supipowers 2.0.1 → 2.1.0

package/README.md

@@ -23,7 +23,7 @@ Run the interactive installer:
 bunx supipowers
 ```
 
-The installer detects Pi (`~/.pi`) and OMP (`~/.omp`) — when both are present it offers a multiselect to install to one or both. It registers the extension, removes legacy external context-mode MCP registrations from `agent/mcp.json` and cleans up the old `settings/mcp.json` if present, and can install missing optional tooling such as LSP servers, `mcpc`, and Playwright CLI.
+The installer detects Pi (`~/.pi`) and OMP (`~/.omp`) — when both are present it offers a multiselect to install to one or both. It registers the extension, removes legacy external context-mode MCP registrations from `agent/mcp.json` and cleans up the old `settings/mcp.json` if present, and can install missing optional tooling such as LSP servers and Playwright CLI.
 
 > [!TIP]
 > Run `/supi:update` at any time to upgrade to the latest version, or `/supi:doctor` to check your setup.
@@ -42,7 +42,6 @@ The installer scans for these and offers to install missing tooling where it can
 
 | Dependency                            | What it enables                                                       |
 | ------------------------------------- | --------------------------------------------------------------------- |
-| [mcpc](https://github.com/apify/mcpc) | MCP server management via `/supi:mcp`                                 |
 | `typescript-language-server`          | TypeScript/JS diagnostics and references in review gates              |
 | `pyright`                             | Python type checking                                                  |
 | `rust-analyzer`                       | Rust language server                                                  |
@@ -70,19 +69,22 @@ The installer scans for these and offers to install missing tooling where it can
 | `/supi:model`            | Configure model assignments per action (plan, review, qa…)    |
 | `/supi:context`          | Show current context window usage and system prompt breakdown |
 | `/supi:optimize-context` | Analyze loaded prompt/context usage and suggest reductions    |
-| `/supi:mcp`              | Manage MCP servers (connect, disconnect, migrate)             |
 | `/supi:config`           | Interactive settings TUI                                      |
 | `/supi:status`           | Show project plans and configuration summary                  |
 | `/supi:doctor`           | Diagnose extension health and missing dependencies            |
 | `/supi:generate`        | Documentation drift detection                                |
 | `/supi:update`           | Update supipowers to the latest version                       |
 | `/supi:agents`           | Manage review agents                                          |
+| `/supi:ultraplan`        | Multi-stage authoring pipeline (intake → scout → discover → research → synthesize → review → approve) |
+| `/supi:harness`          | Harness engineering pipeline and anti-slop guardrails         |
+| `/supi:memory`           | Manage native MemPalace memory integration (`status`, `setup`) |
+| `/supi:clear`            | Clear metrics, cache, session knowledge, and memory           |
 
-Most commands steer the AI session. These are TUI-only — they open native dialogs without triggering the AI: `/supi`, `/supi:config`, `/supi:status`, `/supi:review`, `/supi:update`, `/supi:doctor`, `/supi:mcp`, `/supi:model`, `/supi:context`, `/supi:optimize-context`, `/supi:commit`, `/supi:release`, `/supi:checks`, `/supi:agents`.
+Most commands steer the AI session. These are TUI-only — they open native dialogs without triggering the AI: `/supi`, `/supi:config`, `/supi:status`, `/supi:review`, `/supi:update`, `/supi:doctor`, `/supi:model`, `/supi:context`, `/supi:optimize-context`, `/supi:commit`, `/supi:release`, `/supi:checks`, `/supi:agents`, `/supi:ultraplan`, `/supi:harness`, `/supi:memory`, `/supi:clear`.
 
 ## How it works
 
-**Planning.** `/supi:plan` steers the AI through planning phases (scope → decompose → estimate → verify), saves the result to `.omp/supipowers/plans/`, and presents an approval UI. On approval, tasks execute in the same session.
+**Planning.** `/supi:plan` steers the AI through planning phases (Explore → Clarify → Brainstorm → Design & Save → Review Loop → User Gate → Plan), saves the result to `.omp/supipowers/plans/`, and presents an approval UI. On approval, tasks execute in the same session.
