter.ts and other utility helpers
 ├── runners/
-│   ├── cli/                       # @agent-opfor/cli — `opfor` CLI binary (npm workspace)
+│   ├── cli/                       # @keyvaluesystems/agent-opfor-cli — `opfor` CLI binary (npm workspace)
 │   │   └── src/
 │   │       ├── index.ts           # CLI entrypoint (commander) — registers setup, run, and hunt
 │   │       ├── commands/
@@ -55,18 +55,18 @@ opfor/
 │   │       │   └── hunt.ts        # `opfor hunt` — autonomous red-teaming with agentic orchestration
 │   │       └── lib/
 │   │           └── artifacts.ts   # .opfor/configs/ + .opfor/reports/ path helpers
-│   ├── mcp/                       # @agent-opfor/mcp — MCP server runner (npm workspace)
+│   ├── mcp/                       # @keyvaluesystems/agent-opfor-mcp — MCP server runner (npm workspace)
 │   │   └── src/
 │   │       └── index.ts           # MCP server entrypoint — registers tools, stdio transport
-│   ├── sdk/                       # @agent-opfor/sdk — programmatic SDK (npm workspace)
+│   ├── sdk/                       # @keyvaluesystems/agent-opfor-sdk — programmatic SDK (npm workspace)
 │   │   └── src/
 │   │       └── index.ts           # SDK entrypoint
-│   └── extension/                 # @agent-opfor/extension — Chrome MV3 browser extension (npm workspace)
+│   └── extension/                 # @keyvaluesystems/agent-opfor-extension — Chrome MV3 browser extension (npm workspace)
 │       ├── service_worker.js      # Entry point — message routing only; imports modules below
 │       ├── orchestrator.js        # Main run loop: locate → attack → extract → reset → judge (calls runAllBrowser from bundled core)
 │       ├── llmUiActions.js        # DOM-specific LLM helpers (input picker, UI planner, message shortener)
 │       ├── domTarget.js           # Adapter exposing the DOM send/extract path as a core AgentTarget
-│       ├── dist/core.bundle.js    # esbuild bundle of @agent-opfor/core/browser (attack + judge engine)
+│       ├── dist/core.bundle.js    # esbuild bundle of @keyvaluesystems/agent-opfor-core/browser (attack + judge engine)
 │       ├── frameDiscovery.js      # Frame collection, scoring, chat-frame selection
 │       ├── domActions.js          # chrome.scripting wrappers (send, click, verify, vendor APIs)
 │       ├── responseExtractor.js   # Three-phase polling extractor for bot responses
@@ -135,7 +135,7 @@ opfor/
 │   ├── mcp.md                     # MCP server (runner) setup + tools reference
 │   ├── browser-extension.md       # Browser extension guide
 │   ├── skills.md                  # Skill bundle usage
-│   ├── sdk.md                     # SDK (@agent-opfor/sdk) reference
+│   ├── sdk.md                     # SDK (@keyvaluesystems/agent-opfor-sdk) reference
 │   ├── evaluators.md              # Evaluator + suite reference
 │   ├── evaluator-schema.md        # Evaluator YAML schema
 │   └── telemetry.md               # Trace-aware testing (Langfuse / Netra)
@@ -158,7 +158,7 @@ npm run format:check              # prettier --check
 npm test                          # vitest in core/
 ```
 
-`core` must compile before any runner — `runners/{cli,mcp}` import from `core/dist/`, and `runners/extension` esbuild-bundles `@agent-opfor/core/browser` at build time. Always run `npm run build` from the repo root, never per-package.
+`core` must compile before any runner — `runners/{cli,mcp}` import from `core/dist/`, and `runners/extension` esbuild-bundles `@keyvaluesystems/agent-opfor-core/browser` at build time. Always run `npm run build` from the repo root, never per-package.
 
 ---
 
@@ -197,7 +197,7 @@ npm test                          # vitest in core/
 | `runners/extension/service_worker.js`    | Extension entry point — message routing; imports from focused ES modules                                                                                                    |
 | `runners/extension/orchestrator.js`      | Full adaptive run loop — drives `runAllBrowser` against `DomTarget`                                                                                                         |
 | `runners/extension/domTarget.js`         | Implements the core `AgentTarget` interface against the live chat DOM                                                                                                       |
-| `runners/extension/dist/core.bundle.js`  | esbuild bundle of `@agent-opfor/core/browser`; supplies `runAllBrowser` + `generateNextTurn` + judge                                                                        |
+| `runners/extension/dist/core.bundle.js`  | esbuild bundle of `@keyvaluesystems/agent-opfor-core/browser`; supplies `runAllBrowser` + `generateNextTurn` + judge                                                        |
 
 ---
 
@@ -342,7 +342,7 @@ For a new MCP _transport_ (beyond stdio/url):
 - **No barrel re-exports** — import directly from the file that owns the symbol
 - **Error messages are actionable** — tell the user what to fix, not just what went wrong
 - **Evaluator files are data** — no business logic in `.md` files; logic lives in `core/src/evaluators/`
-- **Never invoke the CLI as a subprocess from the MCP server** — call `@agent-opfor/core` directly
+- **Never invoke the CLI as a subprocess from the MCP server** — call `@keyvaluesystems/agent-opfor-core` directly
 
 ---
 

CONTRIBUTING.md

@@ -456,7 +456,7 @@ Only submit findings for systems you are authorized to test, or where you have c
 ### Project structure
 
