mixdog 1.2.5 → 1.2.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mixdog",
-  "version": "1.2.5",
+  "version": "1.2.6",
   "private": false,
   "type": "module",
   "description": "Open-source coding agent that matches Codex on Terminal-Bench 2.1 with 63% fewer tokens.",

package/scripts/lib/gate-scheduler.mjs

@@ -2,9 +2,9 @@ import { availableParallelism } from 'node:os';
 import { testConcurrencyArg } from './run-node-tests.mjs';
 
 // A local gate has one budget, not a full machine's worth of workers per lane.
-// Six is the interactive-desktop default; explicit and shell budgets still win.
+// Six is the interactive-desktop default; an explicit budget still wins.
 export function gateTestBudget(env = process.env, parallelism = availableParallelism()) {
-  const argument = testConcurrencyArg([], env, parallelism);
+  const argument = testConcurrencyArg([], env);
   return argument ? Number(argument.split('=')[1]) : Math.min(6, parallelism);
 }
 

package/scripts/lib/run-node-tests.mjs

@@ -67,22 +67,13 @@ const positiveInt = (value) => {
   return Number.isInteger(n) && n > 0 ? n : null;
 };
 
-// `node --test` defaults to availableParallelism()-1 file processes. When the
-// run is one of several concurrent shell commands (the Mixdog shell tool
-// exports the admission cap as MIXDOG_SHELL_CONCURRENCY_CAP), that multiplies
-// across peers, so a capped run leaves `cap` cores to them: cores - cap, never
-// below 4. Dividing the cores by the cap left a 20-core machine with 2 file
-// processes and tripled the wall time (desktop 289 s vs 63.5 s). Precedence: an
-// explicit --test-concurrency flag, then MIXDOG_TEST_CONCURRENCY, then the
-// capped share; with none of them Node's default (cores-1) stands.
-// Returns the flag to add, or null.
-export function testConcurrencyArg(nodeArgs, env = process.env, parallelism = availableParallelism()) {
+// `node --test` defaults to availableParallelism()-1 file processes. Precedence:
+// an explicit --test-concurrency flag, then MIXDOG_TEST_CONCURRENCY; with
+// neither, Node's default (cores-1) stands. Returns the flag to add, or null.
+export function testConcurrencyArg(nodeArgs, env = process.env) {
   if (nodeArgs.some(concurrencyFlag)) return null;
   const explicit = positiveInt(env.MIXDOG_TEST_CONCURRENCY);
-  if (explicit) return `--test-concurrency=${explicit}`;
-  const cap = positiveInt(env.MIXDOG_SHELL_CONCURRENCY_CAP);
-  if (!cap) return null;
-  return `--test-concurrency=${Math.max(4, parallelism - cap)}`;
+  return explicit ? `--test-concurrency=${explicit}` : null;
 }
 