 
 **Quality gates.** `/supi:checks` runs deterministic quality gates. Six gates are available: `lsp-diagnostics`, `lint`, `typecheck`, `format`, `test-suite`, and `build`. Each gate can be enabled independently via `/supi:config` or the shared repository config at `.omp/supipowers/config.json`. In monorepos, `/supi:checks` defaults to `All`, which runs the root target plus every workspace target sequentially; use `--target <package>` to narrow the run or `--target all` to request the batch mode explicitly. Gates report issues with severity levels.
 
@@ -123,7 +125,6 @@ Use `/supi:agents` to inspect the merged set that will actually run.
 | Release automation                    | ✅         | ❌               |
 | Commit workflow                       | ✅         | ❌               |
 | Context-window optimizations          | ✅         | ❌               |
-| MCP server management through mcpc    | ✅         | ❌               |
 | Git worktree workflow                 | ❌         | ✅               |
 
 ## Quality gates
@@ -157,6 +158,8 @@ Configuration uses built-in defaults plus two user-managed override layers:
 
 `/supi:config` exposes only `Global` and `Repository`. In monorepos, the repository config is shared across every workspace; there are no per-workspace Supipowers config files for general settings.
 
+MemPalace hook timeouts are configured under `mempalace.timeouts`. Keep `hookMs` at or above `6000` when `mempalace.hooks.autoSearchOnPrompt` is enabled; MemPalace search can now pause before retrying a transient index lookup. The built-in default is `10000`.
+
 ## Release channels
 
 Three built-in channels are available: `github` (GitHub Release via `gh` CLI), `gitlab` (GitLab Release via `glab` CLI), and `gitea` (Gitea Release via `tea` CLI). Channels are selected per-project in `release.channels`.
@@ -213,6 +216,7 @@ Supipowers ships runtime-loaded prompt skills that are also available to the age
 | `release`               | `/supi:release`         |
 | `context-mode`          | Context window guidance |
 | `creating-supi-agents`  | Agent creation guidance  |
+| `harness`               | `/supi:harness`         |
 
 ## Development
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "supipowers",
-  "version": "2.0.1",
+  "version": "2.1.0",
   "description": "Workflow extension for OMP coding agents.",
   "type": "module",
   "scripts": {
@@ -9,9 +9,11 @@
     "test:watch": "bun test --timeout 20000 --watch tests/",
     "test:evals": "bun test --timeout 20000 tests/evals/",
     "build": "tsc -p tsconfig.build.json",
+    "ci": "bun run typecheck && bun run test",
     "install:visual-server": "npm --prefix src/visual/scripts ci --ignore-scripts --no-audit --no-fund",
     "postinstall": "bun run install:visual-server",
-    "prepare": "git config core.hooksPath hooks || true"
+    "prepare": "git config core.hooksPath hooks || true",
+    "dev-install": "bun run bin/dev-install.ts"
   },
   "engines": {
     "bun": ">=1.3.10"

package/skills/harness/SKILL.md

@@ -78,6 +78,7 @@ Scorecard has lenient + strict scores (0–100). Strict counts `wontfix` items a
 - `/supi:harness resolve <id>` — mark an entry resolved.
 - `/supi:harness backlog` — list every open entry.
 - `/supi:harness score` — recompute and display the score.
+- `/supi:harness pr-comment [--dry-run] [--pr=N] [--repo=owner/repo] [--session=<id>] [--mode=every-push|on-status-change]` — render (or post) the sticky PR comment for the latest validate report. See `docs/supipowers/harness/pr-comment.md`.
