diff --git a/.agents/skills/cross-platform-compatibility/references/platform-specific-dependencies.md b/.agents/skills/cross-platform-compatibility/references/platform-specific-dependencies.md
index f61a9ab4..b39a6343 100644
--- a/.agents/skills/cross-platform-compatibility/references/platform-specific-dependencies.md
+++ b/.agents/skills/cross-platform-compatibility/references/platform-specific-dependencies.md
@@ -30,19 +30,25 @@ export async function loadPlatformModule() {
   }
 }
 
-// Graceful fallback for optional dependencies
-export function useFSEvents() {
+// Graceful fallback for optional dependencies. Uses dynamic `import()` so the
+// example works under ESM (`"type": "module"`) and degrades to native
+// `fs.watch` when neither `fsevents` nor `chokidar` is installed.
+export async function useFSEvents() {
   try {
     // fsevents is macOS only
     if (process.platform === "darwin") {
-      const fsevents = require("fsevents");
-      return fsevents;
+      return await import("fsevents");
     }
   } catch (error) {
     console.warn("fsevents not available, using fallback");
   }
 
-  // Fallback to chokidar or fs.watch
-  return require("chokidar");
+  // Fallback to chokidar, then to native fs.watch as a final safety net.
+  try {
+    return await import("chokidar");
+  } catch {
+    const { watch } = await import("node:fs/promises");
+    return { watch };
+  }
 }
 ```
diff --git a/.agents/skills/cross-platform-compatibility/references/process-management.md b/.agents/skills/cross-platform-compatibility/references/process-management.md
index 1f75476f..9bf4b027 100644
--- a/.agents/skills/cross-platform-compatibility/references/process-management.md
+++ b/.agents/skills/cross-platform-compatibility/references/process-management.md
@@ -14,28 +14,34 @@ const execFileAsync = promisify(execFile);
 
 export class ProcessUtils {
   // Kill process by PID with platform-specific signal
-  static kill(pid: number, signal?: string): void {
+  static async kill(pid: number, signal?: string): Promise<void> {
     if (process.platform === "win32") {
-      // Windows doesn't support signals, use taskkill
-      spawn("taskkill", ["/pid", pid.toString(), "/f", "/t"]);
-    } else {
-      process.kill(pid, signal || "SIGTERM");
+      // Windows doesn't support POSIX signals. Await taskkill so spawn
+      // failures and non-zero exits reject this Promise and remain catchable.
+      await execFileAsync("taskkill", ["/pid", pid.toString(), "/f", "/t"]);
+      return;
     }
+
+    process.kill(pid, signal ?? "SIGTERM");
   }
 
   // Spawn process with platform-specific handling
   static spawnCommand(command: string, args: string[] = []): ChildProcess {
     if (process.platform === "win32") {
-      // Windows requires cmd.exe to run commands
+      // Windows requires cmd.exe to run commands. shell: false (the default)
+      // is mandatory whenever an argv array is used — Node emits DEP0190 and
+      // arguments are concatenated without shell escaping when shell: true is
+      // combined with args, which is an injection vector.
       return spawn("cmd", ["/c", command, ...args], {
         stdio: "inherit",
-        shell: true,
       });
     }
 
+    // shell: false (the default) — args are passed directly to execvp without
+    // shell interpretation. If shell semantics are required (e.g. globbing),
+    // build a properly escaped command string and pass it as a single argv[0].
     return spawn(command, args, {
       stdio: "inherit",
-      shell: true,
     });
   }
 
diff --git a/.agents/skills/cross-platform-compatibility/references/shell-commands.md b/.agents/skills/cross-platform-compatibility/references/shell-commands.md
index 87014e8c..f27cf39b 100644
--- a/.agents/skills/cross-platform-compatibility/references/shell-commands.md
+++ b/.agents/skills/cross-platform-compatibility/references/shell-commands.md
@@ -4,13 +4,18 @@
 
 ```typescript
 // shell-utils.ts
-import { exec } from "child_process";
+import { exec, execFile } from "child_process";
+import { readdir } from "fs/promises";
 import { promisify } from "util";
 
 const execAsync = promisify(exec);
+const execFileAsync = promisify(execFile);
 
 export class ShellUtils {
-  // Execute command with platform-specific handling
+  // WARNING: Runs `command` through a shell. NEVER pass untrusted input here —
+  // shell metacharacters (`$()`, backticks, `;`, `|`, `&`, redirections) will
+  // be interpreted. For any value derived from user input, prefer
+  // `executeSafe` below, which uses execFile and does not invoke a shell.
   static async execute(command: string): Promise<string> {
     try {
       const { stdout, stderr } = await execAsync(command, {
@@ -19,10 +24,17 @@ export class ShellUtils {
       if (stderr) console.error(stderr);
       return stdout.trim();
     } catch (error) {
-      throw new Error(`Command failed: ${error.message}`);
+      throw new Error(`Command failed: ${(error as Error).message}`);
     }
   }
 
+  // Safe execution with argv array — no shell interpretation. Use this for
+  // any path or value that may originate outside the program.
+  static async executeSafe(command: string, args: string[] = []): Promise<string> {
+    const { stdout, stderr } = await execFileAsync(command, args);
+    if (stderr) console.error(stderr);
+    return stdout.trim();
+  }
   // Get platform-specific shell
   static getShell(): string {
     if (process.platform === "win32") {
@@ -31,29 +43,25 @@ export class ShellUtils {
     return process.env.SHELL || "/bin/sh";
   }
 
-  // Platform-specific commands
+  // Prefer runtime APIs for filesystem work. Shell builtins like `dir` must
+  // go through cmd.exe, where caller-supplied paths are shell-parsed unless
+  // explicitly escaped.
   static async listFiles(directory: string): Promise<string> {
-    if (process.platform === "win32") {
-      return this.execute(`dir "${directory}"`);
-    }
-    return this.execute(`ls -la "${directory}"`);
+    const entries = await readdir(directory, { withFileTypes: true });
+    return entries.map((entry) => entry.name).join("\n");
   }
 
   static async clearScreen(): Promise<void> {
-    if (process.platform === "win32") {
-      await this.execute("cls");
-    } else {
-      await this.execute("clear");
-    }
+    process.stdout.write("\x1Bc");
   }
 
   static async openFile(filepath: string): Promise<void> {
     if (process.platform === "win32") {
-      await this.execute(`start "" "${filepath}"`);
+      await execFileAsync("explorer.exe", [filepath]);
     } else if (process.platform === "darwin") {
-      await this.execute(`open "${filepath}"`);
+      await execFileAsync("open", [filepath]);
     } else {
-      await this.execute(`xdg-open "${filepath}"`);
+      await execFileAsync("xdg-open", [filepath]);
     }
   }
 }
diff --git a/.omp/omp-audit-config.json b/.omp/omp-audit-config.json
index 4e1a92ef..9aabfc3b 100644
--- a/.omp/omp-audit-config.json
+++ b/.omp/omp-audit-config.json
@@ -1,6 +1,6 @@
 {
-  "lastAnalyzedVersion": "15.1.3",
-  "supipowersVersionAtLastAudit": "2.2.0",
-  "lastAuditDate": "2026-05-17",
+  "lastAnalyzedVersion": "15.1.7",
+  "supipowersVersionAtLastAudit": "2.2.2",
+  "lastAuditDate": "2026-05-19",
   "auditOutputFile": "OMP_CHANGELOG_AUDIT.md"
 }
diff --git a/OMP_CHANGELOG_AUDIT.md b/OMP_CHANGELOG_AUDIT.md
index 23388c88..e170e585 100644
--- a/OMP_CHANGELOG_AUDIT.md
+++ b/OMP_CHANGELOG_AUDIT.md
@@ -1,195 +1,163 @@
-# OMP Changelog Audit — 15.1.0 → 15.1.3
+# OMP Changelog Audit — 15.1.3 → 15.1.7
 
 | Field | Value |
 | --- | --- |
-| OMP range analyzed | 15.1.0 → 15.1.3 (incl. 15.1.2) |
-| supipowers version | 2.2.0 |
-| Audit date | 2026-05-17 |
-| Previous audit | 15.0.1 → 15.1.0 (2026-05-15) |
+| OMP range analyzed | 15.1.3 → 15.1.7 |
+| supipowers version | 2.2.2 |
+| Audit date | 2026-05-19 |
+| Previous audit | 15.1.0 → 15.1.3 (2026-05-17) |
 
-> Evidence convention: every claim cites `file:line` ranges. Read with `read <path>:N-M` to verify in-place.
+> Evidence convention: every code claim cites `file:line` or `file:line-line` from the inspected source.
 