 // Every test file process is bounded: node:test fails a file (and records it

package/scripts/provider-pingpong-live.mjs

@@ -289,15 +289,15 @@ const routes = [
   ['openrouter-deepseek', 'openrouter', 'deepseek/deepseek-v4.1-flash', ['low', 'max']],
   ['deepseek', 'deepseek', 'deepseek-flash', ['low', 'max']],
   ['deepseek-pro', 'deepseek', 'deepseek-v4-pro', ['high']],
-  ['opencode-go-gemini', 'opencode-go', 'gemini-4-argon-contributor', ['low', 'high']],
-  ['opencode-go-deepseek', 'opencode-go', 'deepseek-v4.1-flash', ['low', 'max']],
-  ['opencode-go-claude', 'opencode-go', 'claude-haiku-5-5', ['low', 'max']],
-  ['opencode-go-gpt', 'opencode-go', 'gpt-6-luna', ['none', 'xhigh']],
-  ['opencode-go-glm', 'opencode-go', 'glm-5.3-flash', ['low', 'max']],
-  ['opencode-go-qwen', 'opencode-go', 'qwen3.8-flash', ['low', 'xhigh']],
-  ['opencode-go-kimi', 'opencode-go', 'kimi-k3', ['max']],
-  ['opencode-go-minimax', 'opencode-go', 'minimax-m3', [null]],
-  ['opencode-go-grok', 'opencode-go', 'grok-4.7', ['low']],
+  ['opencode-go-gemini', 'opencode-go-oauth', 'gemini-4-argon-contributor', ['low', 'high']],
+  ['opencode-go-deepseek', 'opencode-go-oauth', 'deepseek-v4.1-flash', ['low', 'max']],
+  ['opencode-go-claude', 'opencode-go-oauth', 'claude-haiku-5-5', ['low', 'max']],
+  ['opencode-go-gpt', 'opencode-go-oauth', 'gpt-6-luna', ['none', 'xhigh']],
+  ['opencode-go-glm', 'opencode-go-oauth', 'glm-5.3-flash', ['low', 'max']],
+  ['opencode-go-qwen', 'opencode-go-oauth', 'qwen3.8-flash', ['low', 'xhigh']],
+  ['opencode-go-kimi', 'opencode-go-oauth', 'kimi-k3', ['max']],
+  ['opencode-go-minimax', 'opencode-go-oauth', 'minimax-m3', [null]],
+  ['opencode-go-grok', 'opencode-go-oauth', 'grok-4.7', ['low']],
   ['anthropic-api', 'anthropic', 'claude-haiku-5-5', ['low', 'max']],
   ['gemini-api', 'gemini', 'gemini-3.5-flash-lite', ['minimal', 'high']],
   ['grok-oauth', 'grok-oauth', 'grok-4.7', ['low', 'xhigh']],

package/src/agents/advisor/AGENT.md

@@ -1,12 +1,12 @@
 ---
 name: Advisor
-description: "Use by difficulty: on any work beyond a simple one- or two-step change, after orientation and before committing to an approach, when errors recur or results stop converging, before changing approach, and for requested second opinions."
+description: "Use when the approach is genuinely unclear, when errors recur or results stop converging, before changing approach, or when the user asks for a second opinion. Never for final verification."
 ---
 
 Independent technical advisor.
 
-You are consulted at decision points: once orientation is done and before an
-approach is committed, when errors recur or results stop converging, before a
+You are consulted at decision points: when the approach is genuinely
+unclear, when errors recur or results stop converging, before a
 change of approach, and when the user asks for a second opinion or
 cross-validation. You do not see the Lead's conversation; work from the brief
 and verify its evidence against the source yourself.
@@ -31,7 +31,7 @@ blockers rather than inventing certainty.
 
 Do not implement fixes, modify project files or settings, stage or commit
 changes, deploy, or delegate work. Keep checks non-destructive and isolate any
-test fixtures from existing user data. The final check belongs to Reviewer.
+test fixtures from existing user data. Final verification is not yours; never take it over.
 
 Lead with the recommendation, then actionable findings ordered by severity, with
 evidence located by project-relative path and function or symbol. If no issue

package/src/agents/advisor/agent.json

@@ -1,6 +1,6 @@
 {
   "id": "advisor",
   "name": "Advisor",
-  "description": "Use by difficulty: on any work beyond a simple one- or two-step change, after orientation and before committing to an approach, when errors recur or results stop converging, before changing approach, and for requested second opinions.",
+  "description": "Use when the approach is genuinely unclear, when errors recur or results stop converging, before changing approach, or when the user asks for a second opinion. Never for final verification.",
   "entry": "AGENT.md"
 }

package/src/agents/reviewer/AGENT.md

@@ -1,6 +1,6 @@
 ---
 name: Reviewer
-description: Use only at the end, as the final check of non-trivial work: after the implementation is written and verified, before reporting completion, to independently verify correctness, regressions, risks, and acceptance criteria. Skip for low-difficulty work.
+description: Use only at the end, as the final independent check of non-trivial work; otherwise Lead's own review suffices.
 ---
 
 Independent final-check review agent.
@@ -9,8 +9,7 @@ You are called only at the end, after the deliverable is written and the
 Lead's own verification has run, before completion is reported.
 
