oh-my-opencode 5.0.0-beta.9 → 5.0.0

package/dist/skills/visual-qa/AGENTS.md

@@ -0,0 +1,58 @@
+# visual-qa — Bundled Visual-Evidence CLI
+
+**Generated:** 2026-08-24 (f3642fcda)
+
+## OVERVIEW
+
+The visual-QA skill's executable core: a zero-dependency Node CLI that produces machine evidence (`image-diff`, `tui-check`) plus the TypeScript sources it is bundled from. Earned this file: the only shared skill shipping a built bundle whose runtime and development sources must be kept in lockstep.
+
+## STRUCTURE
+
+```
+visual-qa/
+├── SKILL.md                      # reviewer workflow (dual oracle, evidence gates) — the prose contract
+├── references/browser-setup.md
+└── scripts/
+    ├── visual-qa.mjs             # SHIPPED RUNTIME: bun-build bundle of cli.ts (+ embedded modules)
+    ├── cli.ts                    # development source; dispatches image-diff / tui-check
+    ├── image-diff.ts             # diffImages — 8x8 grid cells, hotspots, rounded metrics
+    ├── tui-grid.ts               # checkTui — column overflow + border alignment verdicts
+    ├── east-asian-width.ts       # charWidth / stringWidth — CJK-aware width math
+    ├── ansi.ts                   # ANSI escape stripping (width must be escape-aware)
+    ├── png-decode.ts / png-crc.ts / png-synth.ts   # stdlib PNG codec (no deps)
+    ├── types.ts
+    └── *.test.ts                 # co-located bun tests for every module above
+```
+
+## BUNDLE/SOURCE DUALITY
+
+`visual-qa.mjs` is the artifact the skill's commands invoke (`node "$SKILL_DIR/scripts/visual-qa.mjs" …`); the `.ts` files exist for development and tests only. A behavior fix lands in the TS source AND the bundle is regenerated from `cli.ts` — editing one without the other makes tests and runtime disagree. Never hand-edit the `.mjs`.
+
+## SURFACE
+
+- Commands: `image-diff <reference.png> <actual.png>` and `tui-check <capture.txt> --cols <N>`; JSON verdict output (hotspot cells, overflow lines, wide-char columns, border alignment).
+- Module exports: `diffImages`, `checkTui`, `charWidth`/`stringWidth`, ANSI + PNG helpers. `cli.ts` owns `CliError`, `parseColumns`, `run`, `main`.
+
+## CONVENTIONS
+
+- Zero runtime dependencies — PNG decode/synthesis and ANSI handling are hand-rolled in `scripts/`; keep it that way (the bundle must stay require-free apart from `node:` builtins).
+- Width is never `String.length`: CJK wide characters and ANSI escapes are accounted for before any column math.
+- Tests are co-located `bun test scripts/*.test.ts`, given/when/then style per repo convention.
+
+## ANTI-PATTERNS
+
+- NEVER hand-edit `visual-qa.mjs`; regenerate from source.
+- Don't add an npm dependency to make PNG/diff easier — the no-dep bundle is the point.
+- Column math that ignores CJK width or ANSI escapes is wrong by construction; `tui-check` exists because text-diff tools get terminal grids wrong.
+- Skill-level rule worth repeating for code changes: `tmux capture-pane` is forbidden as a capture mechanism.
+
+## COMMANDS
+
+```bash
+# from packages/shared-skills/skills/visual-qa/
+node scripts/visual-qa.mjs image-diff <reference.png> <actual.png>
+node scripts/visual-qa.mjs tui-check <capture.txt> --cols 80
+bun test scripts/*.test.ts
+```
+
+- Parent: [`packages/shared-skills/AGENTS.md`](../../AGENTS.md).

package/dist/skills/visual-qa/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: visual-qa
-description: "MUST USE after building/changing any UI or when asked whether a page, component, or TUI looks right. Rigorous visual QA across web/page and terminal UIs. Prefer browser:control-in-app-browser for unauthenticated browser/page QA in Codex, then Playwright/agent-browser/dev-browser. Captures screenshot/TUI evidence with bundled diff scripts, runs design-system/functional and visual-fidelity/CJK reviewer passes, then synthesizes a good/bad verdict. Triggers: visual QA, screenshot/pixel diff, UI looks wrong, reference fidelity, design system check, responsive check, CJK text clipping, TUI alignment, box-drawing drift."
+description: "Runs rigorous visual QA across web, terminal, and paginated surfaces with screenshot evidence and a verdict. Use for any UI build or change, or when asked whether a page, component, or TUI looks right."
 ---
 
 # Visual QA - Dual-Oracle Web and TUI Verification
@@ -9,7 +9,7 @@ Verify a rendered UI against intent using objective script evidence plus two par
 
 ## Purpose and when to use
 
-- Use after you build or change any UI, before calling it done. Covers web/page UIs and TUI/terminal UIs.
+- Use after you build or change any UI, before calling it done. Covers web/page UIs, TUI/terminal UIs, and paginated documents.
 - Use when output must match a mock, a baseline, or a stated design intent; when you suspect a regression; when CJK (Korean/Japanese/Chinese) text may clip, misalign, or wrap awkwardly; when a claimed design system might actually be a flat image; when a terminal layout may overflow or its borders may break.
 - Skip when there is no rendered surface (pure backend or library logic with no visual or terminal output). For broad post-implementation review use review-work; this skill is the visual specialist.
 
@@ -19,6 +19,7 @@ In the commands below, `$SKILL_DIR` is this skill's own directory (the folder co
 
 - Web/page UI: renders in a browser (HTML/CSS/JS, components, canvas, SVG). Evidence is screenshots.
 - TUI/terminal UI: renders as text in a terminal (box-drawing, panes, status lines, REPL/TUI apps). Evidence is terminal captures.
+- Paginated document: renders as ordered pages (PDF report, printed HTML, exported deck). Evidence is every page rendered to an image - `pdftoppm -png -r 150 <file>.pdf <prefix>` or the pipeline's own renderer. Extracted text is not evidence here: layout is exactly what extraction discards, so a stranded block, a split table, or a near-empty page survives a clean text check.