-> Scope: every breaking item in 15.1.2 and 15.1.3 was probed against `src/`, `tests/`, shipped `skills/`, shipped `.omp/rules/`, and shipped `.omp/skills/`. Auto-indexed batch search returned 0 in-repo hits for `pi://`, `PiProtocolHandler`, `StringEnum`, `*** Cell`, `*** End`, `*** Abort`, `pty: true`, and the SSH/YieldTool surfaces. All "no-impact" verdicts below cite the actual grep commands and their negative results.
+## Audit scope
 
----
+Primary OMP API surfaces inspected:
 
-## 1. Breaking changes that affect supipowers
+- Central adapter and type contract: `src/platform/omp.ts:48-150`, `src/platform/types.ts:45-160`.
+- Agent-session call paths: `src/ai/final-message.ts:96-136`, `src/context-mode/hooks.ts:251-298`, `src/ultraplan/execution/session-runner.ts:157-167`, `src/git/commit.ts:401-418`.
+- Message handoff call paths: `src/commands/qa.ts:398-405`, `src/commands/fix-pr.ts:521-528`, `src/commands/ultraplan.ts:1458-1465`, `src/harness/hooks/post-session-sweep.ts:111-119`, `src/ui-design/session.ts:754-799`.
+- Tool registration call paths: `src/planning/planning-ask-tool.ts:24-55`, `src/context-mode/tools.ts:453-500`, `src/harness/tools.ts:124-133`.
+- Exec call paths and wrappers: `src/platform/omp.ts:11-17`, `src/platform/omp.ts:56-57`, `src/utils/exec-cli.ts:7-30`, `src/commands/ultraplan.ts:710-729`, `src/fix-pr/fetch-comments.ts:111-129`, `src/fix-pr/fetch-comments.ts:220-246`.
+- Planning approval and ask isolation: `src/planning/system-prompt.ts:223-241`, `src/planning/approval-flow.ts:314-331`, `src/planning/approval-flow.ts:431-463`, `src/storage/plans.ts:12-24`.
 
-### 1.1 [NONE] `pi://` → `omp://` URL scheme + `OmpProtocolHandler` rename — no in-repo references
+## Breaking Changes
 
-**Changelog source (15.1.3):**
-> Renamed the embedded-documentation internal URL scheme from `pi://` to `omp://`. `OmpProtocolHandler` replaces `PiProtocolHandler`; update any external references accordingly.
+### Verdict: no hard breaking changes found
 
-**Evidence collected.** Searched the full repo (excluding `node_modules`, `.git`, `dist`, `omp_source/` upstream tree, and runtime data under `.omp/supipowers/`) for:
+No OMP 15.1.4–15.1.7 changelog entry removes or renames an API that supipowers 2.2.2 directly imports or calls. Two items are worth tracking as compatibility risks because they sit near critical workflows, but neither requires an immediate migration.
 
-- `grep -rn 'pi://' src tests skills global-config docs` → **0 matches**
-- `grep -rn 'PiProtocolHandler' src tests` → **0 matches**
-- `grep -rln 'pi://' .omp/rules .omp/skills` → **0 matches**
+### Compatibility risk: `/supi:plan` native `resolve` guard is narrower than the 15.1.6 malformed-title fix
 
-Stray hits all lived inside the upstream-vendored tree `omp_source/packages/coding-agent/` (`internal-urls/pi-protocol.ts` etc.), which we do not author or ship. The previous audit's reference to the embedded-doc scheme was hypothetical; the codebase has never consumed it.
+**Changelog entry:** 15.1.6 fixed native plan-mode `resolve` loops when `extra.title` is not a usable string.
 
-**Verdict.** No code, prompt, skill, rule, or doc in supipowers references the old scheme. The upstream rename is transparent.
+**Impact.** Supipowers intentionally does not use native OMP plan approval in `/supi:plan`; it relies on a file watcher, schema validation, reliability logging, and a custom approval UI. The planning prompt says this is not native OMP plan mode, forbids `resolve` with `extra.title`, forbids `local://PLAN.md`, and requires saving to the supipowers plan directory (`src/planning/system-prompt.ts:233-240`). The approval hook then detects new plan files by filename delta (`src/planning/approval-flow.ts:314-331`) and presents `ctx.ui.select` approval options (`src/planning/approval-flow.ts:431-443`).
 
-**Recommendation.** None.
+The runtime guard currently blocks only `resolve` calls whose input has `action: "apply"` and a string `extra.title` (`src/planning/planning-ask-tool.ts:130-138`, `src/planning/planning-ask-tool.ts:152-160`). Because OMP now accepts/falls back for malformed `extra.title`, a mistaken `resolve({ action: "apply", extra: { title: {} } })` would not match our guard. Normal `/supi:plan` should not have an OMP-native pending action to resolve, so this is not a confirmed break.
 
----
+**Recommendation.** Harden the guard in a follow-up: while `isPlanningActive()` is true, block every `resolve` call with `action: "apply"`, not just the string-title native shape. Add a narrow unit test for string, object, missing, and non-apply inputs.
 
-### 1.2 [LOW – DOCS DRIFT] `StringEnum` removed from `@oh-my-pi/pi-coding-agent`
+### Compatibility risk: ACP command/custom-tool argument preservation is now fixed upstream, but these paths depend on preserved raw inputs
 
-**Changelog source (15.1.3):**
-> Removed the `StringEnum` re-export from `@oh-my-pi/pi-coding-agent`. Custom tools and extensions should use `z.enum([...])` directly via the injected `pi.zod`.
+**Changelog entry:** 15.1.4 fixed ACP command and custom tool-call notifications to carry original tool arguments in replayed/final updates.
 
-**Runtime impact: none.** `grep -rln 'StringEnum' src tests` → **0 matches**. Supipowers never imported `StringEnum`; all our tool `parameters:` are plain JSON-Schema literals (`src/harness/tools.ts`, `src/ultraplan/authoring/authoring-tools.ts`, `src/ultraplan/authoring-tool.ts`, `src/context-mode/tools.ts`, `src/planning/planning-ask-tool.ts`, `src/mempalace/tool.ts`), as documented in the previous audit's §3.
+**Impact.** Supipowers command semantics use direct handler args, not ACP notification payloads. The OMP adapter forwards command registration unchanged (`src/platform/omp.ts:51-52`), `/supi:plan --quick` parses the raw command string (`src/commands/plan.ts:79-83`), and `planning_ask` receives structured params directly in its execute callback (`src/planning/planning-ask-tool.ts:56-63`). Custom tools are forwarded without local wrapping (`src/platform/omp.ts:135-135`). This is an upstream reliability fix, not a local API break.
 
-**Docs-side impact: shipped extension-dev reference is stale.** Two files under the tracked `.omp/skills/omp-extension-dev/references/` skill (loaded by `.omp/rules/omp-extension-dev.md:2`) instruct readers to import the now-removed symbol:
+**Recommendation.** Keep on the smoke-test list after OMP upgrades: `/supi:plan --quick <text with spaces/punctuation>` and one `planning_ask` call. No source change required.
 
