perf(codex): skip recall for control-only prompts

examples/codex-memory-plugin/DESIGN.md

@@ -175,6 +175,13 @@ model turn. Transcript capture may later see that injected context
 adjacent to the prompt, so plugin-generated recall and resume context are
 wrapped in a deterministic boundary:
 
+Before the health check or any search request, strict control-only prompts are
+short-circuited to `{}`. This covers bare acknowledgements, continuations,
+status checks, and fingerprinted approvals. The classifier is
+deliberately exact: prompts with task nouns, constraints, or implementation
+details continue through normal recall. Set
+`OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT=0` to disable it.
+
 ```text
 <openviking-context source="auto-recall" format="digest">
 OpenViking memory digest:
@@ -250,6 +257,7 @@ Env var overrides for tuning without rebuilding:
 | `OPENVIKING_CODEX_ACTIVE_WINDOW_MS` | `120000` (2 min) | rule-3 active window |
 | `OPENVIKING_CODEX_IDLE_TTL_MS` | `1800000` (30 min) | idle sweep TTL |
 | `OPENVIKING_RECALL_TIMEOUT_MS` | `120000` (2 min) | whole UserPromptSubmit auto-recall deadline |
+| `OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT` | `1` | skip external recall for strict control-only prompts; set `0` to disable |
 | `OPENVIKING_RECALL_COMPRESS` | `1` | set `0` / `off` to skip `codex exec` compression |
 | `OPENVIKING_RECALL_COMPRESS_MODEL` | unset | custom first-choice compressor model; `off` disables compression |
 | `OPENVIKING_RECALL_COMPRESS_THINKING` | unset | custom `model_reasoning_effort`; `default` means omit override; alias `OPENVIKING_RECALL_COMPRESS_REASONING_EFFORT` |

examples/codex-memory-plugin/README.md

@@ -202,7 +202,7 @@ On `resume`, the script skips commit/sweep. It still injects the profile block.
 { "hookSpecificOutput": { "hookEventName": "UserPromptSubmit", "additionalContext": "<openviking-context source=\"auto-recall\" format=\"digest\">\nOpenViking memory digest:\n- ...\n</openviking-context>" } }
 ```
 
-Codex injects `additionalContext` into the model turn, so memories arrive without an extra tool call. By default the hook runs a Codex compression pass over recalled candidates before injection, dropping weakly-related memories and preserving only a short digest. If the compressor returns `NO_RELEVANT_MEMORY`, empty text, or non-digest chatter, the hook emits `{}` and injects nothing. The whole hook has its own `OPENVIKING_RECALL_TIMEOUT_MS` deadline (default 120s); the bundled `hooks.json` gives Codex 130s so the script can return `{}` before Codex kills it. Digests may keep `viking://` source URIs and point the model at the OpenViking MCP `read`/`search` tools for details when the inline bullet is intentionally short. The outer `<openviking-context ...>` wrapper is deterministic, not compressor-generated; capture strips it to distinguish recalled context from the user's prompt. Set `OPENVIKING_RECALL_COMPRESS=0` to fall back to deterministic short formatting.
+Codex injects `additionalContext` into the model turn, so memories arrive without an extra tool call. Before contacting OpenViking, the hook skips strict control-only prompts such as a bare acknowledgement, continuation, status request, or fingerprinted approval. Prompts containing an extra task noun, constraint, or implementation detail still take the normal recall path. By default the hook runs a Codex compression pass over recalled candidates before injection, dropping weakly-related memories and preserving only a short digest. If the compressor returns `NO_RELEVANT_MEMORY`, empty text, or non-digest chatter, the hook emits `{}` and injects nothing. The whole hook has its own `OPENVIKING_RECALL_TIMEOUT_MS` deadline (default 120s); the bundled `hooks.json` gives Codex 130s so the script can return `{}` before Codex kills it. Digests may keep `viking://` source URIs and point the model at the OpenViking MCP `read`/`search` tools for details when the inline bullet is intentionally short. The outer `<openviking-context ...>` wrapper is deterministic, not compressor-generated; capture strips it to distinguish recalled context from the user's prompt. Set `OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT=0` to disable the control-prompt shortcut, or `OPENVIKING_RECALL_COMPRESS=0` to fall back to deterministic short formatting.
 