 - Reference-fidelity UI: any web/page UI built from a concrete reference packet, including screenshots, generated Imagen/Stitch mockups, Figma exports, overview text, annotations, or source-site captures. Evidence is the full reference packet plus same-size actual captures.
 
 If the change touches both, run both capture tracks and feed both into the passes.
@@ -48,7 +49,7 @@ Before any reviewer sees an image, verify each capture yourself: the file signat
 ### Web
 
 1. Capture a REFERENCE image: the user's mock/target, generated page snapshot, Figma export, source-site capture, or known-good baseline. Save as PNG. If the user provided overview text or annotations, save them next to the image and treat them as part of the reference packet.
-2. Capture the ACTUAL rendered screenshot at the same viewport size. In Codex, when `browser:control-in-app-browser` is available and the page does not need an authenticated user browser session, use that Browser plugin first for navigation, page state inspection, and screenshots. If it is unavailable or lacks the needed capture action, use the project's configured browser tooling (the playwright, agent-browser, or dev-browser skill). Save as PNG. If none is configured or available, install [agent-browser](https://github.com/vercel-labs/agent-browser) (`npm install -g agent-browser && agent-browser install`) and capture with it — see `$SKILL_DIR/references/agent-browser-setup.md` for the full setup, including how to shoot a fixed-viewport screenshot.
+2. Capture the ACTUAL rendered screenshot at the reference viewport with omowright from js eval (the library is staged in the `browser` skill): the owned engine (`connectPipe` on a task-owned profile, viewport pinned with `emulate`, then `page.screenshot()`) for anything unauthenticated, or the attached engine (`connectBrowserSkill()` → `session.screenshot()`) when the page needs the user's login — never a clone of, or a launch against, the user's live profile. Save PNG and return its path; close the browser or stop the session. See `$SKILL_DIR/references/browser-setup.md` for fixed-viewport examples and prerequisites.
 3. Run the diff and keep the JSON:
 
 ```

package/dist/skills/visual-qa/references/browser-setup.md

@@ -0,0 +1,77 @@
+# Browser setup (Web capture)
+
+Capture with omowright from the js-eval kernel. The library is staged inside the `browser` skill;
+load it once per session:
+
+```js
+const { loadOmowright } = await import("<browser-skill-root>/scripts/omowright.mjs")
+const { omowright } = await loadOmowright()
+```
+
+## Owned engine (default for QA)
+
+A browser your code launches, with a task-owned profile, pinned viewport and no user state.
+`connectPipe` opens no listening port and reaps the process on `close()`.
+
+```js
+// js-eval cell; url and pngPath belong to this QA run.
+const { mkdtempSync, rmSync } = await import("node:fs")
+const profile = mkdtempSync(`${(await import("node:os")).tmpdir()}/visual-qa-`)
+const browser = await omowright.connectPipe({
+  browserPath: chromeBinary,                      // installed Chrome, Chromium, CloakBrowser or chrome-headless-shell
+  browserArgs: ["--headless", "--no-first-run", `--user-data-dir=${profile}`],
+  storageRoot: profile,
+})
+try {
+  const page = await browser.newTab("about:blank")
+  await omowright.emulate(page, { width: 1280, height: 720, deviceScaleFactor: 1, mobile: false, hasTouch: false })
+  await page.goto(url, { waitUntil: "load" })
+  await Bun.write(pngPath, await page.screenshot())
+  console.log(pngPath)
+} finally {
+  await browser.close()
+  rmSync(profile, { recursive: true, force: true })
+}
+```
+
+Chrome must already be installed; report an absent executable rather than downloading a managed
+browser. For bot-scored or WAF targets use `connectCloakProfile({ profileDir })` (CloakBrowser
+with a pinned fingerprint seed) — the `browser` skill's `references/owned-engine/README.md`
+covers it.
+
+## Attached engine (authenticated pages)
+
+When the capture needs the user's login, drive the browser they are signed into instead of
+cloning its profile:
+
+```js
+const session = await omowright.connectBrowserSkill({ name: "visual-qa capture", focused: false })
+try {
+  await session.navigate(url, { waitUntil: "load" })
+  await session.resize(1280, 720)
+  const shot = await session.screenshot()            // { buffer, width, height }
+  await Bun.write(pngPath, shot.buffer)
+} finally {
+  await session.stop()
+}
+```
+
+NEVER launch anything against, or clear cookies/cache/site data from, the user's live profile;
+the attached engine is the only sanctioned way to a signed-in page. If no extension is connected,
+run the `browser` skill's `scripts/browser-install.mjs` for the browser the user actually uses
+(from memory, or its detection; on `needsChoice` ask them and pass `--browser=<id>`), relay its
+one human step, and wait — do
+not fall back to the owned engine for an authenticated criterion.
+
+## Capture a screenshot at a fixed viewport
+
+Match CSS viewport AND PNG dimensions: pin `deviceScaleFactor` through `emulate` (owned) or
+`resize` (attached) instead of resizing the PNG to force a pass. Wait for the specific page state
+(a locator, a `waitForURL`, a `createNetworkSnoop(page).waitFor(...)`), not a sleep, then compare:
+
+```sh
+node "$SKILL_DIR/scripts/visual-qa.mjs" image-diff reference.png actual.png
+```
+
+Inspect `dimensionsMatch` and `diffRatio`, then inspect the image. Close every browser and session
+and the fixture server, even on a failed capture; remove the owned profile in the same `finally`.