 
 ## Conventions you MUST follow
 

package/src/bootstrap.ts

@@ -13,9 +13,7 @@ import { registerQaCommand } from "./commands/qa.js";
 import { registerReleaseCommand, handleRelease } from "./commands/release.js";
 import { registerUpdateCommand, handleUpdate } from "./commands/update.js";
 import { registerDoctorCommand, handleDoctor } from "./commands/doctor.js";
-import { registerMcpCommand, handleMcp, handleMcpCli, parseCliArgs } from "./commands/mcp.js";
 import { registerModelCommand, handleModel } from "./commands/model.js";
-import { executeManagerAction } from "./mcp/manager-tool.js";
 import { registerFixPrCommand } from "./commands/fix-pr.js";
 import { registerContextCommand, handleContext } from "./commands/context.js";
 import { registerOptimizeContextCommand, handleOptimizeContext } from "./commands/optimize-context.js";
@@ -30,10 +28,6 @@ import { registerHarnessPipelineTools } from "./harness/tools.js";
 import { registerHarnessHooks } from "./harness/hooks/register.js";
 import { loadConfig } from "./config/loader.js";
 import { registerContextModeHooks } from "./context-mode/hooks.js";
-import { loadMcpRegistry } from "./mcp/config.js";
-import { McpcClient } from "./mcp/mcpc.js";
-import { parseTags } from "./mcp/activation.js";
-import { initializeMcpServers, shutdownMcpServers } from "./mcp/lifecycle.js";
 import { registerPlanApprovalHook } from "./planning/approval-flow.js";
 import { registerPlanningSystemPromptHook } from "./planning/system-prompt.js";
 import { registerPlanningAskTool, registerPlanningAskToolGuard } from "./planning/planning-ask-tool.js";
@@ -60,7 +54,6 @@ const TUI_COMMANDS: Record<string, (platform: Platform, ctx: any, args?: string)
   "supi:review": (platform, ctx, args) => handleAiReview(platform, ctx, args),
   "supi:update": (platform, ctx) => handleUpdate(platform, ctx),
   "supi:doctor": (platform, ctx) => handleDoctor(platform, ctx),
-  "supi:mcp": (platform, ctx) => handleMcp(platform, ctx),
   "supi:model": (platform, ctx) => handleModel(platform, ctx),
   "supi:context": (platform, ctx) => handleContext(platform, ctx),
   "supi:optimize-context": (platform, ctx, args) => handleOptimizeContext(platform, ctx, args),
@@ -74,8 +67,6 @@ const TUI_COMMANDS: Record<string, (platform: Platform, ctx: any, args?: string)
   "supi:memory": (platform, ctx, args) => handleMemory(platform, ctx, args),
 };
 
-let pendingTags: string[] = [];
-
 function getInstalledVersion(platform: Platform): string | null {
   const pkgPath = platform.paths.agent("extensions", "supipowers", "package.json");
   if (!existsSync(pkgPath)) return null;
@@ -99,7 +90,6 @@ export function bootstrap(platform: Platform): void {
   registerUpdateCommand(platform);
   registerFixPrCommand(platform);
   registerDoctorCommand(platform);
-  registerMcpCommand(platform);
   registerModelCommand(platform);
   registerContextCommand(platform);
   registerOptimizeContextCommand(platform);
@@ -131,17 +121,6 @@ export function bootstrap(platform: Platform): void {
   // message submission, so no chat message appears and no "Working..." indicator
   platform.on("input", (event, ctx) => {
     const text = event.text.trim();
-
-    // Scan for $tags
-    const registry = loadMcpRegistry(platform.paths, ctx.cwd);
-    const registeredNames = new Set(Object.keys(registry.servers));
-    if (registeredNames.size > 0) {
-      const tags = parseTags(event.text, registeredNames);
-      if (tags.length > 0) {
-        pendingTags = tags;
-      }
-    }
-
     if (!text.startsWith("/")) return;
 
     const spaceIndex = text.indexOf(" ");
@@ -157,14 +136,7 @@ export function bootstrap(platform: Platform): void {
 
   // Context-mode integration
   const config = loadConfig(platform.paths, process.cwd());
-  registerActiveToolController(platform, config, {
-    loadMcpRegistryForCwd: (cwd: string) => loadMcpRegistry(platform.paths, cwd),
-    consumePendingTags: () => {
-      const tags = pendingTags;
-      pendingTags = [];
-      return tags;
-    },
-  });
+  registerActiveToolController(platform, config);
   registerContextModeHooks(platform, config);
   registerMempalaceTool(platform, config);
   registerMempalaceHooks(platform, config);
@@ -174,9 +146,9 @@ export function bootstrap(platform: Platform): void {
   registerPlanningSystemPromptHook(platform);
   registerUiDesignSystemPromptHook(platform);
 
-  // Register harness anti-slop hooks (gated by per-repo marker file at session start).
