jtaroreh · Oct 29, 2026 · Oct 29, 2026
diff --git a/README.md b/README.md
@@ -34,7 +34,7 @@ this repository is **agystack**, the agentic engineering framework for Google An
 
 ### prerequisites
 
-- `bun` (v1.0+) is mandatory for PR babysitting (`watch-pr`) and multi-agent orchestration (`orch`). Node.js is not supported.
+- `bun` (v1.0+) is used for PR babysitting (`watch-pr`). Orchestration is native agentic markdown via Brain Artifacts. Node.js is not supported.
 - `gh` (GitHub CLI) is required for PR automation and preflight checks.
 - `gt` (Graphite CLI) is recommended for stacked PRs.
 
@@ -207,6 +207,7 @@ Or trigger `/swarm` right in chat:
 | [`/unslop`](./skills/unslop/SKILL.md) | you're cleaning up writing. removes AI tells. |
 | [`/bro`](./skills/bro/SKILL.md) | you want the last message restated in plain human language, no jargon. |
 | [`/technical-writing`](./skills/technical-writing/SKILL.md) | layered doc standard (Diátaxis + Google developer style + STE + Global English) for docs, RFCs, readmes, PR descriptions, commit messages. |
+| [`/statechart`](./skills/statechart/SKILL.md) | you want to model system states, transitions, invariants, and failure modes as native Mermaid diagrams. |
 