 ```
-core/                       ← @agent-opfor/core — shared engine (npm workspace)
+core/                       ← @keyvaluesystems/agent-opfor-core — shared engine (npm workspace)
   src/
     autonomous/             ← autonomous red-teaming orchestration (opfor hunt)
     catalog/                ← evaluator/suite discovery + loading
@@ -473,9 +473,9 @@ core/                       ← @agent-opfor/core — shared engine (npm workspa
     targets/                ← agent + MCP target adapters
     telemetry/              ← Langfuse + Netra adapters
 runners/
-  cli/                      ← @agent-opfor/cli — `opfor setup`, `opfor run`, `opfor hunt`
-  mcp/                      ← @agent-opfor/mcp — opfor as an MCP server
-  sdk/                      ← @agent-opfor/sdk — programmatic SDK for embedding
+  cli/                      ← @keyvaluesystems/agent-opfor-cli — `opfor setup`, `opfor run`, `opfor hunt`
+  mcp/                      ← @keyvaluesystems/agent-opfor-mcp — opfor as an MCP server
+  sdk/                      ← @keyvaluesystems/agent-opfor-sdk — programmatic SDK for embedding
   extension/                ← Chrome MV3 browser extension
 evaluators/
   agent/                      ← agent evaluators by category

README.md

@@ -39,7 +39,7 @@ Apache 2.0. Built from India.
 ## Quick Start
 
 ```bash
-npm install -g @agent-opfor/cli
+npm install -g @keyvaluesystems/agent-opfor-cli
 export OPENAI_API_KEY=your-key    # or GEMINI_API_KEY, ANTHROPIC_API_KEY, etc.
 ```
 
@@ -75,13 +75,13 @@ Most red-team tooling in this space is excellent at one thing — a probe librar
 
 Different people on your team need different entry points. Opfor ships five.
 
-| Mode                     | How                                                                     | Best for                                                                                  |
-| ------------------------ | ----------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- |
-| 🖥️ **CLI**               | `opfor setup` → `opfor run`                                             | Engineers, CI/CD, terminal-first workflows                                                |
-| 🌐 **Browser extension** | Install the extension, click the icon on any chat interface             | Product managers, designers, QA, security analysts — anyone who can't or won't write code |
-| 🤖 **MCP server**        | Register opfor in Cursor or Claude Desktop, then ask in chat            | AI coding agents that test your other agents                                              |
-| ⚡ **Skills**            | `/opfor-setup` · `/opfor-run` · `/opfor-mcp-setup` · `/opfor-mcp-run`   | Developers who want one-command testing inside their IDE                                  |
-| 📦 **SDK**               | `npm install @agent-opfor/sdk`, then call `run` / `hunt` from your code | Programmatic red-teaming and custom workflows                                             |
+| Mode                     | How                                                                                     | Best for                                                                                  |
+| ------------------------ | --------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------- |
+| 🖥️ **CLI**               | `opfor setup` → `opfor run`                                                             | Engineers, CI/CD, terminal-first workflows                                                |
+| 🌐 **Browser extension** | Install the extension, click the icon on any chat interface                             | Product managers, designers, QA, security analysts — anyone who can't or won't write code |
+| 🤖 **MCP server**        | Register opfor in Cursor or Claude Desktop, then ask in chat                            | AI coding agents that test your other agents                                              |
+| ⚡ **Skills**            | `/opfor-setup` · `/opfor-run` · `/opfor-mcp-setup` · `/opfor-mcp-run`                   | Developers who want one-command testing inside their IDE                                  |
+| 📦 **SDK**               | `npm install @keyvaluesystems/agent-opfor-sdk`, then call `run` / `hunt` from your code | Programmatic red-teaming and custom workflows                                             |
 
