Enhance agent documentation and workflows

- Updated planning-agent.md to include best uses, non-goals, review mode, and output format for plan reviews.
- Revised qa.md to add best uses, escalation criteria, and a new BDD mode for requirements-driven testing.
- Improved reviewer.md with best uses, escalation criteria, and a section for positive signals in code reviews.
- Expanded security.md to include best uses, escalation criteria, and a severity rubric for vulnerabilities.
- Removed test_generator.md as its functionality is integrated into qa.md.
- Consolidated cheatsheet.md to reflect updated agent roles and routing guidelines, emphasizing cost-awareness and model usage.
- Added setup.sh script for streamlined development environment setup, including symlinks and package installations.
- Updated cost-awareness.md to clarify model usage and decision-making guidelines for cost-effective development.
- Introduced personal-profile.md to capture communication, stack preferences, and evaluation criteria for tools and research.

Author: thisguymartin

.claude/rules/personal-profile.md

@@ -0,0 +1 @@
+../../skills/rules/personal-profile.md

.cursor/rules/personal-profile.mdc

@@ -0,0 +1,44 @@
+---
+description: Captures Martin's communication, stack, research, architecture, privacy, and cost preferences. Use when giving recommendations, writing code, evaluating tools, or explaining concepts.
+globs: "**/*"
+---
+
+# Personal Profile
+
+## Communication Style
+- Direct, casual, first-person tone
+- Slack-style brevity
+- No corporate polish
+- Minimal em dashes
+- Use `->` notation for flows
+- Use real-world analogies when teaching new concepts
+
+## Developer Context
+- Default to Go and TypeScript unless told otherwise
+- Prefer terminal-first workflows: Zellij, git worktrees, LazyGit, CLI tools
+- Prefer CLI solutions over GUI recommendations
+- Check current docs first with Context7, MCP, or verified web sources when available
+- Favor production-ready code with error handling, context propagation, and logging
+- Lead architecture discussions with domain modeling before folders or frameworks
+- Do not assume deployment target; ask if infra is unspecified
+- Default to the simplest, cheapest viable option
+
+## Research And Evaluation
+- Include maintenance health, commit frequency, issue load, bus factor, license, adoption signals, and stack fit
+- Cite sources for recommendations
+- Say explicitly when no source is available
+- Include confidence levels for claims and recommendations
+- Flag deprecations, breaking changes, and migration concerns proactively
+
+## Business And Startup
+- Structure idea work as: problem, user, solution, MVP, business model, technical architecture
+- Include competitors and differentiation when relevant
+- Use fresh market data with dates when available
+
+## General Principles
+- Never suggest sending real customer data to third-party AI services
+- Use synthetic data for testing
+- Prefer maintainable solutions for a solo dev or small team
+- Stay cost-conscious by default
+- Calibrate teaching depth to current domain familiarity
+- Label extrapolation and speculation clearly

.opencode/plugins/completion-sound.js

@@ -0,0 +1,39 @@
+const DEFAULT_COOLDOWN_MS = 1500
+
+export const CompletionSoundPlugin = async ({ client, $ }) => {
+  const defaultSound = new URL("../sounds/navi-listen.wav", import.meta.url).pathname
+  const soundFile = process.env.OPENCODE_SOUND_FILE || defaultSound
+  const cooldownMs = Number(process.env.OPENCODE_SOUND_COOLDOWN_MS || DEFAULT_COOLDOWN_MS)
+
+  let lastPlayedAt = 0
+
+  async function playCompletionSound() {
+    const now = Date.now()
+    if (now - lastPlayedAt < cooldownMs) return
+    lastPlayedAt = now
+
+    try {
+      await $`afplay ${soundFile}`
+    } catch (error) {
+      await client.app.log({
+        body: {
+          service: "completion-sound-plugin",
+          level: "warn",
+          message: "Failed to play OpenCode completion sound",
+          extra: {
+            soundFile,
+            error: String(error),
+          },
+        },
+      })
+    }
+  }
+
+  return {
+    event: async ({ event }) => {
+      if (event.type === "session.idle") {
+        await playCompletionSound()
+      }
+    },
+  }
+}