 </details>
 

diff --git a/rules/agystack-models.md → agystack-models.md b/rules/agystack-models.md → agystack-models.md
diff --git a/assets/logo.svg b/assets/logo.svg
diff --git a/docs/guide/01-setup.md b/docs/guide/01-setup.md
@@ -29,7 +29,7 @@ Run:
 /setup-agystack
 ```
 
-[`/setup-agystack`](../../skills/setup-agystack/SKILL.md) detects the models you have access to, shows you each role (code delegates, judgment, the review panels), and asks what you want. Answer the questions. It writes `~/.gemini/config/plugins/agystack/rules/agystack-models.md`, a small rule every agystack skill reads.
+[`/setup-agystack`](../../skills/setup-agystack/SKILL.md) detects the models you have access to, shows you each role (code delegates, judgment, the review panels), and asks what you want. Answer the questions. It writes `~/.gemini/config/plugins/agystack/agystack-models.md`, a configuration file every agystack skill reads.
 
 You only override what you care about. A role with no line in the rule keeps the skill's default. To restore a default later, delete that role's line, or just run `/setup-agystack` again.
 

diff --git a/docs/plans/2026-10-cloud-run-model-heterogeneity.md b/docs/plans/2026-10-cloud-run-model-heterogeneity.md
diff --git a/plugin.json b/plugin.json
@@ -1,6 +1,7 @@
 {
   "$schema": "https://antigravity.google/schemas/v1/plugin.json",
   "name": "agystack",
+  "displayName": "agystack",
   "version": "0.15.0",
   "description": "agystack for Antigravity IDE. Rigorous agent workflows: poteto-mode, how, why, architect, arena, swarm, interrogate, deslop. Port of Lauren Tan's pstack for Antigravity.",
   "author": {
@@ -9,6 +10,7 @@
   "homepage": "https://github.com/jtaroreh/agystack",
   "repository": "https://github.com/jtaroreh/agystack",
   "license": "MIT",
+  "logo": "assets/logo.svg",
   "keywords": [
     "agystack",
     "pstack",
@@ -19,5 +21,10 @@
     "subagents",
     "unslop",
     "deslop"
+  ],
+  "suggestedPrompts": [
+    "Plan and implement this feature with /poteto-mode rigor",
+    "Interrogate this implementation with multi-model adversarial reviews",
+    "Run /arena to compare multiple architectural approaches"
   ]
 }
diff --git a/rules/AGENTS.md b/rules/AGENTS.md
@@ -44,4 +44,4 @@ Always use the most direct, deterministic tool surface for each task:
 2. **CLI / Runtime Tools (`run_command`)**: Test suites, API endpoints, build/server logs, process lifecycles, and exit codes.
 3. **Browser Tools (`browser_subagent` / Chrome DevTools MCP)**: Strictly for live UI interaction, DOM layout, CSS styles, user events, and rendered visual screenshots.
 
-Per-role models live in `agystack-models.md` next to this file. `/setup-agystack` rewrites that file.
+Per-role models live in `agystack-models.md` at the plugin root. `/setup-agystack` rewrites that file.
diff --git a/skills/arena/SKILL.md b/skills/arena/SKILL.md
@@ -21,7 +21,7 @@ Open a todolist tracking the operational phases:
 
 1. **Artifact.** Define the target output each candidate must produce.
 2. **Rubric.** Formulate 3-6 concrete, gradeable criteria (e.g. `Adds a --dry-run flag that skips writes`).
-3. **Runners.** Use `arena runners` from `~/.gemini/config/plugins/agystack/rules/agystack-models.md` if configured, or default across available tiers: `pro`, `flash`, `inherit`.
+3. **Runners.** Use `arena runners` from `~/.gemini/config/plugins/agystack/agystack-models.md` if configured, or default across available tiers: `pro`, `flash`, `inherit`.
 4. **Workspaces.** Assign isolated workspaces via `Workspace: "branch"` or scratch paths (`<appDataDir>/brain/<conversation-id>/scratch/arena-<slug>/candidate-<n>/`) per **principle-separate-before-serializing-shared-state**.
 
 ## Phase B: Fan out
@@ -41,7 +41,7 @@ Provide `{PARENT_CONVERSATION_ID}` in each prompt. Candidates MUST invoke `send_
 
 ## Phase C: Cross-judge
 
-Spawn one judge subagent via `invoke_subagent` (`TypeName: "research"`, `Role: "Arena Cross-Judge"`) on a distinct tier from `pro`, `flash`, `inherit`. The judge scores candidates against rubric criteria and recommends a base with rationale.
+Spawn one judge subagent via `invoke_subagent` (`TypeName: "research"`, `Role: "Arena Cross-Judge"`). The cross-judge must be selected from a different model family than the candidates. For Antigravity, assign the judge role to `Model: "pro"` (Gemini Pro) to ensure deep comparative judgment over Flash candidate drafts. The judge scores candidates against rubric criteria and recommends a base with rationale.
 
 ## Phase D: Pick a base
 

diff --git a/skills/interrogate/SKILL.md b/skills/interrogate/SKILL.md
@@ -25,7 +25,7 @@ State what the code attempts to accomplish based on commit messages, PR descript
 
 Launch all reviewers in a single turn using `invoke_subagent`. Simulating reviewer personas in-context is strictly prohibited.
 
-Use the `interrogate reviewers` list from `~/.gemini/config/plugins/agystack/rules/agystack-models.md` when present, or fall back to defaults:
+Use the `interrogate reviewers` list from `~/.gemini/config/plugins/agystack/agystack-models.md` when present, or fall back to defaults:
 
 | Subagent | Default model tier |
 |---|---|
@@ -42,7 +42,11 @@ For each reviewer:
 Fill `references/reviewer-prompt.md` with:
 1. Stated intent
 2. Diff or file contents
-3. Rubric from `references/rubric.md`
+3. Rubric from `references/rubric.md`, explicitly injecting concrete audit categories:
+   - **Idempotency & Crash Recovery:** Re-execution safety and partial failure state.
+   - **Concurrency & Races:** Shared state serialization and lock invariants.
+   - **Boundary Discipline:** System boundary validation vs internal type trust.
+   - **Contract vs Symptom:** Root cause verification vs shallow guard clauses.
 4. Code-quality lens from `references/code-quality-review.md`
 5. System `{PARENT_CONVERSATION_ID}` for return routing
 
@@ -55,21 +59,24 @@ Fill `references/reviewer-prompt.md` with:
 4. **Full Arrival:** When all reviewers report, cancel the timer via `manage_task(Action: "kill", TaskId: <timer_task_id>)` and proceed to Step 4.
 5. **Watchdog Timeout Fallback:** If the watchdog fires, inspect `manage_subagents(Action: "list")`, terminate unresponsive workers (`manage_subagents(Action: "kill", ConversationIds: [<id>])`), and synthesize surviving findings.
 