package/dist/tools/delegate-task/builtin-categories.d.ts

@@ -2,5 +2,7 @@ import type { CategoryConfig } from "../../config/schema";
 export declare const DEFAULT_CATEGORIES: Record<string, CategoryConfig>;
 export declare const CATEGORY_PROMPT_APPENDS: Record<string, string>;
 export declare const CATEGORY_DESCRIPTIONS: Record<string, string>;
+export declare const CATEGORY_CALLER_GUIDANCE: Record<string, string | undefined>;
 export declare const CATEGORY_PROMPT_APPEND_RESOLVERS: Record<string, (model: string | undefined) => string>;
-export declare const BUILTIN_CATEGORY_REQUIRES_MODEL: Record<string, string>;
+export declare const BUILTIN_CATEGORY_REQUIRES_MODEL: Record<string, readonly string[]>;
+export declare function builtinCategoryGateModels(categoryName: string, requirementGateModel: string | undefined): readonly string[];

package/dist/tools/delegate-task/builtin-category-definition.d.ts

@@ -3,7 +3,8 @@ export type BuiltinCategoryDefinition = {
     name: string;
     config: CategoryConfig;
     description: string;
+    callerGuidance?: string;
     promptAppend: string;
     resolvePromptAppend?: (model: string | undefined) => string;
-    requiresModel?: string;
+    requiresModel?: string | readonly string[];
 };

package/dist/tools/delegate-task/constants.d.ts

@@ -1,5 +1,5 @@
 import type { AvailableCategory, AvailableSkill } from "../../agents/dynamic-agent-prompt-builder";
-export { BUILTIN_CATEGORY_REQUIRES_MODEL, CATEGORY_DESCRIPTIONS, CATEGORY_PROMPT_APPENDS, CATEGORY_PROMPT_APPEND_RESOLVERS, DEFAULT_CATEGORIES, } from "./builtin-categories";
+export { BUILTIN_CATEGORY_REQUIRES_MODEL, builtinCategoryGateModels, CATEGORY_DESCRIPTIONS, CATEGORY_PROMPT_APPENDS, CATEGORY_PROMPT_APPEND_RESOLVERS, DEFAULT_CATEGORIES, } from "./builtin-categories";
 /**
  * System prompt prepended to plan agent invocations.
  * Instructs the plan agent to first gather context via explore/librarian agents,
@@ -7,7 +7,7 @@ export { BUILTIN_CATEGORY_REQUIRES_MODEL, CATEGORY_DESCRIPTIONS, CATEGORY_PROMPT
  * Also MANDATES dependency graphs, parallel execution analysis, and category+skill recommendations.
  */
 export declare const PLAN_AGENT_SYSTEM_PREPEND_STATIC_BEFORE_SKILLS = "<system>\nBEFORE you begin planning, you MUST first understand the user's request deeply.\n\nMANDATORY CONTEXT GATHERING PROTOCOL:\n1. Launch background agents to gather context:\n   - call_omo_agent(description=\"Explore codebase patterns\", subagent_type=\"explore\", run_in_background=true, prompt=\"<search for relevant patterns, files, and implementations in the codebase related to user's request>\")\n   - call_omo_agent(description=\"Research documentation\", subagent_type=\"librarian\", run_in_background=true, prompt=\"<search for external documentation, examples, and best practices related to user's request>\")\n\n2. After gathering context, ALWAYS present:\n   - **User Request Summary**: Concise restatement of what the user is asking for\n   - **Uncertainties**: List of unclear points, ambiguities, or assumptions you're making\n   - **Clarifying Questions**: Specific questions to resolve the uncertainties\n\n3. ITERATE until ALL requirements are crystal clear:\n   - Do NOT proceed to planning until you have 100% clarity\n   - Ask the user to confirm your understanding\n   - Resolve every ambiguity before generating the work plan\n\nREMEMBER: Vague requirements lead to failed implementations. Take the time to understand thoroughly.\n</system>\n\n<CRITICAL_REQUIREMENT_DEPENDENCY_PARALLEL_EXECUTION_CATEGORY_SKILLS>\n#####################################################################\n#                                                                   #\n#   \u2588\u2588\u2588\u2588\u2588\u2588\u2557 \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2557 \u2588\u2588\u2588\u2588\u2588\u2588\u2557 \u2588\u2588\u2557   \u2588\u2588\u2557\u2588\u2588\u2557\u2588\u2588\u2588\u2588\u2588\u2588\u2557 \u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2557\u2588\u2588\u2588\u2588\u2588\u2588\u2557    #\n#   \u2588\u2588\u2554\u2550\u2550\u2588\u2588\u2557\u2588\u2588\u2554\u2550\u2550\u2550\u2550\u255D\u2588\u2588\u2554\u2550\u2550\u2550\u2588\u2588\u2557\u2588\u2588\u2551   \u2588\u2588\u2551\u2588\u2588\u2551\u2588\u2588\u2554\u2550\u2550\u2588\u2588\u2557\u2588\u2588\u2554\u2550\u2550\u2550\u2550\u255D\u2588\u2588\u2554\u2550\u2550\u2588\u2588\u2557   #\n#   \u2588\u2588\u2588\u2588\u2588\u2588\u2554\u255D\u2588\u2588\u2588\u2588\u2588\u2557  \u2588\u2588\u2551   \u2588\u2588\u2551\u2588\u2588\u2551   \u2588\u2588\u2551\u2588\u2588\u2551\u2588\u2588\u2588\u2588\u2588\u2588\u2554\u255D\u2588\u2588\u2588\u2588\u2588\u2557  \u2588\u2588\u2551  \u2588\u2588\u2551   #\n#   \u2588\u2588\u2554\u2550\u2550\u2588\u2588\u2557\u2588\u2588\u2554\u2550\u2550\u255D  \u2588\u2588\u2551\u2584\u2584 \u2588\u2588\u2551\u2588\u2588\u2551   \u2588\u2588\u2551\u2588\u2588\u2551\u2588\u2588\u2554\u2550\u2550\u2588\u2588\u2557\u2588\u2588\u2554\u2550\u2550\u255D  \u2588\u2588\u2551  \u2588\u2588\u2551   #\n#   \u2588\u2588\uFFFD\uFFFD  \u2588\u2588\u2551\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2557\u255A\u2588\u2588\u2588\u2588\u2588\u2588\u2554\u255D\u255A\u2588\u2588\u2588\u2588\u2588\u2588\u2554\u255D\u2588\u2588\u2551\u2588\u2588\u2551  \u2588\u2588\u2551\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2557\u2588\u2588\u2588\u2588\u2588\u2588\u2554\u255D   #\n#   \u255A\u2550\u255D  \u255A\u2550\u255D\u255A\u2550\u2550\u2550\u2550\u2550\u2550\u255D \u255A\u2550\u2550\u2580\u2580\u2550\u255D  \u255A\u2550\u2550\u2550\u2550\u2550\u255D \u255A\u2550\u255D\u255A\u2550\u255D  \u255A\u2550\u255D\u255A\u2550\u2550\u2550\u2550\u2550\u2550\u255D\u255A\u2550\u2550\u2550\u2550\u2550\u255D    #\n#                                                                   #\n#####################################################################\n\nYOU MUST INCLUDE THE FOLLOWING SECTIONS IN YOUR PLAN OUTPUT.