.opencode/sounds/navi-listen.wav

@@ -0,0 +1 @@
+../../sounds/navi-listen.wav

CLAUDE.md

@@ -10,8 +10,8 @@ A portable, reproducible terminal development environment (devkit). Not a softwa
 
 - `brewfile` -- Homebrew package manifest (install with `brew bundle --file=brewfile`)
 - `zellij/layouts/` -- KDL layout files for Zellij workspaces (api, database, debug, golang, migrations, monitor, node, pipeline, testrunner)
-- `opencode/aig_agents/` -- AI agent persona definitions for OpenCode (advisor, architect, engineer, QA, security, etc.)
-- `skills/rules/` -- Always-on coding standards (code quality, engineering principles, security, workflow)
+- `opencode/aig_agents/` -- OpenCode agent persona definitions (planning, engineer, coder, frontend, QA, reviewer, security, docs, Linear, pickle agents)
+- `skills/rules/` -- Always-on coding standards and personal preferences (code quality, engineering principles, security, workflow, personal profile)
 - `skills/commands/` -- On-demand slash commands (refactor-challenge, trace-debug, explain-code, etc.)
 - `skills/plane/` -- Plane.so project management commands
 - `.claude/rules/` -- Symlinked from `skills/rules/` (single source of truth)
@@ -57,7 +57,7 @@ Configs are symlinked from this repo to `~/.config/` and `~/`:
 - `.config/shell/enhancements.zsh` -> sourced from `.zshrc`
 - `.claude/settings.json` -> `~/.claude/settings.json`
 - `opencode/aig_agents/` -> `~/.config/opencode/agents/` (via `scripts/opencode-setup.sh`)
-- `opencode.json` -> `~/.config/opencode/config.json` (via `scripts/opencode-setup.sh`)
+- `opencode.json` -> `~/.config/opencode/opencode.json` (via `scripts/opencode-setup.sh`)
 
 ### OpenCode + Copilot Pro Setup
 
@@ -91,21 +91,21 @@ AI agent and assistant configs are included but optional -- the core workflow (t
 - **Claude Code** -- Kept for work use; devkit includes Claude rules and settings for both tools
 
 ### OpenCode Agents
-14 specialized agents in `opencode/aig_agents/`, invoked with `@agent-name` syntax inside OpenCode:
-- `@agent-advisor` -- routes tasks to appropriate agents and tools
-- `@planning-agent` / `@plan-reviewer` -- architecture and plan review (Gemini 3.1 Pro)
-- `@engineer` -- default builder (GPT 5.4, requires approval)
-- `@coder` -- autonomous test-fix loops (GPT 5.3 Codex)
-- `@frontend` -- UI/vision-to-code (Kimi K2.5)
-- `@lead_dev` -- quick tasks, auto-commits (MiniMax Free)
+11 specialized agents in `opencode/aig_agents/`, invoked with `@agent-name` syntax inside OpenCode:
+- `@planning-agent` -- architecture, task breakdown, and plan review (Gemini 3.1 Pro)
+- `build` mode in `opencode.json` -- default free implementation lane (MiniMax M2.5 Free)
+- `@engineer` -- premium implementation/orchestration for harder work (Gemini 3.1 Pro)
+- `@coder` -- autonomous test-fix loops when the spec and tests are clear (GPT 5.3 Codex)
+- `@frontend` -- cost-sensitive UI and component work (Kimi K2.5)
 - `@reviewer` / `@security` -- code review and security audit (Gemini 3.1 Pro, cross-model)
-- `@qa` / `@test_generator` -- testing (GPT 5.3 Codex)
+- `@qa` -- test generation, execution, and BDD flows (GPT 5.4 Mini)
 - `@docs_generator` -- documentation (Gemini 3 Flash)
-- `@linear` -- Linear project management
-- `@commiter` -- git automation (MiniMax Free)
+- `@linear` -- Linear project management (GPT 5.4 Mini)
+- `@pickle-think` -- free triage and rough planning (Big Pickle)
+- `@pickle-implement` -- free low-risk code changes (Big Pickle)
 
 ### Rules & Commands
-Always-on rules in `skills/rules/` are symlinked into `.claude/rules/` -- they apply to every conversation automatically.
+Always-on rules in `skills/rules/` are symlinked into `.claude/rules/` -- they apply to every conversation automatically. Personal preferences live in `skills/rules/personal-profile.md`.
 