-  // Idempotent — safe to call once per process. The hooks themselves no-op on every
-  // event when the marker is missing, so other repos see no behavior change.
+  // Register harness anti-slop hooks only for repos with a harness marker at extension boot.
+  // Registered handlers also check the marker per event, so removing the marker disables
+  // an already-started process without affecting other repos.
   registerHarnessHooks(platform);
 
 
@@ -196,23 +168,6 @@ export function bootstrap(platform: Platform): void {
     // OMP's StatusLine never clears hook statuses on /new, so extensions must do it.
     ctx.ui?.setStatus?.("supi-model", undefined);
 
-    // MCP: always register mcpc_manager tool (agent needs it even with zero servers)
-    if (platform.registerTool) {
-      registerMcpcManagerTool(platform, ctx);
-    }
-
-    // MCP server initialization (only if servers configured)
-    const mcpRegistry = loadMcpRegistry(platform.paths, ctx.cwd);
-    if (Object.keys(mcpRegistry.servers).length > 0) {
-      const mcpClient = new McpcClient((cmd, args, opts) => platform.exec(cmd, args, opts));
-      const installed = await mcpClient.checkInstalled();
-      if (!installed.installed) {
-        ctx.ui.notify("mcpc not installed — MCP servers won't connect. Run /supi:upgrade to install.", "warning");
-      } else {
-        await initializeMcpServers(mcpRegistry, mcpClient);
-      }
-    }
-
     // Check for updates in the background
     const currentVersion = getInstalledVersion(platform);
     if (!currentVersion) return;
@@ -234,90 +189,7 @@ export function bootstrap(platform: Platform): void {
   });
 
   // Session shutdown
-  platform.on("session_shutdown", async (_event, ctx) => {
+  platform.on("session_shutdown", async () => {
     await stopActiveUiDesignSession();
-
-    const mcpConfig = loadConfig(platform.paths, ctx.cwd ?? process.cwd());
-    if (!mcpConfig.mcp?.closeSessionsOnExit) return;
-
-    const registry = loadMcpRegistry(platform.paths, ctx.cwd ?? process.cwd());
-    const mcpClient = new McpcClient((cmd, args, opts) => platform.exec(cmd, args, opts));
-    const names = Object.keys(registry.servers).filter((n) => registry.servers[n].enabled);
-    await shutdownMcpServers(names, mcpClient, true);
-  });
-}
-
-function registerMcpcManagerTool(platform: Platform, ctx: any): void {
-  platform.registerTool!({
-    name: "mcpc_manager",
-    label: "MCP Server Manager",
-    description: "Add, remove, enable, disable, or refresh MCP servers managed by supipowers. Use this when the user asks to install, set up, or manage MCP servers.",
-    promptSnippet: "mcpc_manager — manage MCP servers (add, remove, enable, disable, refresh, login, logout, list, info)",
-    promptGuidelines: [
-      "Use when the user asks to install, add, or set up an MCP server",
-      "Use when the user asks to remove, disable, or manage an MCP server",
-      "Do NOT use for calling MCP tools — use the mcpc_<name> gateway tools instead",
-    ],
-    parameters: {
-      type: "object",
-      properties: {
-        action: { type: "string", enum: ["add", "remove", "enable", "disable", "refresh", "login", "logout", "set-activation", "set-taggable", "list", "info"], description: "Action to perform" },
-        name: { type: "string", description: "Server name (required for all except list/refresh-all)" },
-        url: { type: "string", description: "Server URL (required for add)" },
-        transport: { type: "string", enum: ["http", "stdio"], description: "Transport type" },
-        docsUrl: { type: "string", description: "Documentation URL for richer README generation" },
-        activation: { type: "string", enum: ["always", "contextual", "disabled"], description: "Activation mode" },
-        taggable: { type: "boolean", description: "Whether $name tag activates this server" },
-      },
-      required: ["action"],
-    },
-    async execute(_toolCallId: string, params: any, _signal: any, _onUpdate: any, toolCtx: any) {