-## Step 4. Synthesize
+## Step 4. Empirical Verification Gate & Synthesis
 
-- Parse reviewer findings
-- Identify multi-model consensus (highest confidence)
-- Isolate lone-model findings (lower confidence)
-- Deduplicate overlapping issues
-- Highlight direct disagreements between models
+- Parse reviewer findings.
+- **Empirical Gate:** Every actionable finding MUST include an executable repro command, curl call, test snippet, or shell one-liner proving the defect or reproducing the failure against the code. Speculative findings lacking executable reproduction are dropped.
+- **Lead Verifier Rerun:** The coordinator reruns each repro command using `run_command`. If the command does not reproduce the failure against the codebase, the finding is dropped.
+- Identify multi-model consensus (highest confidence).
+- Isolate lone-model findings (lower confidence).
+- Deduplicate overlapping issues.
+- Highlight direct disagreements between models.
 
 ## Step 5. Lead Judgment
 
-Classify findings using the framework in `references/lead-judgment.md`:
-- **Act on.** Critical defects in correctness, security, or maintainability that block landing.
+Classify surviving verified findings using the framework in `references/lead-judgment.md`:
+- Filter out "Nitpick Gravity" and subjective preferences ("I would have done it differently").
+- **Act on.** Critical defects in correctness, security, or maintainability with verified repros that block landing.
 - **Consider.** Valid tradeoffs worth operator attention.
 - **Noted.** Context-dependent, premature optimization, or low impact.
-- **Dismissed.** False positives or missing context, with brief rationale.
+- **Dismissed.** Unreproduced findings, false positives, or missing context, with brief rationale.
 
 ## Output Format
 
@@ -78,7 +85,7 @@ Write the complete categorized review, rationales, and agreement map to `<appDat
 Structure:
 - **Intent.** Stated goal of the change.
 - **Reviewers.** One bullet per reviewer (model, finding count).
-- **Act On.** Findings requiring resolution.
+- **Act On.** Findings requiring resolution, with verified repro commands.
 - **Consider.** Tradeoffs to evaluate.
 - **Noted.** Low-priority items.
 - **Dismissed.** Rejected findings with justification.

diff --git a/skills/interrogate/references/lead-judgment.md b/skills/interrogate/references/lead-judgment.md
@@ -15,6 +15,12 @@ You have the full conversation context. Use it.
 
 ## Filtering Principles
 
+### Empirical Verification Rerun
+
+Before acting on any finding, the coordinator reruns the reviewer's provided repro command using `run_command`.
+- If the command does not reproduce the failure against the code, or if the reviewer failed to provide an executable reproduction, drop the finding.
+- A finding must stand on empirical evidence, not theoretical speculation or rhetorical confidence.
+
 ### Nitpick Gravity
 
 Reviewers, especially adversarial ones, tend to fill their review. If they don't find critical issues, they'll inflate nits to fill the space. If a reviewer's findings are all nits and style preferences, the code is probably fine. Say so.

diff --git a/skills/interrogate/references/reviewer-prompt.md b/skills/interrogate/references/reviewer-prompt.md
@@ -38,7 +38,8 @@ For each finding, provide:
    - `nit`: Style, naming, minor improvement. Only include nits if they're genuinely useful, not to pad your review.
 2. **Finding**: What the problem is, in concrete terms. Reference specific lines/functions.
 3. **Evidence**: Why you believe this is a problem. Show your reasoning. Don't just assert.
-4. **Suggestion** (optional): What you'd do instead, if you have a concrete alternative. Skip this if you don't have a clear fix.
+4. **Repro / Verification**: An executable command, curl call, test snippet, or shell one-liner proving the defect or reproducing the failure against the code. Findings without a reproducible check or executable proof are rejected.
+5. **Suggestion** (optional): What you'd do instead, if you have a concrete alternative. Skip this if you don't have a clear fix.
 
 ## What Makes a Good Finding
 
@@ -72,6 +73,7 @@ Format your findings as:
 **Location**: file:line or function name
 **Finding**: What's wrong
 **Evidence**: Why this matters
+**Repro / Verification**: An executable command, curl call, test snippet, or shell one-liner proving the defect or reproducing the failure against the code.
 **Suggestion**: (optional) What to do instead
 
 ### 2. [Severity] Short title

diff --git a/skills/ketchup/SKILL.md b/skills/ketchup/SKILL.md
@@ -27,7 +27,7 @@ The user can respond with a single token or pick such as Option A.
 Do not ask open-ended questions when a structured pick settles the fork.
 When an action is blocked on external input, state the exact input required and the default fallback.
 
-## Native Antigravity Artifacts
+## Native Antigravity Artifacts & Visual Progress
 
 Deliver comprehensive context through native Antigravity Artifacts in `<appDataDir>/brain/<conversation-id>/`.
 Point to artifacts using clickable file links.
@@ -38,6 +38,10 @@ Key companion artifacts include:
 - `decision_trail.md` for sequential forks, rationales, and results.
 - `implementation_plan.md` for roadmap progress and checked boxes.
 
+### Visual Progress Carousels and Generative UI
+Use native markdown carousels (````carousel ... ````) with `<!-- slide -->` comments for sequential visual walkthroughs and before/after progressions.
+Embed `generative_ui` HTML/SVG timeline widgets for visual progress rendering directly in artifacts without external video compilation dependencies like ffmpeg MP4s.
+
 ## Visual Media Embedding
 
 Link or embed visual media directly to provide immediate visual receipts.

diff --git a/skills/poteto-mode/SKILL.md b/skills/poteto-mode/SKILL.md
@@ -1,5 +1,5 @@
 ---
-name: Poteto Mode
+name: poteto-mode
 description: poteto's agent style for concise, detailed responses, deliberate subagents, unslopped prose, simple code, and verified work. Use for poteto, /poteto-mode, or requests to work in this style.
 mode: true
 icon: crown