-| File | Line | Snippet |
-| --- | --- | --- |
-| `.omp/skills/omp-extension-dev/references/custom_tools.md` | 122–124 | ``// Enums (use StringEnum from pi.pi for string enums)\nconst { StringEnum } = pi.pi;\nStringEnum(["staging", "production"], { description: "Target environment" })`` |
-| `.omp/skills/omp-extension-dev/references/api_reference.md` | 270 | ``\| `pi.pi` \| Module \| `@oh-my-pi/pi-coding-agent` exports — `StringEnum`, `logger`, etc. \|`` |
+### No-impact entries
 
-`.omp/skills/omp-extension-dev/SKILL.md:28,31` links both files from its "Custom Tools" and "Configuration" rows, and the rule is the `omp-extension-dev` rulebook entry the model loads whenever a user works on an OMP extension. The model will happily reproduce the dead pattern.
+| Changelog entry | Impact | Evidence | Recommendation |
+| --- | --- | --- | --- |
+| 15.1.7 `debug` launch/attach fixes and directory-valued launch rejection | Supipowers does not invoke OMP DAP launch/attach. Its own debug logger is `SUPI_DEBUG` file logging, not OMP `debug`. | `src/debug/logger.ts:19-24`, `src/debug/logger.ts:36-52`, `src/platform/types.ts:45-55`, `src/platform/omp.ts:101-133` | No migration. |
+| 15.1.7 hashline edit payload warning | Runtime code does not construct hashline `edit` payloads. Planning mode forbids implementation edits and only tells agents to save plan markdown. | `src/planning/system-prompt.ts:193-218`, `src/planning/system-prompt.ts:223-241`, `skills/context-mode/SKILL.md:64-68` | No runtime migration. Optional docs polish in Opportunities. |
+| 15.1.7 ACP bash permission metadata | `Platform.exec` exposes only `cwd`, `timeout`, and `env`; non-env calls delegate to `api.exec`, env calls use `Bun.spawn`. Supipowers does not read ACP permission metadata. | `src/platform/types.ts:66-77`, `src/platform/omp.ts:11-17`, `src/platform/omp.ts:56-57` | No migration. |
+| 15.1.7 fast-mode status-line predicate | Supipowers writes its own `supi-model` footer status for model overrides but does not inspect OMP fast-mode state. | `src/config/model-resolver.ts:172-185`, `src/config/model-resolver.ts:187-195`, `src/platform/types.ts:103-115` | Smoke-test status-line coexistence; no code migration unless `ctx.ui.setStatus` semantics change. |
+| 15.1.7 scoped `serviceTier` values | Supipowers model config has only `model` and `thinkingLevel`; no service-tier field is persisted or passed. | `src/types.ts:613-626`, `src/config/model-resolver.ts:10-55`, `src/config/model-resolver.ts:130-170`, `src/platform/omp.ts:67-78` | No migration. |
+| 15.1.7 provider-agnostic `/fast` | `/fast` remains OMP-owned. Supipowers controls only `setModel` and `setThinkingLevel`. | `src/platform/omp.ts:67-78`, `src/platform/types.ts:146-153`, `src/config/model-resolver.ts:142-170` | No migration. |
+| 15.1.5 `ast_grep` / `ast_edit` `parseErrors` cap | Supipowers does not post-process native AST tool `parseErrors`. Its `parseErrors` usage in doctor is config parsing, not OMP AST results. | `src/platform/omp.ts:135-148`, `src/context-mode/tools.ts:453-458`, `src/commands/doctor.ts:76-80` | No migration. |
+| 15.1.4 `normalizePlanTitle` spaces/punctuation | Supipowers does not use native plan title normalization. It watches `.md` files under its own plans directory. | `src/planning/system-prompt.ts:233-238`, `src/storage/plans.ts:12-24`, `src/planning/approval-flow.ts:318-331` | No migration. |
+| 15.1.4 built-in `ask` prompt example | `/supi:plan` uses a custom `planning_ask` schema and blocks generic `ask` while planning/UI-design is active. | `src/planning/planning-ask-tool.ts:24-55`, `src/planning/planning-ask-tool.ts:108-149`, `src/planning/system-prompt.ts:230-232` | No migration. |
+| 15.1.4 ACP async-job owner scoping/status/deferred turns | Supipowers does not implement ACP job draining. Its agent workers consume OMP agent sessions and dispose them normally. | `src/ultraplan/execution/session-runner.ts:157-167`, `src/ultraplan/batch/worker.ts:17-25`, `src/commands/ultraplan.ts:936-980` | No migration. See Opportunities for future status integration. |
+| 15.1.4 edit/write/ast_edit permission behavior | Supipowers prompts may instruct agents to write plan artifacts, but production code does not invoke OMP edit/write/ast_edit APIs directly. | `src/planning/system-prompt.ts:193-218`, `src/planning/system-prompt.ts:233-241`, `src/planning/approval-flow.ts:431-463` | No migration. |
+| 15.1.4 session tree selector readability | Supipowers only calls `ctx.newSession()` for plan execution handoff and checks cancellation. It does not use the session tree selector API. | `src/planning/approval-flow.ts:238-247`, `src/planning/approval-flow.ts:253-284`, `src/commands/plan.ts:183-192` | No migration. |
 
-`git ls-files .omp/skills/omp-extension-dev` confirms supipowers ships these files; they are not user-side state. The repo owns this drift.
+## Opportunities
 
-**Recommendation.** Replace both snippets with the `pi.zod` form named in the changelog:
+### P1 — Keep command-driven hidden steer handoffs on `sendMessage(..., triggerTurn: true)`
 
-```ts
-// .omp/skills/omp-extension-dev/references/custom_tools.md:122-124
-// Enums (use pi.zod for string enums)
-const { z } = pi.zod;
-z.enum(["staging", "production"]).describe("Target environment")
-```
+**Why.** OMP 15.1.4 fixed deferred agent-initiated turns during ACP async-job draining. Supipowers already uses the safer pattern for orchestration commands that persist state before dispatching an agent turn.
 
-```md
-<!-- .omp/skills/omp-extension-dev/references/api_reference.md:270 -->
-| `pi.zod` | Module | `zod/v4` — preferred schema builder (`z.object()`, `z.enum()`, …) |
-| `pi.pi`  | Module | `@oh-my-pi/pi-coding-agent` exports — `logger`, lifecycle types, etc. |
-```
+**Evidence.** `/supi:qa` creates the E2E ledger before dispatch (`src/commands/qa.ts:371-372`) and sends hidden `supi-qa` content with `{ deliverAs: "steer", triggerTurn: true }` (`src/commands/qa.ts:398-405`). `/supi:fix-pr` persists or initializes a running ledger (`src/commands/fix-pr.ts:413-427`) and sends `supi-fix-pr` the same way (`src/commands/fix-pr.ts:521-528`). UltraPlan authoring uses the same handoff (`src/commands/ultraplan.ts:1456-1465`). The adapter currently forwards `deliverAs` and `triggerTurn` (`src/platform/omp.ts:58-63`).
 
-No tests gate these docs; the change is content-only.
+**Effort.** S.
 
----
+**Guidance.** Do not replace these flows with `sendUserMessage`. Add a release/support note that OMP ≥15.1.4 is recommended for command-driven agent handoffs.
 