-      // First validate via routeManagerAction
-      const result = await executeManagerAction(params, {
-        hasUI: toolCtx.hasUI ?? true,
-        ui: toolCtx.ui ?? {},
-        cwd: toolCtx.cwd ?? ctx.cwd,
-      }, {
-        addServer: () => {},
-        removeServer: () => {},
-        updateServer: () => {},
-      });
-
-      if (result.error) {
-        throw new Error(result.content[0]?.text ?? "Manager action failed");
-      }
-
-      // For actual operations, delegate to handleMcpCli
-      const cliArgs = buildCliArgsFromParams(params);
-      await handleMcpCli(platform, {
-        cwd: toolCtx.cwd ?? ctx.cwd,
-        hasUI: toolCtx.hasUI ?? true,
-        ui: {
-          notify: (msg: string) => {
-            // Collect notifications — they'll be in the tool result
-          },
-          ...toolCtx.ui,
-        },
-      }, cliArgs);
-
-      return {
-        content: result.content,
-        details: { action: params.action, name: params.name },
-      };
-    },
   });
 }
-
-function buildCliArgsFromParams(params: any): ReturnType<typeof parseCliArgs> {
-  return {
-    subcommand: params.action === "set-activation" ? "activation"
-      : params.action === "set-taggable" ? "tag"
-      : params.action,
-    name: params.name,
-    url: params.url,
-    transport: params.transport,
-    docsUrl: params.docsUrl,
-    activation: params.activation,
-    taggable: params.taggable,
-  };
-}

package/src/commands/clear.ts

@@ -269,14 +269,14 @@ function parseArgs(args: string | undefined): ParsedArgs {
   return { scope, dryRun };
 }
 
-export function handleClear(
+export async function handleClear(
   platform: Platform,
   ctx: PlatformContext,
   args?: string,
-): void {
+): Promise<void> {
   if (!ctx.hasUI) return;
 
-  void (async () => {
+  try {
     const { scope, dryRun } = parseArgs(args);
     const store = getMetricsStore();
     const sessionId = getSessionId();
@@ -419,16 +419,16 @@ export function handleClear(
         "error",
       );
     }
-  })().catch((err) => {
+  } catch (err) {
     ctx.ui.notify(`Clear error: ${(err as Error).message}`, "error");
-  });
+  }
 }
 
 export function registerClearCommand(platform: Platform): void {
   platform.registerCommand("supi:clear", {
     description: "Clear metrics, cache, current-session knowledge, and memory for the active session (or `all` for the project)",
     async handler(args: string | undefined, ctx: any) {
-      handleClear(platform, ctx, args);
+      await handleClear(platform, ctx, args);
     },
   });
 }

package/src/commands/release.ts

@@ -51,6 +51,9 @@ import {
 import {
   checkDocDrift,
   buildFixPrompt,
+  loadState as loadDriftState,
+  saveState as saveDriftState,
+  getHeadCommit,
 } from "../docs/drift.js";
 import { runQualityGates } from "../quality/runner.js";
 import { REVIEW_GATE_REGISTRY } from "../quality/review-gates.js";
@@ -492,7 +495,6 @@ export async function handleRelease(platform: Platform, ctx: any, args?: string)
           progress.activate("doc-drift", "Fixing documentation");
           notifyInfo(ctx, "Updating documentation", driftResult.summary);
           const fixPrompt = buildFixPrompt(driftResult.findings);
-          const { loadState: loadDriftState, saveState: saveDriftState, getHeadCommit } = await import("../docs/drift.js");
           const driftHead = await getHeadCommit(platform, repoRoot);
           const driftState = loadDriftState(platform.paths, repoRoot);
           saveDriftState(platform.paths, repoRoot, { ...driftState, lastCommit: driftHead, lastRunAt: new Date().toISOString() });

package/src/commands/update.ts

@@ -106,7 +106,7 @@ async function updateSupipowers(
       cpSync(binSource, join(extDir, "bin"), { recursive: true });
     }
     // skills/ must live inside the extension dir — src/commands/agents.ts
-    // uses a static `import from "../../skills/..."` resolved relative to src/.