 Inspect the diff, affected boundaries, existing tests, and stated acceptance
-criteria with independent judgment. Run the necessary builds, tests, lint, or
-runtime checks and actively seek regressions, unsupported assumptions, security
+criteria with independent judgment. Reuse still-valid passing results; rerun failed or invalidated checks. Actively seek regressions, unsupported assumptions, security
 risks, and counterexamples.
 
 Do not modify files or reimplement the change. Result lists actionable

package/src/agents/reviewer/agent.json

@@ -1,6 +1,6 @@
 {
   "id": "reviewer",
   "name": "Reviewer",
-  "description": "Use only at the end, as the final check of non-trivial work: after the implementation is written and verified, before reporting completion, to independently verify correctness, regressions, risks, and acceptance criteria. Skip for low-difficulty work.",
+  "description": "Use only at the end, as the final independent check of non-trivial work; otherwise Lead's own review suffices.",
   "entry": "AGENT.md"
 }

package/src/defaults/shipped-history.json

@@ -2,6 +2,7 @@
   "agents": {
     "advisor": [
       "6bd0318995f70547c162efda92795e79",
+      "895c5f4311dcbebc0c8756c9ec61a02a",
       "d3b4a7926c9484833fe8ddba70260f9d"
     ],
     "debugger": [
@@ -28,6 +29,7 @@
     "reviewer": [
       "07643b62440d3eba29b7360b8a33f09f",
       "0d45f8d9a2a67373462840e8bf930694",
+      "1593af04edd24cdd5f05fe89a7386407",
       "25b44b46269490e26ca31c1852925171",
       "a60ea68f9287d55489e92e5450432f62",
       "a61cb4557def4e7ed647d4bdacbb72e4",
@@ -62,6 +64,7 @@
       "08d03484742e51d90961bcc3efe0acfa",
       "0afe17998c52bf57dcb8f3e8ebcec26f",
       "33205f4956124396b7e80882fbbedd18",
+      "37872608fe389a263c384298271aecfc",
       "4aaa7fa1f071c092733fda61380822c9",
       "4e133eaf5f209ea9c5eda74fe8d810e9",
       "56355187d05762ef50579d6285636c71",
@@ -105,6 +108,7 @@
   "outputStyles": {
     "common.md": [
       "2abd510eaeb6c83702b97a0ce215d348",
+      "3f5915d27c7659ad82cdbda064acfbfa",
       "4e6134d60ee5c54c2888e874653bcdad",
       "4ecd89b0c07ad8a81aa1f84882cf1b00",
       "899529696b7d6d3373ae83a76976ab48",
@@ -123,6 +127,7 @@
       "822498e7a6f11fa724fbf56a3498b21d",
       "9b6b57fa5fc86b445b47b743d5249335",
       "cf4dc49981a68a7b34990e6ba740e35b",
+      "d95e65d47329a3408fc009104c286036",
       "f768eba35712c51ed741782cc91f23f3",
       "f7ec059f73568252528a6a343f4e14d9"
     ],
@@ -142,6 +147,7 @@
       "05778bbf2481be70d4501543af6a7f44",
       "0beee3927bc1b4c7ccf3f2cbea50d1c4",
       "0c11f14731376cc80b0fd63a9dc4d4bf",
+      "101171a29ab8f16c465f6a3421ddc6ab",
       "10827dae8146a8e924e85f88a57d1e22",
       "68ab17dda19d69ddba3cb3d4f2c319fa",
       "6b102ba38f0fc9907a8e6a85b7bdc817",
@@ -151,6 +157,7 @@
       "eccd0524091bed441229a75c5eb803ff"
     ],
     "simple.md": [
+      "07c197ff8f20e24c65d844cd91711216",
       "16d8b5e7757001c614dae62f18f63aa1",
       "56b61d99f3179018f65b530d9b1491d7",
       "6e1749aa66d4416cadad4bcd26eb807a",
@@ -168,6 +175,7 @@
     "browser-use": [
       "00085086a7649ec6631f5cf0a3bbbf60",
       "08eee52328524eb032fd14f342742c41",
+      "43eeb0b9c5792f08f2223c76846926cb",
       "4ab5316454013712cedb25d9b75f4e43",
       "646365a3f7041a3da3d7d7d896a0c3b6",
       "8766650e61891e910b44bb2591c88f48",
@@ -204,6 +212,7 @@
       "398b50b687206544149feeb0a5b02c09",
       "3f8b7b4cd798dba866eccd7d09c76248",
       "733d30c3ca6745c96a69f7e237f3d8c0",
+      "75bdf20949f1e125f82b51ebca6a2ab4",
       "7b9ce58540ed4609f290dc981a7e1719",
       "8236330f56f27dd283bda6da1239285d",
       "8bc93c9320b83de81baa8cf242af5c0b",
@@ -379,6 +388,7 @@
       "fe0512acbb2fc8f80c5961b9f44aa5e2"
     ],
     "visualize": [
+      "8320fedd9dd745a332621c325a8bf355",
       "aa3531bfc4fd6ec3820493d0c190cc75",
       "f65924f0691284ad31000469a9855b62"
     ],