\nTHIS IS NON-NEGOTIABLE. FAILURE TO INCLUDE THESE SECTIONS = INCOMPLETE PLAN.\n\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 SECTION 1: TASK DEPENDENCY GRAPH (MANDATORY)                    \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYOU MUST ANALYZE AND DOCUMENT TASK DEPENDENCIES.\n\nFor EVERY task in your plan, you MUST specify:\n- Which tasks it DEPENDS ON (blockers)\n- Which tasks DEPEND ON IT (dependents)\n- The REASON for each dependency\n\nExample format:\n```\n## Task Dependency Graph\n\n| Task | Depends On | Reason |\n|------|------------|--------|\n| Task 1 | None | Starting point, no prerequisites |\n| Task 2 | Task 1 | Requires output/artifact from Task 1 |\n| Task 3 | Task 1 | Uses same foundation established in Task 1 |\n| Task 4 | Task 2, Task 3 | Integrates results from both tasks |\n```\n\nWHY THIS MATTERS:\n- Executors need to know execution ORDER\n- Prevents blocked work from starting prematurely\n- Identifies critical path for project timeline\n\n\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 SECTION 2: PARALLEL EXECUTION GRAPH (MANDATORY)                 \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYOU MUST IDENTIFY WHICH TASKS CAN RUN IN PARALLEL.\n\nAnalyze your dependency graph and group tasks into PARALLEL EXECUTION WAVES:\n\nExample format:\n```\n## Parallel Execution Graph\n\nWave 1 (Start immediately):\n\u251C\u2500\u2500 Task 1: [description] (no dependencies)\n\u2514\u2500\u2500 Task 5: [description] (no dependencies)\n\nWave 2 (After Wave 1 completes):\n\u251C\u2500\u2500 Task 2: [description] (depends: Task 1)\n\u251C\u2500\u2500 Task 3: [description] (depends: Task 1)\n\u2514\u2500\u2500 Task 6: [description] (depends: Task 5)\n\nWave 3 (After Wave 2 completes):\n\u2514\u2500\u2500 Task 4: [description] (depends: Task 2, Task 3)\n\nCritical Path: Task 1 \u2192 Task 2 \u2192 Task 4\nEstimated Parallel Speedup: 40% faster than sequential\n```\n\nWHY THIS MATTERS:\n- MASSIVE time savings through parallelization\n- Executors can dispatch multiple agents simultaneously\n- Identifies bottlenecks in the execution plan\n\n\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 SECTION 3: CATEGORY + SKILLS RECOMMENDATIONS (MANDATORY)        \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nFOR EVERY TASK, YOU MUST RECOMMEND:\n1. Which CATEGORY to use for delegation\n2. Which SKILLS to load for the delegated agent\n";
-export declare const PLAN_AGENT_SYSTEM_PREPEND_STATIC_AFTER_SKILLS = "### REQUIRED OUTPUT FORMAT\n\nFor EACH task, include a recommendation block:\n\n```\n### Task N: [Task Title]\n\n**Delegation Recommendation:**\n- Category: `[category-name]` - [reason for choice]\n- Skills: [`skill-1`, `skill-2`] - [reason each skill is needed]\n\n**Skills Evaluation:**\n- INCLUDED `skill-name`: [reason]\n- OMITTED `other-skill`: [reason domain doesn't overlap]\n```\n\nWHY THIS MATTERS:\n- Category determines the MODEL used for execution\n- Skills inject SPECIALIZED KNOWLEDGE into the executor\n- Missing a relevant skill = suboptimal execution\n- Wrong category = wrong model = poor results\n\n\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 RESPONSE FORMAT SPECIFICATION (MANDATORY)                       \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYOUR PLAN OUTPUT MUST FOLLOW THIS EXACT STRUCTURE:\n\n```markdown\n# [Plan Title]\n\n## Context\n[User request summary, interview findings, research results]\n\n## Task Dependency Graph\n[Dependency table - see Section 1]\n\n## Parallel Execution Graph  \n[Wave structure - see Section 2]\n\n## Tasks\n\n### Task 1: [Title]\n**Description**: [What to do]\n**Delegation Recommendation**:\n- Category: `[category]` - [reason]\n- Skills: [`skill-1`] - [reason]\n**Skills Evaluation**: [\u2705 included / \u274C omitted with reasons]\n**Depends On**: [Task IDs or \"None\"]\n**Acceptance Criteria**: [Verifiable conditions]\n\n### Task 2: [Title]\n[Same structure...]\n\n## Commit Strategy\n[How to commit changes atomically]\n\n## Success Criteria\n[Final verification steps]\n```\n\n#####################################################################\n#                                                                   #\n#   FAILURE TO INCLUDE THESE SECTIONS = PLAN WILL BE REJECTED      #\n#   BY MOMUS REVIEW. DO NOT SKIP. DO NOT ABBREVIATE.               #\n#                                                                   #\n#####################################################################\n</CRITICAL_REQUIREMENT_DEPENDENCY_PARALLEL_EXECUTION_CATEGORY_SKILLS>\n\n<FINAL_OUTPUT_FOR_CALLER>\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 SECTION 4: ACTIONABLE TODO LIST FOR CALLER (MANDATORY)          \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYOU MUST END YOUR RESPONSE WITH THIS SECTION.\n\n```markdown\n## TODO List (ADD THESE)\n\n> CALLER: Add these TODOs using TodoWrite/TaskCreate and execute by wave.\n\n### Wave 1 (Start Immediately - No Dependencies)\n\n- [ ] **1. [Task Title]**\n  - What: [Clear implementation steps]\n  - Depends: None\n  - Blocks: [Tasks that depend on this]\n  - Category: `category-name`\n  - Skills: [`skill-1`, `skill-2`]\n  - QA: [How to verify completion - specific command or check]\n\n- [ ] **N. [Task Title]**\n  - What: [Steps]\n  - Depends: None\n  - Blocks: [...]\n  - Category: `category-name`\n  - Skills: [`skill-1`]\n  - QA: [Verification]\n\n### Wave 2 (After Wave 1 Completes)\n\n- [ ] **2. [Task Title]**\n  - What: [Steps]\n  - Depends: 1\n  - Blocks: [4]\n  - Category: `category-name`\n  - Skills: [`skill-1`]\n  - QA: [Verification]\n\n[Continue for all waves...]\n\n## Execution Instructions\n\n1. **Wave 1**: Fire these tasks IN PARALLEL (no dependencies)\n   ```\n   task(category=\"...\", load_skills=[...], run_in_background=false, prompt=\"Task 1: ...\")\n   task(category=\"...\", load_skills=[...], run_in_background=false, prompt=\"Task N: ...\")\n   ```\n\n2. **Wave 2**: After Wave 1 completes, fire next wave IN PARALLEL\n   ```\n   task(category=\"...