+    // uses a static skills markdown import resolved relative to src/.
     const skillsDirSource = join(downloadedRoot, "skills");
     if (existsSync(skillsDirSource)) {
       cpSync(skillsDirSource, join(extDir, "skills"), { recursive: true });

package/src/config/defaults.ts

@@ -1,5 +1,6 @@
 // src/config/defaults.ts
 import type { SupipowersConfig } from "../types.js";
+import { MEMPALACE_PACKAGE_VERSION } from "../mempalace/upstream-limits.js";
 
 export const DEFAULT_CONFIG: SupipowersConfig = {
   version: "1.0.0",
@@ -39,7 +40,7 @@ export const DEFAULT_CONFIG: SupipowersConfig = {
     lazyTools: {
       enabled: true,
       mode: "balanced",
-      alwaysKeep: ["ctx_execute", "ctx_search", "ctx_open_cached", "mcpc_manager"],
+      alwaysKeep: ["ctx_execute", "ctx_search", "ctx_open_cached"],
       commandAllowlist: {},
       keywordTools: {},
     },
@@ -61,12 +62,9 @@ export const DEFAULT_CONFIG: SupipowersConfig = {
       focusChainCadence: 6,
     },
   },
-  mcp: {
-    closeSessionsOnExit: false,
-  },
   mempalace: {
     enabled: true,
-    packageVersion: "3.3.4",
+    packageVersion: MEMPALACE_PACKAGE_VERSION,
     managedVenvPath: "~/.omp/supipowers/mempalace-venv",
     palacePath: "~/.mempalace/palace",
     defaultWingStrategy: "repo-name",
@@ -87,6 +85,8 @@ export const DEFAULT_CONFIG: SupipowersConfig = {
       diaryChars: 8000,
       autoSearchTokens: 150,
       wakeUpInjectionEvery: 10,
+      autoSearchSimilarityFloor: 0.55,
+      autoSearchBm25Floor: 0.3,
     },
     timeouts: {
       setupMs: 120000,

package/src/config/loader.ts

@@ -349,6 +349,7 @@ function migrateConfig(config: Record<string, unknown>): Record<string, unknown>
   const legacyProfile = typeof migrated.defaultProfile === "string" ? migrated.defaultProfile : null;
   delete migrated.defaultProfile;
   delete migrated.orchestration;
+  delete migrated.mcp;
 
   const qa = asRecord(migrated.qa);
   const legacyTestCommand =

package/src/config/schema.ts

@@ -120,12 +120,6 @@ export const ConfigSchema = Type.Object(
       },
       { additionalProperties: false },
     ),
-    mcp: Type.Object(
-      {
-        closeSessionsOnExit: Type.Boolean(),
-      },
-      { additionalProperties: false },
-    ),
     mempalace: Type.Object(
       {
         enabled: Type.Boolean(),
@@ -158,6 +152,8 @@ export const ConfigSchema = Type.Object(
             diaryChars: Type.Integer({ minimum: 1 }),
             autoSearchTokens: Type.Integer({ minimum: 1 }),
             wakeUpInjectionEvery: Type.Integer({ minimum: 1 }),
+            autoSearchSimilarityFloor: Type.Number({ minimum: 0, maximum: 1 }),
+            autoSearchBm25Floor: Type.Number({ minimum: 0 }),
           },
           { additionalProperties: false },
         ),

package/src/context/analyzer.ts

@@ -59,41 +59,71 @@ export interface ParsedSkill {
 export function parseIndividualSkills(systemPrompt: string): ParsedSkill[] {
   if (!systemPrompt) return [];
 
-  // OMP renders skills as markdown headings under "# Skills":
-  //   ## skill-name
-  //   Description text...
-  // Find the Skills section bounded by the next h1 heading or end of text.