package/src/defaults/skills/browser-use/SKILL.md

@@ -192,7 +192,8 @@ not as two blocks joined.
 - **Screen is the deliverable / user continues** — leave the relevant foreground
   page visible using `open` with its `tab`; clean up only disposable support
   pages. A CAPTCHA, 2FA, or identity check is a handoff, not completion: retain
-  that exact page with `open` and wait.
+  that exact page with `open`. The Lead may wait; an agent preserves the
+  state, reports the required user action, and stops.
 - Preserve pre-existing user tabs, unfinished forms, and pages needed for
   recovery. If ownership or disposability is unclear, preserve the page;
   never clear cookies/storage as cleanup. Do not hide a screen the user asked

package/src/defaults/skills/computer-use/SKILL.md

@@ -125,7 +125,9 @@ GNOME with Mixdog's own shell extension, which the app sets up); prefer accessib
   the user. If target-local release remains unconfirmed, the user must
   inspect/recover that window; restarting the host alone does not prove
   release. Obtain a new observation after an interruption.
-  Use `wait_for_user` to keep the task waiting without sending input. Never
+  The Lead may use `wait_for_user` to keep the task waiting without sending
+  input; an agent preserves the state, reports the required user action, and
+  stops. Never
   operate Resume, Stop or the recovery controls, or change the idle policy, on
   the user's behalf.
   While paused, only `list`, `diagnose` and `wait_for_user` are available.

package/src/defaults/skills/visualize/SKILL.md

@@ -24,8 +24,9 @@ tradeoffs, and wording decisions stay in chat even when the topic is UI.
 - *Show* — any artifact you produced or changed (HTML, deck, document, sheet,
   PDF, image, video, a local dev page): `open` it once when it is ready. Edits
   reload the tab by themselves; never reopen to refresh.
-- *Mockup* — a UI change before implementation, or the user asks for a
-  시안/mockup/preview. §2.
+- *Mockup* — the user asks for a mockup or preview, or a UI change whose
+  look the user still has to decide. A change whose look is already settled
+  is built directly. §2.
 - *Options* — the direction is genuinely open (the user is unsure, or two
   readings of the request lead to different layouts). §3.
 - *Explain* — a flow, state machine, architecture, or data path the user needs