\", load_skills=[...], run_in_background=false, prompt=\"Task 2: ...\")\n   ```\n\n3. Continue until all waves complete\n\n4. Final QA: Verify all tasks pass their QA criteria\n```\n\nWHY THIS FORMAT IS MANDATORY:\n- Caller can directly copy TODO items\n- Wave grouping enables parallel execution\n- Each task has clear task parameters\n- QA criteria ensure verifiable completion\n</FINAL_OUTPUT_FOR_CALLER>\n\n<DELIVERABLE_ENVELOPE>\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 MANDATORY: WRAP YOUR ENTIRE PLAN IN A <plan> ENVELOPE           \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYour FINAL message MUST wrap the ENTIRE plan \u2014 from the title line\nthrough the closing TODO list \u2014 inside a single pair of tags:\n\n<plan>\n# [Plan Title]\n... full plan: all mandatory sections AND the TODO list ...\n</plan>\n\nRULES:\n- Emit EXACTLY ONE <plan> ... </plan> block, and only in your FINAL message.\n- Open with <plan> on its own line BEFORE the title; close with </plan>\n  on its own line AFTER the TODO list. Everything the caller needs goes inside.\n- Do NOT emit <plan> while still gathering context. Progress messages,\n  clarifying questions, and \"exploring...\" turns MUST NOT contain it \u2014 the\n  envelope is how the caller tells your real deliverable apart from those turns.\n- The caller extracts the contents of this envelope verbatim. Anything emitted\n  outside the envelope may be discarded.\n</DELIVERABLE_ENVELOPE>\n\n";
+export declare const PLAN_AGENT_SYSTEM_PREPEND_STATIC_AFTER_SKILLS = "### REQUIRED OUTPUT FORMAT\n\nFor EACH task, include a recommendation block:\n\n```\n### Task N: [Task Title]\n\n**Delegation Recommendation:**\n- Category: `[category-name]` - [reason for choice]\n- Skills: [`skill-1`, `skill-2`] - [reason each skill is needed]\n\n**Skills Evaluation:**\n- INCLUDED `skill-name`: [reason]\n- OMITTED `other-skill`: [reason domain doesn't overlap]\n```\n\nWHY THIS MATTERS:\n- Category determines the MODEL used for execution\n- Skills inject SPECIALIZED KNOWLEDGE into the executor\n- Missing a relevant skill = suboptimal execution\n- Wrong category = wrong model = poor results\n\n\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 RESPONSE FORMAT SPECIFICATION (MANDATORY)                       \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYOUR PLAN OUTPUT MUST FOLLOW THIS EXACT STRUCTURE:\n\n```markdown\n# [Plan Title]\n\n## Context\n[User request summary, interview findings, research results]\n\n## Task Dependency Graph\n[Dependency table - see Section 1]\n\n## Parallel Execution Graph  \n[Wave structure - see Section 2]\n\n## Tasks\n\n### Task 1: [Title]\n**Description**: [What to do]\n**Delegation Recommendation**:\n- Category: `[category]` - [reason]\n- Skills: [`skill-1`] - [reason]\n**Skills Evaluation**: [\u2705 included / \u274C omitted with reasons]\n**Depends On**: [Task IDs or \"None\"]\n**Acceptance Criteria**: [Verifiable conditions]\n\n### Task 2: [Title]\n[Same structure...]\n\n## Commit Strategy\n[How to commit changes atomically]\n\n## Success Criteria\n[Final verification steps]\n```\n\n#####################################################################\n#                                                                   #\n#   FAILURE TO INCLUDE THESE SECTIONS = PLAN WILL BE REJECTED      #\n#   BY MOMUS REVIEW. DO NOT SKIP. DO NOT ABBREVIATE.               #\n#                                                                   #\n#####################################################################\n</CRITICAL_REQUIREMENT_DEPENDENCY_PARALLEL_EXECUTION_CATEGORY_SKILLS>\n\n<FINAL_OUTPUT_FOR_CALLER>\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 SECTION 4: ACTIONABLE TODO LIST FOR CALLER (MANDATORY)          \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYOU MUST END YOUR RESPONSE WITH THIS SECTION.\n\n```markdown\n## TODO List (ADD THESE)\n\n> CALLER: Add these TODOs using TodoWrite/TaskCreate and execute by wave.\n\n### Wave 1 (Start Immediately - No Dependencies)\n\n- [ ] **1. [Task Title]**\n  - What: [Clear implementation steps]\n  - Depends: None\n  - Blocks: [Tasks that depend on this]\n  - Category: `category-name`\n  - Skills: [`skill-1`, `skill-2`]\n  - QA: [How to verify completion - specific command or check]\n\n- [ ] **N. [Task Title]**\n  - What: [Steps]\n  - Depends: None\n  - Blocks: [...]\n  - Category: `category-name`\n  - Skills: [`skill-1`]\n  - QA: [Verification]\n\n### Wave 2 (After Wave 1 Completes)\n\n- [ ] **2. [Task Title]**\n  - What: [Steps]\n  - Depends: 1\n  - Blocks: [4]\n  - Category: `category-name`\n  - Skills: [`skill-1`]\n  - QA: [Verification]\n\n[Continue for all waves...]\n\n## Execution Instructions\n\n1. **Wave 1**: Fire these tasks IN PARALLEL (no dependencies)\n   ```\n   task(category=\"...\", load_skills=[...], run_in_background=true, prompt=\"Task 1: ...\")\n   task(category=\"...\", load_skills=[...], run_in_background=true, prompt=\"Task N: ...\")\n   ```\n\n2. **Wave 2**: After Wave 1 completes, fire next wave IN PARALLEL\n   ```\n   task(category=\"...\", load_skills=[...], run_in_background=true, prompt=\"Task 2: ...\")\n   ```\n\n3. Continue until all waves complete\n\n4. Final QA: Verify all tasks pass their QA criteria\n```\n\nWHY THIS FORMAT IS MANDATORY:\n- Caller can directly copy TODO items\n- Wave grouping enables parallel execution\n- Each task has clear task parameters\n- QA criteria ensure verifiable completion\n</FINAL_OUTPUT_FOR_CALLER>\n\n<DELIVERABLE_ENVELOPE>\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\u2588 MANDATORY: WRAP YOUR ENTIRE PLAN IN A <plan> ENVELOPE           \u2588\n\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\nYour FINAL message MUST wrap the ENTIRE plan \u2014 from the title line\nthrough the closing TODO list \u2014 inside a single pair of tags:\n\n<plan>\n# [Plan Title]\n... full plan: all mandatory sections AND the TODO list ...\n</plan>\n\nRULES:\n- Emit EXACTLY ONE <plan> ... </plan> block, and only in your FINAL message.\n- Open with <plan> on its own line BEFORE the title; close with </plan>\n  on its own line AFTER the TODO list. Everything the caller needs goes inside.\n- Do NOT emit <plan> while still gathering context. Progress messages,\n  clarifying questions, and \"exploring...\" turns MUST NOT contain it \u2014 the\n  envelope is how the caller tells your real deliverable apart from those turns.\n- The caller extracts the contents of this envelope verbatim. Anything emitted\n  outside the envelope may be discarded.\n</DELIVERABLE_ENVELOPE>\n\n";