 All five share the same evaluators, attack templates, and judge logic.
 
@@ -160,10 +160,10 @@ This is the path for the half of every product team that doesn't open a terminal
 
 ## SDK — embed red-teaming in your code
 
-The SDK is opfor's programmatic path. Install `@agent-opfor/sdk`, call `run` or `hunt`, and get structured results back — no CLI, no config files, no subprocess.
+The SDK is opfor's programmatic path. Install `@keyvaluesystems/agent-opfor-sdk`, call `run` or `hunt`, and get structured results back — no CLI, no config files, no subprocess.
 
 ```typescript
-import { Opfor } from "@agent-opfor/sdk";
+import { Opfor } from "@keyvaluesystems/agent-opfor-sdk";
 
 const opfor = new Opfor({ apiKey: process.env.ANTHROPIC_API_KEY });
 

core/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "@agent-opfor/core",
+  "name": "@keyvaluesystems/agent-opfor-core",
   "version": "0.9.0",
   "description": "Opfor core engine — attacker prompt generation, judge, and execution shared by all runners",
   "license": "Apache-2.0",

core/src/autonomous/index.ts

@@ -1,8 +1,8 @@
 /**
  * Autonomous red-team engine — public API.
  *
- * Consumers import from "@agent-opfor/core/autonomous/index.js" (or sub-paths
- * like "@agent-opfor/core/autonomous/report/types.js" for deeper access).
+ * Consumers import from "@keyvaluesystems/agent-opfor-core/autonomous/index.js" (or sub-paths
+ * like "@keyvaluesystems/agent-opfor-core/autonomous/report/types.js" for deeper access).
  */
 
 export { runAutonomous } from "./orchestrator/run.js";

core/src/autonomous/knowledge/load.ts

@@ -2,6 +2,7 @@
 // Uses core's shared frontmatter parser. Resolves the bundled `data/`
 // directory relative to this module so it works regardless of the caller's cwd.
 
+import { existsSync } from "node:fs";
 import { readFile, readdir, writeFile, mkdir } from "node:fs/promises";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
@@ -12,13 +13,18 @@ import type { KnowledgeBase, VulnClass, Persona, Strategy, KnowledgeKind } from
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 
 /**
- * Resolve the seed data directory. At runtime this module lives in
- * `core/dist/autonomous/knowledge/load.js`, so the bundled seeds are at
- * `../../../../runners/cli/data` (the CLI's `data/` dir shipped via
- * package.json `files`). Callers should always pass `seedDir`
- * explicitly; this fallback is a best-effort for in-repo dev use.
+ * Resolve the seed data directory. Two layouts are supported:
+ * - Published package: `<core>/data` (vendored into the tarball at pack time). This module
+ *   lives at `core/dist/autonomous/knowledge/load.js`, so the package root is 3 levels up.
+ * - Monorepo dev: `<repo>/runners/cli/data` (the source of truth, 4 levels up).
+ *
+ * Callers may still pass `seedDir` explicitly to override.
  */
 function defaultSeedDir(): string {
+  const packageLocal = path.resolve(__dirname, "../../../data");
+  if (existsSync(packageLocal)) {
+    return packageLocal;
+  }
   return path.resolve(__dirname, "../../../../runners/cli/data");
 }
 

core/src/autonomous/lib/types.ts

@@ -1,5 +1,5 @@
 // Shared option/config types for the autonomous red-team runner.
-// This package is fully standalone — it does NOT import from @agent-opfor/core.
+// This package is fully standalone — it does NOT import from @keyvaluesystems/agent-opfor-core.
 
 /** How the target HTTP agent maintains conversation state. */