-### 1.3 [NONE] `eval` tool replaced LARK input with structured `cells` array — no in-repo callers
+### P2 — Harden the `/supi:plan` native `resolve` block
 
-**Changelog source (15.1.3):**
-> Replaced the `eval` tool's LARK-grammar `input` string with a structured `cells` array. Each cell is `{ language: "py" | "js", code, title?, timeout?, reset? }`. Removed the implicit/sniffed language path, the `*** Cell` / `*** End` / `*** Abort` markers, and the per-cell `t:<duration>` unit suffixes — `timeout` is now seconds (1-600).
+**Why.** The 15.1.6 fallback made malformed native plan-title resolution non-looping upstream. That is good for OMP, but our plan-mode policy is broader: native `resolve` should not be used at all during `/supi:plan`.
 
-**Evidence collected.**
+**Evidence.** Policy: `src/planning/system-prompt.ts:233-240`. Guard: `src/planning/planning-ask-tool.ts:130-138`. Narrow matcher: `src/planning/planning-ask-tool.ts:152-160`. Supipowers approval path: `src/planning/approval-flow.ts:314-331`, `src/planning/approval-flow.ts:431-463`.
 
-- `grep -rln -E '\*\*\* (Cell|End|Abort)' src tests skills global-config` → **0 matches**.
-- `grep -rn -E 'name:\s*"eval"|registerTool.*name.*eval' src tests` → **0 matches**.
-- `grep` for `"eval"` / `` `eval` `` / `eval tool` / `/eval` inside `src`, `tests`, `skills/`, `global-config/`, and `docs/`, after stripping the `evalua…` substring noise, returns only references to:
-  - our own behavior-test framework `tests/evals/` (`tests/evals/{harness,fixtures}.ts`, `tests/evals/README.md`, `docs/supipowers/failure-mining.md:35-90`) — unrelated to OMP's `eval` tool;
-  - `src/harness/artifacts/lint-configs.ts:2,125,130` and `src/harness/stages/implement-apply.ts:312,317,324,431,424` — these emit/consume `tooling.eval` config keys describing a *user project's* mutation-test/eval framework, not the OMP runtime `eval` tool.
-- No subagent prompt instructs the model to call the OMP `eval` tool (`grep "use the eval"|"run eval"|"\beval("` → 0 hits outside the items above).
+**Effort.** S.
 
-**Verdict.** Supipowers neither calls the OMP `eval` tool from TS code nor instructs subagents to use it through prompt text. The LARK→`cells` migration and the dropped `*** Cell` markers are transparent to us.
+**Guidance.** Replace `isPlanApprovalResolveInput` with an apply-shape check that returns true for any object where `action === "apply"`. Keep non-apply `resolve` inputs unaffected. Add unit tests for string title, object title, missing `extra`, missing `title`, and non-apply actions.
 
-**Recommendation.** None.
+### P2 — Future-proof `Platform.sendMessage` for new ACP messaging metadata
 
----
+**Why.** The 15.1.4 ACP fixes are about preserving command/custom-tool arguments in updates. Supipowers currently reconstructs the send options object and would silently drop future send metadata unless the type and adapter are updated together.
 
-### 1.4 [NONE] 15.1.2 fixed items audit
+**Evidence.** `SendMessageOptions` has only `deliverAs` and `triggerTurn` (`src/platform/types.ts:81-84`). `createOmpAdapter.sendMessage` forwards only those two keys (`src/platform/omp.ts:58-63`). High-value callers are QA, Fix-PR, and UltraPlan (`src/commands/qa.ts:398-405`, `src/commands/fix-pr.ts:521-528`, `src/commands/ultraplan.ts:1458-1465`).
 
-| Changelog item | Probe | Result |
-| --- | --- | --- |
-| SSH host add/remove not refreshing live `ssh` tool | `grep -rn -E '"ssh"\|/ssh add\|/ssh remove\|sshHost' src tests` → **0 matches** | We never drive the `ssh` tool. |
-| `YieldTool` loose object output schemas | `grep -rn -E 'YieldTool\|outputSchema:\s*(true\|\{\|\?)' src` → 0 *runtime* hits; existing `outputSchema:` matches in `src/review/{multi-agent-runner,runner,validator}.ts` and `src/quality/gates/ai-review.ts` are **prompt-template strings** (e.g. `REVIEW_OUTPUT_SCHEMA_TEXT` from `src/ai/schema-text.ts`), not OMP `YieldTool` config | We never configure OMP's `YieldTool`; schema validation flows through our own `parseStructuredOutput` (`src/ai/structured-output.ts:124`). |
-| Unconstrained `outputSchema` modes (`outputSchema: true` / absent) | Same as above | We never pass `outputSchema` to `createAgentSession`; `src/ai/final-message.ts:89-99` shows the full options surface we use. |
-| Bash `pty: true` hang on Windows | `grep -rn -E 'pty:\s*true' src tests` → **0 matches** | We never request a PTY. |
-| MCP/theme schema metadata → JSON Schema draft-2020-12 | `.omp/mcp.json:2` references the live raw GitHub URL; we already render our own schemas at draft-2020-12 (`src/ai/schema-text.ts:146`) | Nothing to migrate. |
+**Effort.** S when OMP exposes a stable field.
 
----
+**Guidance.** Do not blindly spread unknown options. When OMP exposes a typed metadata field, add it to `SendMessageOptions`, forward it explicitly, and add adapter unit coverage that the field reaches `api.sendMessage`.
 
-### 1.5 [NONE] Other 15.1.3 fixed items
+### P2 — Use future in-flight completion status to close orchestration ledgers on dispatch failure
 
-All other "Fixed" entries are scoped to subsystems we don't touch:
+**Why.** OMP 15.1.4 now reports in-flight completions more accurately internally. Supipowers ledgers currently mark work as started/running after dispatch but do not observe whether the triggered turn actually starts and completes.
 
-| Changelog item | Why we are unaffected |
-| --- | --- |
-| Streaming auth recovery (invalidate + retry on stale credentials) | We have no retry-on-401 wrapper around `createAgentSession`. The upstream improvement is a transparent reliability win for every supipowers AI gate; no code change required. |
-| `auth-broker migrate` / `auth-gateway` startup / `discoverAuthStorage` fast-fail on non-200 snapshot | `grep -rn -E 'auth-broker\|auth-gateway' src tests` → 0 matches; we don't operate the broker/gateway ourselves. |
-| `auth-broker migrate` skip `<authenticated>` placeholder credentials | Same — we never invoke broker migration. |
-| `auth-gateway` token-init race fix, protocol-control rejection, usage cached-token fix, abort-before-dispatch fix | Same. |
-| `/login` and `/logout` selector overflow → 10-item scroll window + PageUp/PageDown | OMP-internal TUI; our `src/commands/model-picker.ts:286-316` already implements its own scrolling window for the supipowers model picker. |
-
-### 1.6 [NONE] TTSR `interruptMode` semantics change
-
-**Changelog source (15.1.3):**
-> Changed TTSR `interruptMode` semantics so a non-interrupting decision on a tool-source match now folds the rule reminder into that specific tool's `toolResult` content instead of queuing a loop-wide deferred follow-up turn. Text/thinking matches keep the previous deferred-injection behavior.
-
-**Probe.** `src/commands/runbook.ts:30,145,345` and `src/context/runbook-extension-template.ts:7-179` reference TTSR rules by reading the `interruptMode` frontmatter field for display only — they do not implement or modify firing semantics. Our shipped rules at `.omp/rules/*.md` are 22 files (`acp.md`, `code-review.md`, …, `writing-skills.md`); a `grep -l 'interruptMode' .omp/rules/*.md` returns **0 matches**. None of them set `interruptMode`, so all of them inherit OMP's default. The behavior change only affects rules that combine `interruptMode: prose-only` (or similar non-interrupting modes) with tool-source matchers, neither of which any supipowers rule does.
-
-**Verdict.** No behavioral drift in our shipped rules. The `runbook` command continues to print `Interrupt: ${rule.interruptMode ?? "default"}` unchanged.
+**Evidence.** QA creates a session and immediately notifies started after `sendMessage` (`src/commands/qa.ts:371-372`, `src/commands/qa.ts:398-411`). Fix-PR initializes `status: "running"` before dispatch (`src/commands/fix-pr.ts:413-427`) and then notifies started (`src/commands/fix-pr.ts:521-535`). UltraPlan batch sets state to `running`, runs supervisor passes, and persists transitions (`src/commands/ultraplan.ts:979-993`). `Platform.sendMessage` has no completion/status hook today (`src/platform/types.ts:81-84`).
 