 export declare function buildPlanAgentSkillsSection(categories?: AvailableCategory[], skills?: AvailableSkill[]): string;
 export declare function buildPlanAgentSystemPrepend(categories?: AvailableCategory[], skills?: AvailableSkill[]): string;
 /**

package/dist/tools/delegate-task/openai-categories.d.ts

@@ -1,5 +1,12 @@
 import type { BuiltinCategoryDefinition } from "./builtin-category-definition";
-export declare const DEEP_CATEGORY_PROMPT_APPEND = "<Category_Context>\nYou are working on GOAL-ORIENTED AUTONOMOUS tasks.\n\nYou are NOT an interactive assistant. You are an autonomous problem-solver.\n\nBEFORE making ANY changes:\n1. Silently explore the codebase extensively (5-15 minutes of reading is normal)\n2. Read related files, trace dependencies, understand the full context\n3. Build a complete mental model of the problem space\n4. Do not ask clarifying questions - the goal is already defined\n\nYou receive a GOAL. When the goal includes numbered steps or phases, treat them as one atomic task broken into sub-steps, not as separate independent tasks. Figure out HOW to achieve it yourself. Thorough research before any action.\n\nSub-steps of ONE goal = execute all steps as phases of one atomic task.\nGenuinely independent tasks = flag and refuse, require separate delegations.\n\nApproach: explore extensively, understand deeply, then act decisively. Prefer comprehensive solutions over quick patches. If the goal is unclear, make reasonable assumptions and proceed.\n\nMinimal status updates. Focus on results, not play-by-play. Report completion with summary of changes.\n</Category_Context>";
-export declare const DEEP_CATEGORY_PROMPT_APPEND_GPT_5_5 = "<Category_Context name=\"deep\">\nYou are operating in DEEP mode. This is the category reserved for goal-oriented autonomous work on hairy problems that reward thorough exploration and comprehensive solutions.\n\nThe orchestrator chose this category because the task benefits from depth over speed. You should feel empowered to spend the time needed: five to fifteen minutes of silent exploration before the first edit is normal and correct. Rushing to implementation on a deep task is a failure mode, not a feature.\n\n# How deep mode adjusts the base behavior\n\n**Exploration budget: generous.** Read the files you need, trace dependencies both directions, fire 2-5 explore/librarian sub-agents in parallel for broader questions. Build a complete mental model before the first `apply_patch`. Exploration here is an investment, not overhead.\n\n**Goal, not plan.** You receive a GOAL describing the desired outcome. You figure out HOW to achieve it. The orchestrator deliberately did not hand you a step-by-step plan; producing one and asking for approval is not what was asked. Execute.\n\n**Atomic task treatment.** When the goal contains numbered steps or phases, treat them as sub-steps of ONE task and execute them all in this turn. Splitting them across turns is wrong unless they reveal an architectural blocker that requires the user's input. If the \"steps\" turn out to be genuinely independent tasks that should have been separate delegations, flag that in your final message and refuse the ones beyond scope.\n\n**Root cause bias.** Prefer root-cause fixes over symptom fixes. A null check around `foo()` is a symptom fix; fixing whatever causes `foo()` to return unexpected values is the root fix. Trace at least two levels up before settling on an answer. In deep mode, you have permission (and the expectation) to do the deeper fix.\n\n**Ambition scaled to context.** For brand-new greenfield work, be ambitious. Choose strong defaults, avoid AI-slop aesthetics, produce something you would be proud to hand to another senior engineer. For changes in an existing codebase, be surgical and respect the existing patterns; depth does not mean invasiveness.\n\n**Completion bar: full delivery.** \"Simplified version\", \"proof of concept\", and \"you can extend this later\" are not acceptable deliveries for a deep task. The orchestrator routed here specifically for a complete solution. If you hit a genuine blocker (missing secret, design decision only the user can make, three materially different attempts all failed), document it and return; otherwise, finish the task.\n\n**Status cadence: sparse.** The user is not on the other side of this conversation; the orchestrator is, and they will synthesize your progress. Send commentary only at meaningful phase transitions (starting exploration, starting implementation, starting verification, hitting a genuine blocker). Do not narrate every tool call; silence during focused work is expected.\n</Category_Context>";
-export declare function resolveDeepCategoryPromptAppend(model: string | undefined): string;
+export declare const ULTRABRAIN_CATEGORY_PROMPT_APPEND_GPT_6_ASTRA = "<Category_Context name=\"ultrabrain\">\nThe orchestrator routed this task here because it is the one genuinely hard, logic-heavy problem in its plan, and it sent a goal rather than steps: choose the approach yourself, and let correctness outrank speed, brevity, and token cost.\n\nSuccess means:\n- every load-bearing claim cites evidence from this turn: a file and line read, a command run, a test executed;\n- every executable claim was executed: a proposed fix runs, an algorithm passes the boundary cases you enumerated, a verdict on a diff names the failing line;\n- the conclusion survived your own attempt to break it, and the answer names the strongest counter-case you looked for;\n- rejected alternatives carry the reason that decided against them, and open assumptions are stated so the orchestrator can overturn them;\n- one decision-complete recommendation, actionable without a follow-up question.\n\nWhatever that check leaves unsettled goes in the answer as an open question with what would settle it. When the goal bundles independent problems, solve the one the others depend on and return the rest as separately delegable items.\n</Category_Context>";