-  const skillsSectionMatch = systemPrompt.match(
-    /^# Skills\n[\s\S]*?\n(?=##\s)/m,
-  );
-  if (!skillsSectionMatch) return [];
-
-  // Extract the region from first ## to the next # (h1) or end of text
-  const sectionStart = skillsSectionMatch.index! + skillsSectionMatch[0].length;
-  const afterSection = systemPrompt.slice(sectionStart);
-  const nextH1 = afterSection.search(/^# [^#]/m);
-  const skillsBody =
-    skillsSectionMatch[0].slice(skillsSectionMatch[0].indexOf("\n## ") + 1) +
-    (nextH1 === -1 ? afterSection : afterSection.slice(0, nextH1));
-
-  // Split on ## headings
-  const skills: ParsedSkill[] = [];
+  // Locate `# Skills` and bound the section by the next h1 heading or end of
+  // text. The legacy bound used `^##\s`, which let the section bleed past
+  // sibling h2 headings (e.g. `## MCP Server Instructions`) into unrelated
+  // content and misidentify them as skills.
+  const headerMatch = systemPrompt.match(/^# Skills\b[^\n]*\n/m);
+  if (!headerMatch) return [];
+  const bodyStart = headerMatch.index! + headerMatch[0].length;
+  const after = systemPrompt.slice(bodyStart);
+  const nextH1 = after.search(/^# [^#]/m);
+  const skillsBody = nextH1 === -1 ? after : after.slice(0, nextH1);
+
+  // Modern OMP (≥14.7) renders skills as a bullet list: "- name: description"
+  // with descriptions that may wrap across multiple lines.
+  const bulletRegex = /^- ([a-zA-Z0-9._-]+):/gm;
+  const bullets: { name: string; index: number }[] = [];
+  let bm: RegExpExecArray | null;
+  while ((bm = bulletRegex.exec(skillsBody)) !== null) {
+    bullets.push({ name: bm[1], index: bm.index });
+  }
+
+  if (bullets.length > 0) {
+    // Defensive upper bound for the last bullet: stop at any inline markdown
+    // heading inside the body. Real OMP prompts already terminate at an h1
+    // boundary, but synthetic / older prompts may not.
+    let bodyEnd = skillsBody.length;
+    const headingScan = /^#{1,6}\s/gm;
+    let hsm: RegExpExecArray | null;
+    while ((hsm = headingScan.exec(skillsBody)) !== null) {
+      if (hsm.index > bullets[bullets.length - 1].index) {
+        bodyEnd = hsm.index;
+        break;
+      }
+    }
+
+    const bulletSkills: ParsedSkill[] = [];
+    for (let i = 0; i < bullets.length; i++) {
+      const start = bullets[i].index;
+      const end = i + 1 < bullets.length ? bullets[i + 1].index : bodyEnd;
+      const content = skillsBody.slice(start, end).trimEnd();
+      bulletSkills.push({
+        name: bullets[i].name,
+        bytes: byteLength(content),
+        tokens: estimateTokens(content),
+        content,
+      });
+    }
+    return bulletSkills;
+  }
+
+  // Legacy / synthetic shape: "## name" h2 sub-headings under `# Skills`.
   const headingRegex = /^## (.+)$/gm;
   const headings: { name: string; index: number }[] = [];
-  let match;
-
-  while ((match = headingRegex.exec(skillsBody)) !== null) {
-    headings.push({ name: match[1].trim(), index: match.index });
+  let hm: RegExpExecArray | null;
+  while ((hm = headingRegex.exec(skillsBody)) !== null) {
+    headings.push({ name: hm[1].trim(), index: hm.index });
   }
 
+  const skills: ParsedSkill[] = [];
   for (let i = 0; i < headings.length; i++) {
     const start = headings[i].index;
     const end = i + 1 < headings.length ? headings[i + 1].index : skillsBody.length;
     const content = skillsBody.slice(start, end).trimEnd();
-    const bytes = byteLength(content);
     skills.push({
       name: headings[i].name,
-      bytes,
+      bytes: byteLength(content),
       tokens: estimateTokens(content),
       content,
     });
@@ -218,26 +248,45 @@ function extractXmlSections(
   sections: PromptSection[],
   consumed: Set<number>,
 ): void {
-  // Project section FIRST (so nested <file> tags inside <project> are consumed)
-  const projMatch = text.match(/<project>([\s\S]*?)<\/project>/);
-  if (projMatch) {
+  // Project section FIRST (so nested <file> tags inside the wrapper are consumed).