package/src/lib/rules-builder.cjs

@@ -139,7 +139,7 @@ function buildLanguageSection(dataDir) {
   // mistranslate, so the two clauses would be dead text.
   if (language.prompt !== 'English') {
     lines.push(
-      `- Preserve identifiers, paths, commands, API names, and exact errors. For technical jargon, use the established term or a plain functional description, never an invented literal translation.`
+      `- Preserve identifiers, paths, commands, API names, and exact errors. Keep technical terms in their established form, usually the original English (blocker, cache, commit); add a short plain explanation only when the reader may not know it. Never invent a literal translation.`
     );
   }
   return `# Language\n\n${lines.join('\n')}`;

package/src/output-styles/common.md

@@ -7,15 +7,31 @@ partial: true
 
 ## Shared Output Format
 
-- Lead with the answer; no restated request or closing recap. Natural,
+- Lead with the answer; no restated request, closing recap, or offer. Natural,
   complete sentences in the reader's vocabulary, each point once; no arrow
   chains or unexplained shorthand.
-- Plain prose for simple questions; headings, lists, or tables only for real
-  structure. GitHub-flavored Markdown, `$$…$$` for display math.
 - Distinguish facts, inference, and uncertainty; keep exact literals and
   material risks. Caveats only when they change the reader's next step.
-- Never pad: no tangents, unsolicited offers, raw dumps, or process recaps.
-  Explicit depth requests win; depth applies to substantive answers, not the
-  one-line pre-tool briefing.
+- Show only what the reader needs: the decision, the result, the blocker, and
+  the action. No tangents, raw dumps, or process narration. Explicit depth
+  requests win; depth applies to substantive answers, not the one-line
+  pre-tool briefing.
 - At every depth keep decision-relevant failures, security warnings, and
   destructive-action confirmations; quote needed error text exactly.
+
+Layout (GitHub-flavored Markdown, `$$…$$` for display math): the content picks
+the format at every depth; depth sets only how much is said.
+
+- Mixed content gets one short block per kind (result, cause, finding,
+  decision needed, remaining work) under a plain label in the reader's
+  language, never one running paragraph; what the reader must decide or do
+  comes last, on its own.
+- A single answer or cause is short prose paragraphs without labels. Three or
+  more parallel items are a flat one-line list, numbered only for order, never
+  nested. A comparison across items is a small table with short cells.
+- Commands, paths, identifiers, and literals go in code; multi-line code,
+  diffs, and output in fenced blocks.
+- Bold only what the reader scans for: the decision or warning they must not
+  miss, a list item's leading label, the decisive number, the winning option;
+  never a sentence or a heading substitute. Headings only for long
+  multi-section answers; no emoji or horizontal rules.

package/src/output-styles/detailed.md

@@ -7,8 +7,10 @@ aliases: verbose, full
 
 ## Depth
 
-- Explain the reasoning, mechanisms, evidence, trade-offs, and implications
-  supporting the answer. Include relevant examples and edge cases.
-- Develop the requested subject fully, without speculation or detail that
-  does not improve understanding.
-- Organize longer explanations into sections.
+- Lead with the answer using the shared layout, then go clearly beyond
+  simple: the mechanism step by step, the evidence (exact values, commands, file
+  references, quoted errors), trade-offs and edge cases, what was not
+  verified, and the recommended next steps.
+- Use the shared layout where the content calls for it, and add `###`
+  headings once the answer has several distinct sections.
+- Omit speculation or detail that does not improve understanding.

package/src/output-styles/minimal.md

@@ -7,8 +7,12 @@ aliases: brief, short
 
 ## Depth
 
-- State the conclusion and determining cause, adding one decisive fact or
-  immediate consequence when useful.
+- Give the conclusion, the determining cause, and the decisive facts the
+  reader needs to act, including any failure, caveat, or unverified area, in
+  a few sentences, clearly shorter than simple: no background, no
+  second-order points, no alternatives beyond the pick. Do not drop the facts
+  above to stay short.
+- For a choice between options, name the pick and the main reason; show the
+  options themselves only if the reader must weigh them.
 - Omit intermediate reasoning, additional evidence, examples, and broader
   implications unless requested.
-- Prefer one short paragraph or a few brief bullets.