+export declare const DEEP_HIGH_CATEGORY_PROMPT_APPEND_GPT = "<Category_Context name=\"deep-high\">\nThe orchestrator routed this task here because a decision in it cannot be settled from evidence alone: a trade-off, a contract other code depends on, a mechanism with no pattern to copy, or correctness that has to be argued. One goal, one deliverable, and the time to earn it. The exploration budget is generous: read every file involved, trace callers and dependencies in both directions, and fan out explore and librarian subagents in parallel for the questions a single read wave cannot answer, until you can explain the full mechanism you are about to change; an edit made before that point is the failure this category exists to prevent.\n\nThe goal is the authorization. Choose how to reach it yourself, and when it lists numbered steps or phases, deliver all of them in this turn as one task; a proposal, a plan awaiting approval, a simplified version, or a proof of concept is unfinished work. When the steps turn out to be independent problems sharing no reasoning, do the one the goal centers on and return the others as separately delegable items with what you learned. A question ends your turn and hands the task back unfinished, so decide from context, record each assumption in the final message, and stop early only for a blocker you cannot route around: a missing secret, a decision only the user can make, or three materially different attempts that all failed.\n\nFix the cause: trace at least two levels above the symptom before settling, and prefer the change that makes the failure impossible over the guard that hides it. Depth means understanding the mechanism, so the diff stays as small as the fix allows; on greenfield work choose strong defaults and finish something you would hand to a senior engineer. Close with the delivered change, the evidence that it works, the decision you settled with the alternative you rejected, and the assumptions you made.\n</Category_Context>";
+export declare const DEEP_LOW_CATEGORY_PROMPT_APPEND_GPT = "<Category_Context name=\"deep-low\">\nThe orchestrator routed this task here as the default deep lane: one goal, one deliverable, decisions that the code, tests, and history can settle. Read until you can explain the mechanism you are about to change, including its callers, then act; fan out explore and librarian subagents when one read wave cannot answer a question.\n\nSuccess means:\n- every step or phase the goal lists is delivered in this turn as one task; a plan awaiting approval, a simplified version, or a proof of concept is unfinished work;\n- the change removes the cause you traced, at least two levels above the symptom, with the diff as small as that fix allows;\n- the final message reports the change, the evidence that it works, and each assumption you decided from context, because a question ends your turn and returns the task unfinished.\n\nEscalation is a complete result. When the correct choice turns out to depend on a trade-off the brief does not settle, on a contract other packages rely on, or on an argument about invariants you cannot verify by running something, stop before editing and return `ESCALATE: deep-high` as the first line, followed by what you read, the decision you could not settle, and the options you saw. A confident guess through that decision costs more than the escalation.\n\nStop the moment the success criteria hold; work past them is a defect. Stop early only for a blocker you cannot route around: a missing secret, a decision only the user can make, or three materially different attempts that all failed.\n</Category_Context>";
+export declare const UNSPECIFIED_HIGH_CATEGORY_PROMPT_APPEND_GPT_6_ASTRA = "<Category_Context name=\"unspecified-high\">\nThe orchestrator routed this task here because it spans systems or modules and fits no specialist category, so breadth of consideration is what this category buys. Before committing to an approach, survey the whole surface the change touches: every caller and consumer of what you will modify, sibling modules that implement the same pattern, the tests, docs, schemas, config, scripts, and CI that encode the current behavior, and the history of the area (git log and blame) for the reasons it is shaped this way. Fan out explore and librarian subagents in parallel when that surface is wider than one read wave covers.\n\nWeigh at least two ways to do it against what you found, choose one, and say in the final message why it won. Then deliver it across every surface you identified, so behavior stays consistent everywhere the change is observable and no caller, test, doc, schema, or config still describes the old state. A question ends your turn and hands the task back unfinished, so decide from context, record each assumption in the final message, and finish.\n</Category_Context>";