-**Recommendation.** None. If we ever author a TTSR rule with `interruptMode: prose-only` *and* a `condition` that matches against tool output, the reminder will now appear inside that tool's `toolResult` instead of as a separate next-turn nudge — keep that in mind during future authoring.
+**Effort.** M after public API support exists.
 
----
+**Guidance.** Wait for a public OMP extension API; do not synthesize completion with timers. Once available, extend the platform abstraction and update QA/Fix-PR/UltraPlan ledgers to a blocked/error state if dispatch fails before the orchestrator starts.
 
-## 2. Opportunities
+### P3 — Add optional exec-operation reason metadata only if OMP exposes it
 
-### 2.1 [P3, Trivial] Streaming-auth retry is a free reliability win
+**Why.** OMP 15.1.7 improves ACP bash permission prompts. Supipowers has several high-impact command paths where a reason string would help if OMP exposes a command metadata field.
 
-**Changelog (15.1.3 Fixed):**
-> Fixed streaming API requests to recover from provider auth errors by invalidating stale credentials and retrying with a fresh key.
+**Evidence.** `ExecOptions` currently supports only `cwd`, `timeout`, and `env` (`src/platform/types.ts:66-70`). Non-env exec delegates directly to OMP (`src/platform/omp.ts:56-57`). High-impact calls include `git worktree add` (`src/commands/ultraplan.ts:710-729`) and `gh api` review fetching (`src/fix-pr/fetch-comments.ts:111-129`, `src/fix-pr/fetch-comments.ts:220-246`).
 
-Our AI gates (`src/quality/gates/ai-review.ts`, `src/lsp/{capabilities,bridge}.ts`, `src/fix-pr/assessment.ts`, `src/docs/drift.ts`, `src/review/*`, `src/commands/release.ts`, `src/git/commit.ts`, `src/ultraplan/authoring/stages/*`) all route through `runWithOutputValidation` (`src/ai/structured-output.ts:160-225`) which retries only on *parse / schema* failures, not on transport errors. Transient 401s from rotated OAuth tokens previously surfaced as `agent-error` outcomes in `src/storage/reliability-metrics.ts` (`"retry-exhausted"` / `"agent-error"`); after 15.1.3 they should self-heal inside OMP and the rate of `agent-error` records in the reliability log should drop. Worth noting in release-notes; no code change.
+**Effort.** S if OMP exposes the field; none otherwise.
 
-### 2.2 [P3, Trivial] Provider auth-broker / auth-gateway is a deployment story for container users
+**Guidance.** If OMP adds a stable `description`/`reason`/metadata option for `api.exec`, thread it through `ExecOptions` and annotate destructive or networked operations. Do not add a parallel supipowers permission prompt.
 
-**Changelog (15.1.3 Added):**
-> Added `omp auth-broker` subcommand … `omp auth-gateway` subcommand …  
-> Added `providers.<name>.transport: "pi-native"` to `models.yml`. … Use case: containerized omp installs (robomp slots, swarm extension) where the slot must stay credential-free and a sidecar gateway holds the real provider tokens.
+### P3 — Decide deliberately before exposing provider-scoped `serviceTier` in supipowers model config
 
-Supipowers needs no code change — every credential we use is owned by OMP. The opportunity is **documentation-only**: users who run supipowers inside containers (robomp slot, swarm extension, CI runner) can now hide provider tokens behind a sidecar `omp auth-gateway` and pin `providers.<name>.transport: "pi-native"` plus `providers.<name>.apiKey: <gateway-bearer>` in `~/.omp/agent/models.yml`. Worth a paragraph in `README.md` under deployment guidance once we publish against an OMP version that includes 15.1.3.
+**Why.** OMP 15.1.7 adds provider-scoped priority tiers and makes `/fast` provider-agnostic. Supipowers per-action model config cannot request those tiers today.
 
-### 2.3 [P3, Trivial] `auth.broker.url` / `auth.broker.token` config tier closes a model-resolution edge case for us
+**Evidence.** `ModelAssignment` has only `model` and `thinkingLevel` (`src/types.ts:613-619`), and `ModelConfig` has only `version`, `default`, and `actions` (`src/types.ts:621-626`). Resolution and application return/pass only `model`, `thinkingLevel`, and `source` (`src/config/model-resolver.ts:16-24`, `src/config/model-resolver.ts:142-185`). The OMP adapter exposes `setModel`, `setThinkingLevel`, `getCurrentModel`, and `getModelForRole`, not service-tier controls (`src/platform/omp.ts:67-78`).
 
-**Changelog (15.1.3 Added):**
-> `ModelRegistry` now promotes `models.yml` `providers.<name>.apiKey` entries to `AuthStorage`'s new config-override tier (above OAuth, below `--api-key`). Pinning a bearer in `models.yml` was previously a no-op when the broker had an OAuth credential for the same provider …
+**Effort.** M.
 
-`src/commands/model.ts:184-218` (`selectModelFromList`) calls `ctx.modelRegistry?.getAvailable?.()`, then renders a "no key" warning in `src/commands/model-picker.ts:303,306,323`. Pre-15.1.3, a user with OAuth credentials *and* a `models.yml`-pinned bearer for the same provider would see the OAuth resolution win silently and (post-redirect to a gateway) hit 401 — the picker would still show the model as "configured" even though dispatch was broken. 15.1.3 makes the `models.yml` bearer authoritative, so our picker's "configured" badge is now accurate end-to-end. No code change required.
+**Guidance.** Do nothing unless users ask for per-action fast/priority behavior. If they do, first confirm the public OMP API shape, then extend `ModelAssignment`, `resolveModelForAction`, `resolveAllCandidates`, and `applyModelOverride` together. Keep `/fast` itself documented as OMP-owned.
 
-### 2.4 [N/A] `dev.autoqaPush.*` and `getInstallId()` — not applicable
+### P3 — Refresh hashline/edit reference docs if they mention edit payloads
 
-- `dev.autoqaPush.enabled` / `endpoint` / `token` and the `report_tool_issue` push pipeline target OMP's tool-quality grievance database. `grep -rn 'report_tool_issue' src tests` → 0 hits. Supipowers does not emit grievances; nothing to wire up.
-- `getInstallId()` (new export from `@oh-my-pi/pi-utils`) provides a stable per-install UUID at `~/.omp/install-id`. Our project-scoped state already uses repo-identity slugs (`src/workspace/state-paths.ts:79-83`, `src/harness/project-paths.ts:81-82`, `src/ultraplan/project-paths.ts:73-81`) which deliberately key on repo, not host. Install-level identity is a different axis and we have no current consumer for it.
+**Why.** OMP 15.1.7 now warns on separator-padding-shaped hashline payload mistakes. Supipowers does not generate edit payloads, but its agent-facing reference docs mention the edit contract.
 
----
+**Evidence.** Context-mode docs mention OMP read anchors for the edit contract (`skills/context-mode/SKILL.md:64-68`). Runtime `ctx_open_cached` slices cached text and does not synthesize edit anchors (`src/context-mode/tools.ts:880-950`).
 
-## 3. Files not touched and why
+**Effort.** S.
 