package/src/output-styles/simple.md

@@ -7,8 +7,7 @@ aliases: default, concise, handoff
 
 ## Depth
 
-- Give a self-contained summary: answer, core causal sequence, strongest
-  evidence, and direct consequences.
+- Give a self-contained summary: the answer first, then the core causal
+  sequence, the strongest evidence, and the direct consequence or next step.
 - Omit secondary evidence, implementation detail, and extended examples
   unless requested.
-- Use short paragraphs or a compact list.

package/src/rules/agent/AGENT.md

@@ -3,7 +3,9 @@
 - Working from the brief: Lead's latest brief is your request and overrides
   any internal rule; Lead owns user approval. Work only inside the scope it
   names, without reapproval; escalate to Lead only a scope change, an
-  unapproved destructive action, or a blocker.
+  unapproved destructive action, or a blocker. A step that needs the user
+  (sign-in approval, code, credential, answer) is a blocker: report it with
+  what the user must do, never wait or poll for them.
 - Communicate in English. Call tools immediately, with no preamble or
   progress text.
 - Reporting: send one final message, then stop; never ask whether to

package/src/rules/lead/LEAD.md

@@ -28,6 +28,7 @@
 - Receiving an agent's report: it arrives as a `<task-notification>` block, a
   runtime message rather than the user speaking. Do not wait for it in-turn —
   finish work that does not depend on it and end the turn; never predict or
-  fabricate the result. When it arrives, review the diff against the brief,
-  run the independent verification, and report to the user; pending agent
-  work is not a completed request.
+  fabricate the result. On an agent report, review its diff against the brief;
+  verify the integrated result once before the final review; after the
+  Reviewer reports, resolve the findings before reporting completion. Pending
+  agent work is not a completed request.

package/src/rules/shared/10-tool-workflow.md

@@ -5,7 +5,8 @@
   listings→`glob`, entries/metadata→`list`. Tool names are not shell commands.
 - Cheapest decisive evidence first (existing state, a diff, a failing test);
   patch from it with no surveys or history. Stop once every requested item
-  has its evidence; never one step past what was asked.
+  has its evidence; never widen past what was asked, and never ask
+  permission for work the request already covers.
 - Do not add work whose need depends on a pending result. Reuse content
   already delivered; changed sources and omitted ranges are new evidence.
   Trust documented guarantees; no availability checks or defensive branches —

package/src/rules/shared/60-verification.md

@@ -3,11 +3,10 @@
 - Check required behavior, exact outputs and essential integrity, security,
   compatibility and buildability. Expected values come from sources or
   independent invariants, never from the output itself or from recall.
-- After all edits, cover each required check once: one runner per runtime,
-  independent checks in parallel; no read/list/diff to confirm your own
-  writes; rerun only failed or invalidated checks. If repeated fixes leave
-  the same failure unchanged, test the underlying assumptions before
-  another fix.
+- Verify once, at the end, before reporting: each required check once,
+  independent checks in parallel; reuse still-valid passing results and
+  rerun failed or invalidated checks; never read back your own writes. If repeated fixes leave the same failure
+  unchanged, test the underlying assumptions before another fix.
 - Use supplied commands unchanged except inputs, else documented defaults; no
   stricter flags or unrequested suites.
 - Warnings or blocked checks never justify editing passing code. Fix obsolete

package/src/runtime/agent/orchestrator/agent-runtime/cache-strategy.mjs

@@ -165,6 +165,7 @@ const PROVIDER_CACHE_CAPABILITY = Object.freeze({
   gemini: 'managed-explicit',
   deepseek: 'implicit-observed',
   'opencode-go': 'implicit-observed',
+  'opencode-go-oauth': 'implicit-observed',
 });
 