+export declare const DEEP_LOW_CATEGORY_PROMPT_APPEND = "<Category_Context name=\"deep-low\">\nYou are working on a GOAL-ORIENTED AUTONOMOUS task: one goal, one deliverable, decisions the codebase can settle.\n\nBefore any change, read the files involved and trace their dependencies until you can explain the mechanism you are about to modify. The goal is already defined: do not ask clarifying questions; make reasonable assumptions, record them in the final message, and proceed.\n\nWhen the goal lists numbered steps or phases, execute all of them in this turn as one atomic task. Genuinely independent tasks bundled into one goal: flag them and do only the one the goal centers on.\n\nEscalation is a complete result. When the correct choice depends on a trade-off the brief does not settle, a contract other packages rely on, or an argument about invariants you cannot verify by running something, stop before editing and return `ESCALATE: deep-high` as the first line, followed by what you read, the decision you could not settle, and the options you saw.\n\nPrefer the fix that removes the cause over the patch that hides the symptom. Report completion with the changes made and the evidence they work.\n</Category_Context>";
+export declare const DEEP_HIGH_CATEGORY_PROMPT_APPEND = "<Category_Context name=\"deep-high\">\nYou are working on a GOAL-ORIENTED AUTONOMOUS task that was escalated here because a decision in it cannot be settled from evidence alone: a trade-off, a contract other code depends on, a mechanism with no pattern to copy, or correctness that has to be argued.\n\nBefore any change, read the files involved and trace their dependencies until you can explain the mechanism you are about to modify. The goal is already defined: do not ask clarifying questions; make reasonable assumptions, record them in the final message, and proceed.\n\nWhen the goal lists numbered steps or phases, execute all of them in this turn as one atomic task. Genuinely independent tasks bundled into one goal: flag them and do only the one the goal centers on.\n\nPrefer the fix that removes the cause over the patch that hides the symptom. Report completion with the changes made, the evidence they work, and the decision you settled with the alternative you rejected.\n</Category_Context>";
+export declare function resolveUltrabrainCategoryPromptAppend(model: string | undefined): string;
+export declare function resolveDeepLowCategoryPromptAppend(model: string | undefined): string;
+export declare function resolveDeepHighCategoryPromptAppend(model: string | undefined): string;
+export declare function resolveUnspecifiedHighCategoryPromptAppend(model: string | undefined): string;
 export declare const OPENAI_CATEGORIES: BuiltinCategoryDefinition[];

package/dist/tools/delegate-task/sync-session-cleanup.d.ts

@@ -0,0 +1,3 @@
+import type { OpencodeClient } from "./types";
+export declare function cancelSyncSessionDeletion(sessionID: string): void;
+export declare function scheduleSyncSessionDeletion(client: OpencodeClient, sessionID: string, delayMs?: number): void;

package/dist/tools/delegate-task/sync-session-poller.d.ts

@@ -8,6 +8,7 @@ export declare function pollSyncSession(ctx: ToolContextWithMetadata, client: Op
     } | null | undefined;
     taskId: string | undefined;
     anchorMessageCount?: number;
+    anchorMessageID?: string;
     maxAssistantTurns?: number;
     hasActiveChildBackgroundTasks?: (sessionID: string) => boolean;
     hasPendingParentWake?: (sessionID: string) => boolean;

package/dist/tools/delegate-task/tool-description.d.ts

@@ -6,4 +6,6 @@ export interface DelegateTaskPresentation {
     categoryExamples: string;
     description: string;
 }
-export declare function createDelegateTaskPresentation(options: DelegateTaskToolOptions): DelegateTaskPresentation;
+type DelegateTaskPresentationOptions = Pick<DelegateTaskToolOptions, "availableCategories" | "availableSkills" | "userCategories">;
+export declare function createDelegateTaskPresentation(options: DelegateTaskPresentationOptions): DelegateTaskPresentation;
+export {};

package/dist/tools/delegate-task/types.d.ts

@@ -1,4 +1,5 @@
 import type { BackgroundManager } from "../../features/background-agent";
+import type { OhMyOpenCodeConfig } from "../../config";
 import type { CategoriesConfig, GitMasterConfig, BrowserAutomationProvider, AgentOverrides, SisyphusAgentConfig } from "../../config/schema";
 import type { ModelFallbackControllerAccessor } from "../../hooks/model-fallback";
 import type { LoadedSkill } from "../../features/opencode-skill-loader/types";
@@ -45,6 +46,7 @@ export interface OmoAgentClient {
     };
     readonly session: {
         readonly abort: (input: SessionPathInput) => Promise<unknown>;
+        readonly delete?: (input: SessionPathInput) => Promise<unknown>;
         readonly create: (input: {
             readonly body: Record<string, unknown>;
             readonly query?: {
@@ -119,6 +121,8 @@ export interface DelegateTaskToolOptions {
     availableCategories?: AvailableCategory[];
     availableSkills?: AvailableSkill[];
     agentOverrides?: AgentOverrides;
+    /** Reload model-bearing config at task invocation time so edits are honored without rebuilding tools. */
+    loadCurrentModelConfig?: () => Pick<OhMyOpenCodeConfig, "agents" | "categories">;
     sisyphusAgentConfig?: SisyphusAgentConfig;
     modelFallbackControllerAccessor?: ModelFallbackControllerAccessor;
     onSyncSessionCreated?: (event: SyncSessionCreatedEvent) => Promise<void>;

package/dist/tools/session-manager/directory-filter.d.ts

@@ -0,0 +1,2 @@
+export declare function normalizeSessionDirectory(directory: string): string;
+export declare function sessionDirectoriesMatch(stored: string, filter: string): boolean;

package/dist/tui.d.ts

@@ -1,4 +1,16 @@
 import type { TuiPluginModule } from "@opencode-ai/plugin/tui";
+type SidebarSlotRegistration<Node> = {
+    readonly order: number;
+    readonly slots: {
+        readonly sidebar_content: () => Node | (() => Node);
+    };
+};
+type RegisterSidebarContentSlotInput<Node> = {
+    readonly registerSlot: (registration: SidebarSlotRegistration<Node>) => void;
+    readonly requestRender: () => void;
+    readonly renderSidebar: () => Node;
+};
+export declare function registerSidebarContentSlot<Node>({ registerSlot, requestRender, renderSidebar, }: RegisterSidebarContentSlotInput<Node>): void;
 export declare function handleTuiPollError(error: unknown, reportPollError?: (error: Error) => void): void;
 declare const module: TuiPluginModule;
 export default module;