 export type TargetMode = "stateless" | "stateful";

core/src/browser.ts

@@ -1,4 +1,4 @@
-// Browser-safe entry for @agent-opfor/core. Re-exports only the parts of the engine
+// Browser-safe entry for @keyvaluesystems/agent-opfor-core. Re-exports only the parts of the engine
 // that are reachable from a Chrome MV3 extension or any other non-Node runtime.
 //
 // What's IN: attacker prompt generation, multi-turn escalation, judge logic,

core/src/catalog/loadEvaluatorCatalog.ts

@@ -4,6 +4,7 @@ import { type EvaluatorCategory } from "../config/evaluatorsLayout.js";
 import { discoverEvaluatorFiles, discoverSuiteFiles } from "./discoverEvaluators.js";
 import { resolveStandardsFromFrontmatter } from "../evaluators/standards.js";
 import { loadAtlasTechniqueIdSet } from "../standards/atlas.js";
+import { log } from "../lib/logger.js";
 import type { StandardsMap } from "../evaluators/schema.js";
 
 export interface EvaluatorMeta {
@@ -111,7 +112,18 @@ export async function loadEvaluatorCatalog(category: EvaluatorCategory): Promise
   suites: SuiteMeta[];
 }> {
   const validateAtlas = process.env.OPFOR_VALIDATE_ATLAS !== "0"; // On by default now
-  const atlasTechniqueIds = validateAtlas ? await loadAtlasTechniqueIdSet() : null;
+  let atlasTechniqueIds: Set<string> | null = null;
+  if (validateAtlas) {
+    try {
+      atlasTechniqueIds = await loadAtlasTechniqueIdSet();
+    } catch (e: unknown) {
+      // ATLAS data is an authoring-time validation aid. If it's unavailable at runtime
+      // (e.g. a published install where the data wasn't bundled), skip validation rather
+      // than crash a scan. Set OPFOR_VALIDATE_ATLAS=0 to silence this.
+      const msg = e instanceof Error ? e.message : String(e);
+      log.warn(`Skipping ATLAS technique-id validation: ${msg.split("\n")[0]}`);
+    }
+  }
 
   // Discover and load evaluators
   const discoveredEvaluators = await discoverEvaluatorFiles(category);

core/src/config/evaluatorsLayout.ts

@@ -1,4 +1,4 @@
-import { realpathSync } from "node:fs";
+import { existsSync, realpathSync } from "node:fs";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
 
@@ -7,6 +7,15 @@ const __dirname = path.dirname(realpathSync(fileURLToPath(import.meta.url)));
 export type EvaluatorCategory = "agent" | "mcp";
 
 export function getRepoRoot(): string {
+  // Bundled runner: data dirs at package root, one level up from dist/
+  // Check for evaluators/agent/ (not just evaluators/) to avoid false-matching
+  // core/src/evaluators/ which is a TypeScript source directory, not data.
+  const oneUp = path.resolve(__dirname, "..");
+  if (existsSync(path.join(oneUp, "evaluators", "agent"))) return oneUp;
+  // Compiled core (npm installed): data dirs colocated at core package root, 2 levels up from dist/config/
+  const twoUp = path.resolve(__dirname, "../..");
+  if (existsSync(path.join(twoUp, "evaluators", "agent"))) return twoUp;
+  // Monorepo: data dirs at repo root, 3 levels up from core/dist/config/
   return path.resolve(__dirname, "../../..");
 }
 

core/src/report/types.ts

@@ -5,7 +5,7 @@
 
 import type { JudgeResult } from "../lib/judgeTypes.js";
 
-/** @deprecated Use JudgeResult from @agent-opfor/core/lib/judgeTypes.js directly. */
+/** @deprecated Use JudgeResult from @keyvaluesystems/agent-opfor-core/lib/judgeTypes.js directly. */
 export type ReportJudge = JudgeResult;
 
 export type DetailCard =