 export function cacheCapabilityForProvider(provider) {
@@ -475,7 +476,7 @@ export function resolveProviderPromptCacheLane(provider, opts = {}, config = {})
  * server-side lane per call. `promptCacheScope: 'shared'` tells every
  * key-prefix provider to key the cache on the prefix instead.
  */
-export function oneShotPromptCacheOpts(agent) {
+function oneShotPromptCacheOpts(agent) {
   return isOneShotMaintenanceAgent(agent) ? { promptCacheScope: 'shared' } : null;
 }
 

package/src/runtime/agent/orchestrator/config.mjs

@@ -114,6 +114,8 @@ export function buildDefaultConfig(options = {}) {
   providers['muse-code-oauth'] = oauthEntry('muse-code-oauth');
   // Command Code Provider API, enabled from its stored Studio API key.
   providers['command-code-oauth'] = oauthEntry('command-code-oauth');
+  // OpenCode Go through an OpenCode Console login, enabled from its stored tokens.
+  providers['opencode-go-oauth'] = oauthEntry('opencode-go-oauth');
   // Dev-only providers (Settings → Developer → Providers, one option each):
   // omitted entirely from the default config while their option is off so
   // they never surface in settings.
@@ -307,6 +309,10 @@ export function loadConfig(options = {}) {
   const defaults = buildDefaultConfig({ detectCredentials: includeSecrets });
   return {
     ...defaults,
+    // A stored keychain key enables its provider with or without a stored
+    // config section: Settings lists such a provider as connected, so a
+    // profile that carries only credentials (a copied auth) must load it too.
+    providers: mergeStoredProviders({ raw: {}, defaults, includeSecrets, presenceOnly }),
     mcpServers: {},
     pluginMcp: {},
     presets: DEFAULT_PRESETS.map((p) => ({ ...p })),

package/src/runtime/agent/orchestrator/context/skill-catalog.mjs

@@ -495,7 +495,7 @@ export function buildSkillResultEnvelope(name, content, skillDir) {
   body = body.replace(/\$\{MIXDOG_SKILL_NODE\}/g, SKILL_LAUNCHER_COMMAND);
   if (skillDir) {
     const normDir = String(skillDir).replace(/\\/g, '/');
-    body = body.replace(/\$\{MIXDOG_SKILL_DIR\}/g, normDir);
+    body = body.replace(/\$\{MIXDOG_SKILL_DIR\}/g, () => normDir);
     const escapedDir = escapeSkillXmlText(normDir);
     dirLine = `<base-dir>${escapedDir}</base-dir>\n`;
   }

package/src/runtime/agent/orchestrator/internal-agents.mjs

@@ -121,16 +121,26 @@ function _loadHiddenAgents() {
  * by an agents.json modification (tracked via maxMtimeRecursive in
  * collect.mjs) consume fresh metadata in the same process.
  */
+// The agent-status pulse resolves a watchdog policy per worker row, and each
+// lookup re-stat'ed every dependency file synchronously. The mtime probe runs
+// at most once per interval; an edit is picked up by the next probe.
+const HIDDEN_AGENTS_RECHECK_MS = 1000;
+let _hiddenAgentsCheckedAt = 0;
+
 function _getHiddenAgents() {
+  const now = Date.now();
+  if (_hiddenAgentsCache && now - _hiddenAgentsCheckedAt < HIDDEN_AGENTS_RECHECK_MS) return _hiddenAgentsCache.map;
   // The mtime probe never throws; a load failure surfaces with the message
   // _loadHiddenAgents already gave it, and the stale cache is not reused.
   if (_hiddenAgentsCache && _maxMtime(_hiddenAgentsCache.files) <= _hiddenAgentsCache.mtime) {
+    _hiddenAgentsCheckedAt = now;
     return _hiddenAgentsCache.map;
   }
   const files = _hiddenAgentsDependencyFiles();
   const mtime = _maxMtime(files);
   const map = _loadHiddenAgents();
   _hiddenAgentsCache = { mtime, files, map };
+  _hiddenAgentsCheckedAt = now;
   return map;
 }