-- `src/platform/omp.ts:101-133` — `createAgentSession` continues to extract `model` → `modelPattern`. No new option in 15.1.0–15.1.3 needs plumbing here.
-- `src/ai/structured-output.ts` — still the single canonical retry/parse path; no new schema-validation hook in 15.1.3.
-- `src/ai/final-message.ts:89-99` — verified that we do not pass an `outputSchema` to OMP's `createAgentSession`, so the 15.1.2 `YieldTool` loose/strict fix is irrelevant.
-- `src/commands/model-picker.ts:286-316` — our own scroll/window logic is independent of OMP's `/login` selector fix.
-- `src/commands/runbook.ts:25-345` — reads but does not enforce `interruptMode`; the 15.1.3 semantic change has nothing to update here.
-- `.omp/mcp.json:2` — `$schema` pulls the live OMP MCP schema URL; the draft-2020-12 metadata update is transparent.
+**Guidance.** If editing agent docs, add one sentence: do not include `|content` in anchors, do not fabricate anchors, and payload lines must start with `~` immediately followed by intended file content.
 
----
+### P3 — Update `/supi:doctor` recommendations after an OMP minimum-version bump
 
-## 4. Summary table
+**Why.** Doctor recommendations still mention OMP 14.7-era UX fixes only. If the project decides to require or recommend OMP ≥15.1.7, doctor can surface the high-value runtime fixes from this audit.
 
-| ID | Type | Severity / Priority | Effort | Surface |
-| --- | --- | --- | --- | --- |
-| 1.1 | Breaking | **None** — no in-repo references to `pi://` / `PiProtocolHandler` | — | — |
-| 1.2 | Breaking (docs only) | **Low** — model will keep reproducing dead `StringEnum` snippet | S (2 markdown edits) | `.omp/skills/omp-extension-dev/references/custom_tools.md:122-124`, `.omp/skills/omp-extension-dev/references/api_reference.md:270` |
-| 1.3 | Breaking | **None** — no caller invokes OMP `eval` tool | — | — |
-| 1.4 | Breaking (15.1.2 cluster) | **None** — every probed feature (SSH tool, YieldTool, PTY bash, MCP/theme schema) is unused by supipowers | — | — |
-| 1.5 | Breaking (15.1.3 fixed cluster) | **None** — all in `auth-broker` / `auth-gateway` / streaming retry / `/login` UI; out of our surface | — | — |
-| 1.6 | Breaking | **None** — no shipped rule sets `interruptMode` | — | — |
-| 2.1 | Opportunity (auto) | P3, Trivial | — | Reliability log; release-notes mention only |
-| 2.2 | Opportunity (deployment) | P3, Trivial | S (README paragraph) | Docs only |
-| 2.3 | Opportunity (auto) | P3, Trivial | — | `src/commands/model-picker.ts` "no key" badge accuracy |
-| 2.4 | N/A | — | — | `report_tool_issue` and install-id are not in use |
+**Evidence.** Current doctor recommendations are static strings for OMP ≥14.7.0 and ≥14.7.2 (`src/commands/doctor.ts:493-505`).
 
----
+**Effort.** S.
 
-## 5. Recommended landing order
+**Guidance.** After the minimum/recommended OMP version is bumped, add a non-failing tip for OMP ≥15.1.7 covering `/fast` status correctness and ACP command/permission prompt reliability. Do not make it a failing health check unless supipowers has a direct version detector.
 
-1. **§1.2 docs fix** — single-PR, 2 file edits in `.omp/skills/omp-extension-dev/references/`. Removes the only post-15.1.3 misdirection supipowers ships. No tests to update, no behaviour change.
-2. **§2.2 README paragraph** — fold into the next release that bumps the required OMP version past 15.1.3. Cite `transport: "pi-native"` + `omp auth-broker` + `omp auth-gateway` for container deployments.
+## Summary table
 
-Everything else is no-op for supipowers (15.1.0 → 15.1.3 is the cleanest delta we've audited so far).
+| Changelog entry | Breaking impact | Opportunity | Priority | Effort | Primary evidence |
+| --- | --- | --- | --- | --- | --- |
+| 15.1.7 debug launch/attach fixes | None | None | — | — | `src/platform/omp.ts:101-133`, `src/debug/logger.ts:19-24` |
+| 15.1.7 hashline edit payload warning | None | Optional doc refresh | P3 | S | `src/planning/system-prompt.ts:223-241`, `skills/context-mode/SKILL.md:64-68` |
+| 15.1.7 ACP bash permission metadata | None | Future exec reason metadata | P3 | S if API exists | `src/platform/types.ts:66-70`, `src/platform/omp.ts:56-57` |
+| 15.1.7 fast-mode status-line predicate | Low UI coexistence risk only | Doctor note after version bump | P3 | S | `src/config/model-resolver.ts:172-195`, `src/commands/doctor.ts:493-505` |
+| 15.1.7 scoped `serviceTier` values | None | Per-action service tier only if requested | P3 | M | `src/types.ts:613-626`, `src/config/model-resolver.ts:16-24` |
+| 15.1.7 provider-agnostic `/fast` | None | Keep `/fast` OMP-owned | P3 | — | `src/platform/omp.ts:67-78` |
+| 15.1.6 `resolve.extra.title` fallback | No hard break; guard gap | Broaden `/supi:plan` `resolve` guard | P2 | S | `src/planning/planning-ask-tool.ts:130-160` |
+| 15.1.5 AST parse error cap | None | None | — | — | `src/platform/omp.ts:135-148`, `src/commands/doctor.ts:76-80` |
+| 15.1.4 `normalizePlanTitle` | None | None | — | — | `src/storage/plans.ts:12-24`, `src/planning/approval-flow.ts:318-331` |
+| 15.1.4 built-in `ask` prompt example | None | None | — | — | `src/planning/planning-ask-tool.ts:24-55`, `src/planning/planning-ask-tool.ts:108-149` |
+| 15.1.4 ACP command/custom notification args | None | Smoke-test command args/tool params | P2 | S | `src/commands/plan.ts:79-83`, `src/planning/planning-ask-tool.ts:56-63` |
+| 15.1.4 async job scoping/status/defer | None | Future ledger status integration | P2 | M | `src/commands/qa.ts:398-411`, `src/commands/fix-pr.ts:521-535`, `src/commands/ultraplan.ts:979-993` |
+| 15.1.4 edit/write/ast_edit permission behavior | None | None | — | — | `src/planning/system-prompt.ts:193-218`, `src/planning/approval-flow.ts:431-463` |
+| 15.1.4 session tree selector | None | None | — | — | `src/planning/approval-flow.ts:253-284` |
diff --git a/README.md b/README.md
index 460b3b2f..5395c7d4 100644
--- a/README.md
+++ b/README.md
@@ -36,6 +36,9 @@ The installer detects Pi (`~/.pi`) and OMP (`~/.omp`) — when both are present
 | [Bun](https://bun.sh)                                 | Runtime — required for installation and the built-in SQLite FTS index |
 | [Git](https://git-scm.com)                            | Used by the installer and git-based workflows                         |
 
+> [!TIP]
+> OMP ≥15.1.7 is recommended for best reliability with supipowers command-driven agent handoffs and accurate provider-scoped `/fast` status indicators. Older compatible OMP versions can run supipowers but lack those runtime fixes.
+
 ### Optional dependencies
 
 The installer scans for these and offers to install missing tooling where it can. Everything works without them, but each one unlocks additional capabilities.
diff --git a/skills/context-mode/SKILL.md b/skills/context-mode/SKILL.md
index fada69a0..aca18a0d 100644
--- a/skills/context-mode/SKILL.md
+++ b/skills/context-mode/SKILL.md
@@ -63,7 +63,7 @@ When OMP's `shellMinimizer` is active, large bash output ends with a `[raw outpu
 
 ### Read
 
-Reads are never blocked — OMP's native open/read tool preserves hashline anchors (e.g., `120th|content` after 14.4.1) for the edit contract. Large reads (>110 lines) are auto-compressed to head (80) + tail (30) with a `sel` hint.
+Reads are never blocked — OMP's native open/read tool preserves hashline anchors (e.g., `120th|content` after 14.4.1) for the edit contract. Copy edit anchors exactly, without the `|content` body, and never fabricate anchors. Edit payload lines must start with `~` immediately followed by intended file content; avoid a readability space after `~` unless that space is intentional file content. Large reads (>110 lines) are auto-compressed to head (80) + tail (30) with a `sel` hint.
 