 The compressor profile is recreated on every `SessionStart` and cached under `OPENVIKING_CODEX_STATE_DIR` so cross-session config changes are picked up but each `UserPromptSubmit` does not probe models. Default fallback order:
 
@@ -216,6 +216,7 @@ Config knobs:
 | Env var | Default | Meaning |
 |---|---|---|
 | `OPENVIKING_RECALL_LIMIT` | `10` | Legacy quota-scaling input; explicit values are converted to six coding quotas, not enforced as a final result cap. |
+| `OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT` | `1` | Skip external recall for strict control-only prompts; set `0` to disable. |
 | `OPENVIKING_RECALL_COMPRESS` | `1` | Set `0` / `off` to disable `codex exec` compression. |
 | `OPENVIKING_RECALL_COMPRESS_MODEL` | unset | Custom first-choice compressor model. Set `off` to disable compression. |
 | `OPENVIKING_RECALL_COMPRESS_THINKING` | unset | Custom `model_reasoning_effort`; `default` omits the Codex config override. Alias: `OPENVIKING_RECALL_COMPRESS_REASONING_EFFORT`. |

examples/codex-memory-plugin/scripts/auto-recall.mjs

@@ -17,6 +17,7 @@ import { mkdtemp, readFile, rm } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { loadConfig } from "./config.mjs";
+import { classifyControlPrompt } from "./control-prompt.mjs";
 import { trySpawnCodex } from "./codex-launch.mjs";
 import { createLogger } from "./debug-log.mjs";
 import {
@@ -587,6 +588,15 @@ async function main() {
     return;
   }
 
+  const controlPromptKind = cfg.recallControlPromptShortCircuit
+    ? classifyControlPrompt(userPrompt)
+    : null;
+  if (controlPromptKind) {
+    log("skip", { stage: "query_check", reason: "control prompt", kind: controlPromptKind });
+    emit();
+    return;
+  }
+
   const health = await fetchJSON("/health");
   if (!health.ok) {
     logError("health_check", "server unreachable or unhealthy");

examples/codex-memory-plugin/scripts/auto-recall.test.mjs

@@ -7,6 +7,7 @@ import { dirname, join } from "node:path";
 import test from "node:test";
 import { fileURLToPath } from "node:url";
 import { resolveCodexLaunch, trySpawnCodex } from "./codex-launch.mjs";
+import { classifyControlPrompt } from "./control-prompt.mjs";
 
 const SCRIPT_DIR = dirname(fileURLToPath(import.meta.url));
 
@@ -200,6 +201,82 @@ async function runEndpointCompressionCase({
   }
 }
 
+test("control prompt classifier is strict about prompts with task details", () => {
+  const controls = [
+    ["确认", "acknowledgement"],
+    ["继续执行", "continuation"],
+    ["现在什么进度？", "status"],
+    ["批准 PUBLISH_APPROVED，指纹 66a86081040c620b9ec92e52b22ca7a74d62f634fc921264030414943b1b2eae", "approval"],
+  ];
+  for (const [prompt, kind] of controls) assert.equal(classifyControlPrompt(prompt), kind);
+
+  for (const prompt of [
+    "继续优化 OpenViking 的召回算法",
+    "批准预算为 100 美元并导入历史记录",
+    "现在 OpenViking 的召回机制是什么",
+    "please continue the migration but keep Python 3.10 compatibility",
+    "这个 PR 完成了吗？如果没有请修复 CI",
+  ]) {
+    assert.equal(classifyControlPrompt(prompt), null, prompt);
+  }
+});
+
+test("auto-recall short-circuits control prompts before any HTTP request", async () => {
+  let requestCount = 0;
+  await withMockOpenViking(async (_req, res) => {
+    requestCount += 1;
+    writeJson(res, { status: "ok", result: { ok: true } });
+  }, async (baseUrl) => {
+    const result = await runAutoRecall(
+      { prompt: "继续执行", session_id: "codex:control" },
+      {
+        OPENVIKING_AUTO_RECALL: "1",
+        OPENVIKING_CREDENTIAL_SOURCE: "env",
+        OPENVIKING_MIN_QUERY_LENGTH: "1",
+        OPENVIKING_RECALL_TIMEOUT_MS: "10000",
+        OPENVIKING_URL: baseUrl,
+      },
+    );
+    assert.deepEqual(JSON.parse(result.stdout.trim()), {});
+  });
+  assert.equal(requestCount, 0);
+});
+
+test("control prompt short-circuit can be disabled", async () => {
+  let requestCount = 0;
+  await withMockOpenViking(async (req, res) => {
+    requestCount += 1;
+    const url = new URL(req.url, "http://127.0.0.1");
+    if (req.method === "GET" && url.pathname === "/health") {
+      writeJson(res, { status: "ok", result: { ok: true } });
+      return;
+    }
+    if (req.method === "POST" && url.pathname === "/api/v1/search/search") {
+      writeJson(res, { status: "ok", result: { entries: [], rendered: "" } });
+      return;
+    }
+    if (req.method === "POST" && url.pathname === "/api/v1/search/recall") {
+      writeJson(res, { status: "ok", result: { entries: [], rendered: "" } });
+      return;
+    }
+    writeStatusJson(res, 404, { status: "error", error: "not found" });
+  }, async (baseUrl) => {
+    const result = await runAutoRecall(
+      { prompt: "继续执行", session_id: "codex:control-disabled" },
+      {
+        OPENVIKING_AUTO_RECALL: "1",
+        OPENVIKING_CREDENTIAL_SOURCE: "env",
+        OPENVIKING_MIN_QUERY_LENGTH: "1",
+        OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT: "0",
+        OPENVIKING_RECALL_TIMEOUT_MS: "10000",
+        OPENVIKING_URL: baseUrl,
+      },
+    );
+    assert.deepEqual(JSON.parse(result.stdout.trim()), {});
+  });
+  assert.ok(requestCount > 0);
+});
+
 test("auto-recall asks the context face with the derived OpenViking session id", async () => {
   const stateDir = await mkdtemp(join(tmpdir(), "ov-auto-recall-state-"));
   const requests = [];