+  // Modern OMP uses `<|START_PROJECT|>...<|END_PROJECT|>` pipe markers; older OMP
+  // and synthetic test inputs use the legacy `<project>...</project>` XML form.
+  const projectPatterns: RegExp[] = [
+    /<\|START_PROJECT\|>[\s\S]*?<\|END_PROJECT\|>/,
+    /<project>[\s\S]*?<\/project>/,
+  ];
+  for (const pattern of projectPatterns) {
+    const projMatch = text.match(pattern);
+    if (!projMatch) continue;
     sections.push({
       label: "Project context",
       bytes: byteLength(projMatch[0]),
       content: projMatch[0],
     });
     markConsumed(consumed, projMatch.index!, projMatch.index! + projMatch[0].length);
+    break;
   }
 
-  // Instructions section
-  const instrMatch = text.match(/<instructions>([\s\S]*?)<\/instructions>/);
-  if (instrMatch) {
+  // Environment envelope (OMP ≥14.9.3) — workstation, tool catalog, LSP guidance.
+  const envMatch = text.match(/<\|START_ENV\|>[\s\S]*?<\|END_ENV\|>/);
+  if (envMatch) {
     sections.push({
-      label: "Extension instructions",
-      bytes: byteLength(instrMatch[0]),
-      content: instrMatch[0],
+      label: "Environment",
+      bytes: byteLength(envMatch[0]),
+      content: envMatch[0],
     });
-    markConsumed(consumed, instrMatch.index!, instrMatch.index! + instrMatch[0].length);
+    markConsumed(consumed, envMatch.index!, envMatch.index! + envMatch[0].length);
+  }
+
+  // Contract envelope (OMP ≥14.9.3) — inviolable rules, yielding criteria.
+  const contractMatch = text.match(/<\|START_CONTRACT\|>[\s\S]*?<\|END_CONTRACT\|>/);
+  if (contractMatch) {
+    sections.push({
+      label: "Contract",
+      bytes: byteLength(contractMatch[0]),
+      content: contractMatch[0],
+    });
+    markConsumed(consumed, contractMatch.index!, contractMatch.index! + contractMatch[0].length);
   }
 
   // File sections — skip if already consumed (e.g., nested inside <project>)
@@ -318,6 +367,26 @@ function extractHeadingSections(
     }
   }
 
+  // Skills aggregate (OMP ≥14.7): markdown bullet list under `# Skills`.
+  // The legacy `<skills>` XML form is handled in extractXmlSections; this picks
+  // up the modern markdown shape rendered by OMP runtime. Bounded by the next
+  // h1/h2 heading so we don't swallow MCP instructions / Tools blocks.
+  const skillsHeading = text.match(/^# Skills\b[^\n]*\n/m);
+  if (skillsHeading && !consumed.has(skillsHeading.index!)) {
+    const start = skillsHeading.index!;
+    const afterHeading = text.slice(start + skillsHeading[0].length);
+    const nextHeading = afterHeading.search(/^#{1,2}\s/m);
+    const end = nextHeading === -1
+      ? text.length
+      : start + skillsHeading[0].length + nextHeading;
+    const content = text.slice(start, end);
+    const bulletCount = (content.match(/^- [a-zA-Z0-9._-]+:/gm) || []).length;
+    if (bulletCount > 0) {
+      sections.push({ label: `Skills (${bulletCount})`, bytes: byteLength(content), content });
+      markConsumed(consumed, start, end);
+    }
+  }
+
   // Also recognize bare memory:// blocks without a heading
   if (!sections.some((s) => s.label === "Memory")) {
     const memoryMatch = text.match(/memory:\/\/\S+/);