 For analysis-only reads where anchors are not needed, prefer `ctx_execute_file(path, language, code)` — only your printed summary enters context.
 
diff --git a/src/commands/doctor.ts b/src/commands/doctor.ts
index 2b5e1a4c..fd22e944 100644
--- a/src/commands/doctor.ts
+++ b/src/commands/doctor.ts
@@ -502,6 +502,7 @@ export function getDoctorRecommendations(): string[] {
   return [
     "Set `tools.elideFileMutationInputs: true` (OMP ≥14.7.0) — elides `write`/`edit`/`apply_patch` payloads from history after success. Saves significant context on long sessions like `/supi:ultraplan execute` and `/supi:harness implement`.",
     "Update to OMP ≥14.7.2 — fixes the `Working…` spinner staying active after read-only commands such as `/supi:status`, `/supi:doctor`, `/supi:context`, and `/supi:clear`. (oh-my-pi#927)",
+    "Use OMP ≥15.1.7 for best reliability — includes ACP fixes for command-driven agent handoffs and permission prompts, plus accurate provider-scoped `/fast` status-line indicators.",
   ];
 }
 
diff --git a/src/mempalace/git-hook.ts b/src/mempalace/git-hook.ts
index e0be0bb6..3e6a9d7e 100644
--- a/src/mempalace/git-hook.ts
+++ b/src/mempalace/git-hook.ts
@@ -94,9 +94,15 @@ async function resolveHookContext(exec: ExecFn, cwd: string): Promise<MempalaceG
   }
 
   const coreHooksPathResult = await gitValue(exec, repoRoot, ["config", "--get", "core.hooksPath"]);
-  const coreHooksPath = coreHooksPathResult.ok && coreHooksPathResult.value.length > 0
-    ? coreHooksPathResult.value
-    : null;
+  let coreHooksPath: string | null = null;
+  if (coreHooksPathResult.ok) {
+    coreHooksPath = coreHooksPathResult.value.length > 0 ? coreHooksPathResult.value : null;
+  } else if (coreHooksPathResult.code !== 1) {
+    // git exits 1 when the config key is unset (expected). Any other code is
+    // a real failure (corrupt config, permission error) and must not silently
+    // fall back to the default hooks dir.
+    return { ok: false, code: "git_failed", message: coreHooksPathResult.message };
+  }
   const hooksDir = coreHooksPath !== null
     ? resolveMaybeRelative(repoRoot, coreHooksPath)
     : path.join(resolveMaybeRelative(repoRoot, commonDirResult.value), "hooks");
@@ -378,25 +384,30 @@ export async function installMempalacePostCommitHook(options: InstallMempalacePo
 
   const resolved = resolveMempalaceConfig(options.config, context.repoRoot, options.paths);
   const runnerPath = options.paths.global("bin", REINDEX_RUNNER_NAME);
-  writeExecutableFile(runnerPath, buildReindexRunnerScript());
 
   const desiredHook = buildPostCommitHookScript(snapshot, runnerPath, resolved.palacePath, resolved.defaultAgentName);
   const existingHook = readTextIfPresent(context.hookPath);
-  let action: MempalacePostCommitHookInstallAction = "installed";
+
+  // Refuse the install before writing any artifacts: when there is a
+  // non-managed active hook AND the chained user slot is already taken,
+  // we have nowhere to move the existing hook to.
+  if (existingHook !== null && !isManagedHook(existingHook) && fs.existsSync(context.userHookPath)) {
+    return {
+      ok: false,
+      code: "user_hook_conflict",
+      message: `Cannot install MemPalace post-commit hook because both ${context.hookPath} and ${context.userHookPath} already exist and the active hook is not managed by supipowers.`,
+    };
+  }
+
+  writeExecutableFile(runnerPath, buildReindexRunnerScript());
 
   if (existingHook === desiredHook) {
     return { ...buildStatus(context, runnerPath), action: "already-installed" };
   }
 
   fs.mkdirSync(context.hooksDir, { recursive: true });
+  let action: MempalacePostCommitHookInstallAction = "installed";
   if (existingHook !== null && !isManagedHook(existingHook)) {
-    if (fs.existsSync(context.userHookPath)) {
-      return {
-        ok: false,
-        code: "user_hook_conflict",
-        message: `Cannot install MemPalace post-commit hook because both ${context.hookPath} and ${context.userHookPath} already exist and the active hook is not managed by supipowers.`,
-      };
-    }
     fs.renameSync(context.hookPath, context.userHookPath);
     action = "chained-user-hook";
   } else if (existingHook !== null) {
diff --git a/src/planning/planning-ask-tool.ts b/src/planning/planning-ask-tool.ts
index 900cd938..cf0fdca3 100644
--- a/src/planning/planning-ask-tool.ts
+++ b/src/planning/planning-ask-tool.ts
@@ -129,11 +129,11 @@ function getAskRedirectReason(): string | null {
  */
 export function registerPlanningAskToolGuard(platform: Platform): void {
   platform.on("tool_call", (event) => {
-    if (event.toolName === "resolve" && isPlanApprovalResolveInput(event.input) && isPlanningActive()) {
+    if (event.toolName === "resolve" && isResolveApplyInput(event.input) && isPlanningActive()) {
       return {
         block: true,
         reason:
-          "Planning mode: /supi:plan uses a file-based approval hook. Do not call `resolve` with `extra.title` because it is OMP's native approval path and bypasses supipowers plan tracking.",
+          "Planning mode: /supi:plan uses a file-based approval hook. Native OMP plan approval is blocked because it bypasses supipowers plan tracking.",
       };
     }
 
@@ -149,13 +149,7 @@ export function registerPlanningAskToolGuard(platform: Platform): void {
   });
 }
 
-function isPlanApprovalResolveInput(input: unknown): boolean {
+function isResolveApplyInput(input: unknown): boolean {
   if (input === null || typeof input !== "object" || Array.isArray(input)) return false;
-  const candidate = input as { action?: unknown; extra?: unknown };
-  if (candidate.action !== "apply") return false;
-  const extra = candidate.extra;
-  return extra !== null
-    && typeof extra === "object"
-    && !Array.isArray(extra)
-    && typeof (extra as { title?: unknown }).title === "string";
+  return (input as { action?: unknown }).action === "apply";
 }
diff --git a/tests/commands/doctor.test.ts b/tests/commands/doctor.test.ts
index 03f1131d..73d4af6a 100644
--- a/tests/commands/doctor.test.ts
+++ b/tests/commands/doctor.test.ts
@@ -352,4 +352,14 @@ describe("getDoctorRecommendations", () => {
     expect(spinner!).toContain("14.7.2");
     expect(spinner!).toContain("927");
   });
+
+  test("recommends OMP ≥15.1.7 for ACP handoffs and scoped fast-mode status", () => {
+    const tips = getDoctorRecommendations();
+    const omp1517 = tips.find((t) => t.includes("15.1.7"));
+    expect(omp1517).toBeDefined();
+    expect(omp1517!).toContain("ACP");
+    expect(omp1517!).toContain("handoffs");
+    expect(omp1517!).toContain("permission prompts");
+    expect(omp1517!).toContain("/fast");
+  });
 });
\ No newline at end of file
diff --git a/tests/mempalace/git-hook.test.ts b/tests/mempalace/git-hook.test.ts
index 99a3f865..a9075fe1 100644
--- a/tests/mempalace/git-hook.test.ts
+++ b/tests/mempalace/git-hook.test.ts
@@ -163,6 +163,7 @@ describe("MemPalace post-commit git hook", () => {
     const hooksDir = path.join(repoDir, ".git", "hooks");
     fs.writeFileSync(path.join(hooksDir, "post-commit"), "#!/bin/sh\necho active\n");
     fs.writeFileSync(path.join(hooksDir, "post-commit.user"), "#!/bin/sh\necho backup\n");
+    const runnerPath = paths.global("bin", "supi-mempalace-reindex.py");
 