examples/codex-memory-plugin/scripts/config.mjs

@@ -33,6 +33,7 @@
  *   OPENVIKING_RECALL_TIMEOUT_MS, OPENVIKING_RECALL_COMPRESS_TIMEOUT_MS
  *   OPENVIKING_RECALL_COMPRESS_MODEL, OPENVIKING_RECALL_COMPRESS_THINKING
  *   OPENVIKING_RECALL_LIMIT, OPENVIKING_SCORE_THRESHOLD
+ *   OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT
  *   OPENVIKING_WORKSPACE_PEER, OPENVIKING_RECALL_PEER_SCOPE
  *   OPENVIKING_NO_AUTO_INJECT, OPENVIKING_PROFILE_TOKEN_BUDGET
  *   OPENVIKING_DEBUG=1, OPENVIKING_DEBUG_LOG
@@ -180,6 +181,9 @@ export function loadConfig() {
       process.env.OPENVIKING_MIN_QUERY_LENGTH,
       num(cx.minQueryLength, 3),
     ))),
+    recallControlPromptShortCircuit:
+      envBool("OPENVIKING_RECALL_CONTROL_PROMPT_SHORT_CIRCUIT") ??
+      configBool(cx.recallControlPromptShortCircuit, true),
     logRankingDetails: envBool("OPENVIKING_LOG_RANKING_DETAILS") ?? (cx.logRankingDetails === true),
     recallPeerScope,
     recallCompress: envBool("OPENVIKING_RECALL_COMPRESS") ?? configBool(cx.recallCompress, true),

examples/codex-memory-plugin/scripts/control-prompt.mjs

@@ -0,0 +1,24 @@
+const APPROVAL_FINGERPRINT_RE = /(?:指纹|fingerprint)\s*[:：]?\s*[a-f0-9]{64}\b/i;
+const APPROVAL_PREFIX_RE = /^(?:批准|同意|approve(?:d)?)(?:\s|[,:，：])/iu;
+
+const ACKNOWLEDGEMENT_RE = /^(?:ok(?:ay)?|yes|yep|sure|confirmed?|approved?|sounds good|lgtm|好|好的|可以|没问题|确认|同意|批准|行|收到|就这样)[。.!！]?$/iu;
+const CONTINUATION_RE = /^(?:continue|proceed|go ahead|keep going|do it|start|继续|继续执行|接着做|开始吧|执行吧|按计划继续|就按这个做|按这个做)[。.!！]?$/iu;
+const STATUS_RE = /^(?:status|progress|(?:any|status) update|what(?:'s| is) the (?:current )?(?:status|progress)|is it done|are we done|进度|现在进度怎么样|进度怎么样|什么进度|现在什么进度|汇报(?:一下)?进度|完成了吗|整体完成了吗|现在整体完成了吗)[?？。.!！]?$/iu;
+
+/**
+ * Classify prompts whose meaning is entirely carried by the active turn state.
+ * These prompts should not trigger external memory retrieval: the model already
+ * has the approval, continuation, or status request in the conversation.
+ *
+ * The match is intentionally strict. Any extra task noun, constraint, or
+ * implementation detail falls through to normal recall.
+ */
+export function classifyControlPrompt(prompt) {
+  const text = String(prompt || "").normalize("NFKC").replace(/\s+/g, " ").trim();
+  if (!text || text.length > 512) return null;
+  if (APPROVAL_PREFIX_RE.test(text) && APPROVAL_FINGERPRINT_RE.test(text)) return "approval";
+  if (ACKNOWLEDGEMENT_RE.test(text)) return "acknowledgement";
+  if (CONTINUATION_RE.test(text)) return "continuation";
+  if (STATUS_RE.test(text)) return "status";
+  return null;
+}