     const result = await installMempalacePostCommitHook({
       paths,
@@ -176,6 +177,35 @@ describe("MemPalace post-commit git hook", () => {
     if (result.ok) throw new Error("expected conflict");
     expect(result.code).toBe("user_hook_conflict");
     expect(fs.readFileSync(path.join(hooksDir, "post-commit"), "utf-8")).toContain("active");
+    // Reject the install before writing any artifacts — the runner must not
+    // exist after a failed install.
+    expect(fs.existsSync(runnerPath)).toBe(false);
+  });
+
+  test("surfaces git_failed when core.hooksPath lookup errors out (non-unset exit code)", async () => {
+    const exec = mock(async (_cmd: string, args: string[]) => {
+      const key = args.join(" ");
+      if (key === "rev-parse --show-toplevel") return { code: 0, stdout: `${repoDir}\n`, stderr: "" };
+      if (key === "rev-parse --git-common-dir") return { code: 0, stdout: ".git\n", stderr: "" };
+      if (key === "config --get core.hooksPath") {
+        // Exit code 128 is what git emits for a corrupt config or other real
+        // failures, distinct from the code 1 it returns for "key unset".
+        return { code: 128, stdout: "", stderr: "fatal: bad config line 1 in file .git/config\n" };
+      }
+      return { code: 1, stdout: "", stderr: `unexpected git args: ${key}` };
+    }) as Platform["exec"];
+
+    const status = await getMempalacePostCommitHookStatus({
+      paths,
+      cwd: repoDir,
+      config: DEFAULT_CONFIG,
+      exec,
+    });
+
+    expect(status.ok).toBe(false);
+    if (status.ok) throw new Error("expected git_failed");
+    expect(status.code).toBe("git_failed");
+    expect(status.message).toContain("bad config");
   });
 
   test("respects a configured core.hooksPath", async () => {
diff --git a/tests/planning/planning-ask-tool.test.ts b/tests/planning/planning-ask-tool.test.ts
index ad2edc27..fb270c3a 100644
--- a/tests/planning/planning-ask-tool.test.ts
+++ b/tests/planning/planning-ask-tool.test.ts
@@ -33,7 +33,7 @@ function makePlatform() {
     registerTool: mock((def: any) => {
       toolDef = def;
     }),
-    fireToolCall: async (toolName: string, input: Record<string, unknown> = {}) => {
+    fireToolCall: async (toolName: string, input: unknown = {}) => {
       if (!toolCallHandler) return undefined;
       return await toolCallHandler({ toolName, input });
     },
@@ -177,11 +177,11 @@ describe("registerPlanningAskToolGuard — runtime ask redirect", () => {
     expect(result).toBeDefined();
     expect(result?.block).toBe(true);
     expect(result?.reason).toContain("file-based approval hook");
-    expect(result?.reason).toContain("extra.title");
+    expect(result?.reason).toContain("Native OMP plan approval");
     expect(result?.reason).toContain("bypasses supipowers");
   });
 
-  test("leaves ordinary resolve calls alone while planning is active", async () => {
+  test("blocks native resolve apply inputs with missing or malformed titles while planning is active", async () => {
     const platform = makePlatform();
     registerPlanningAskToolGuard(platform as any);
 
@@ -191,7 +191,33 @@ describe("registerPlanningAskToolGuard — runtime ask redirect", () => {
       global: (..._parts: string[]) => "/tmp/does-not-exist-global",
     } as any);
 
-    expect(await platform.fireToolCall("resolve", { action: "apply" })).toBeUndefined();
+    for (const input of [
+      { action: "apply" },
+      { action: "apply", extra: {} },
+      { action: "apply", extra: { title: {} } },
+      { action: "apply", extra: { title: null } },
+    ]) {
+      const result = (await platform.fireToolCall("resolve", input)) as
+        | { block: true; reason: string }
+        | undefined;
+
+      expect(result?.block).toBe(true);
+      expect(result?.reason).toContain("Native OMP plan approval");
+    }
+  });
+
+  test("leaves non-apply resolve inputs alone while planning is active", async () => {
+    const platform = makePlatform();
+    registerPlanningAskToolGuard(platform as any);
+
+    startPlanTracking("/cwd", {
+      dotDirDisplay: ".omp",
+      project: (_cwd: string, ..._parts: string[]) => "/tmp/does-not-exist",
+      global: (..._parts: string[]) => "/tmp/does-not-exist-global",
+    } as any);
+
+    expect(await platform.fireToolCall("resolve", { action: "discard" })).toBeUndefined();
+    expect(await platform.fireToolCall("resolve", {})).toBeUndefined();
   });
 
   test("blocks the `ask` tool when ui-design is active", async () => {
diff --git a/tests/platform/static-hazards.test.ts b/tests/platform/static-hazards.test.ts
index fbcc8e97..a4dd24ba 100644
--- a/tests/platform/static-hazards.test.ts
+++ b/tests/platform/static-hazards.test.ts
@@ -4,6 +4,13 @@ import { describe, expect, test } from "bun:test";
 
 const REPO_ROOT = path.resolve(import.meta.dir, "../..");
 const SRC_ROOT = path.join(REPO_ROOT, "src");
+const CROSS_PLATFORM_SKILL_REFERENCES_ROOT = path.join(
+  REPO_ROOT,
+  ".agents",
+  "skills",
+  "cross-platform-compatibility",
+  "references",
+);
 const SKIPPED_DIRS = new Set(["node_modules"]);
 const SOURCE_EXTENSIONS = new Set([".ts", ".tsx", ".js", ".jsx"]);
 const CHILD_PROCESS_IMPORT_ALLOWLIST = new Set([
@@ -94,4 +101,30 @@ describe("cross-platform static hazards", () => {
 
     expect(violations).toEqual([]);
   });
+
+  test("cross-platform skill process docs expose taskkill failures to callers", () => {
+    const processManagement = fs.readFileSync(
+      path.join(CROSS_PLATFORM_SKILL_REFERENCES_ROOT, "process-management.md"),
+      "utf8",
+    );
+
+    expect(processManagement).toContain(
+      "static async kill(pid: number, signal?: string): Promise<void>",
+    );
+    expect(processManagement).toContain('await execFileAsync("taskkill"');
+    expect(processManagement).not.toContain('proc.on("error"');
+    expect(processManagement).not.toContain('proc.on("exit"');
+  });
+
+  test("cross-platform skill shell docs do not pass caller paths through cmd builtins", () => {
+    const shellCommands = fs.readFileSync(
+      path.join(CROSS_PLATFORM_SKILL_REFERENCES_ROOT, "shell-commands.md"),
+      "utf8",
+    );
+
+    expect(shellCommands).toContain('import { readdir } from "fs/promises";');
+    expect(shellCommands).toContain("const entries = await readdir(directory");
+    expect(shellCommands).not.toContain('execFileAsync("cmd", ["/c", "dir", directory]');
+    expect(shellCommands).not.toContain('execFileAsync("cmd", ["/c", "start"');
+  });
 });