 On-demand commands in `skills/commands/` are invoked as slash commands:
 - `/refactor-challenge` and `/trace-debug` -- Socratic tutors (guide, don't fix)
@@ -126,6 +126,6 @@ Frontend design skills (`/audit`, `/polish`, `/critique`, `/animate`, etc.) are
 - Shell scripts should use `set -e` and follow existing patterns in `scripts/`
 - `skills/rules/` is the single source of truth for always-on rules -- `.claude/rules/` files are symlinks. Edit files in `skills/rules/`, not `.claude/rules/`
 - `opencode/aig_agents/` is the single source of truth for OpenCode agents -- `~/.config/opencode/agents/` is a symlink
-- Agent model IDs use the `opencode/` prefix for Zen models (e.g., `opencode/gpt-5.4`)
+- Agent model IDs use the `opencode/` prefix for Zen models (e.g., `opencode/minimax-m2.5-free`, `opencode/gemini-3.1-pro`)
 - `skills/commands/` and `skills/plane/` are slash commands symlinked into `~/.claude/skills/`
 - `.cursor/rules/` is maintained separately in `.mdc` format -- update when modifying standards in `skills/`

README.md

@@ -82,26 +82,23 @@ The repo includes configs for AI coding tools — these are entirely optional an
 | PR review | Copilot | Included | Assign Copilot as reviewer on PRs. |
 | Heavy agentic work | OpenCode + Zen | ~$90/mo | Architecture, multi-file refactors, autonomous coding. |
 
-### OpenCode Agents (14 agents)
+### OpenCode Agents (11 agents)
 
 | Agent | Specialty | Model | Cost | Invoke With |
 | :--- | :--- | :--- | :--- | :--- |
-| **Advisor** | Task routing & tool selection | Gemini 3 Flash | Budget | `@agent-advisor` |
-| **Architect** | System design & planning | Gemini 3.1 Pro | $2/$12 | `@planning-agent` |
-| **Plan Reviewer** | Architecture review | Gemini 3.1 Pro | $2/$12 | `@plan-reviewer` |
-| **Engineer** | Default builder (approval) | GPT 5.4 | $2.50/$15 | `@engineer` |
+| **Architect** | System design, task breakdown, plan review | Gemini 3.1 Pro | $2/$12 | `@planning-agent` |
+| **PM** | Linear integration | GPT 5.4 Mini | $0.75/$4.50 | `@linear` |
+| **Engineer** | Premium builder (approval) | Gemini 3.1 Pro | $2/$12 | `@engineer` |
 | **Coder** | Autonomous test-fix loops | GPT 5.3 Codex | $1.75/$14 | `@coder` |
-| **Frontend** | UI/vision-to-code | Kimi K2.5 | $0.60/$3 | `@frontend` |
-| **Lead Dev** | Quick tasks (auto-commit) | MiniMax Free | FREE | `@lead_dev` |
+| **Frontend** | UI/component development | Kimi K2.5 | $0.60/$3 | `@frontend` |
 | **Reviewer** | Code quality review | Gemini 3.1 Pro | $2/$12 | `@reviewer` |
 | **Security** | Vulnerability scanning | Gemini 3.1 Pro | $2/$12 | `@security` |
-| **QA** | Test generation & execution | GPT 5.3 Codex | $1.75/$14 | `@qa` |
-| **Test Gen** | BDD/requirements-driven tests | GPT 5.3 Codex | $1.75/$14 | `@test_generator` |
-| **Docs** | Inline + external docs | Gemini 3 Flash | Budget | `@docs_generator` |
-| **PM** | Linear integration | Gemini 3 Flash | Budget | `@linear` |
-| **Committer** | Git automation | MiniMax Free | FREE | `@commiter` |
+| **QA** | Test generation, execution, BDD | GPT 5.4 Mini | $0.75/$4.50 | `@qa` |
+| **Docs** | Inline + external docs | Gemini 3 Flash | ~$0.50/$3 | `@docs_generator` |
+| **Pickle Think** | Free triage & rough planning | Big Pickle | FREE | `@pickle-think` |
+| **Pickle Implement** | Free low-risk code changes | Big Pickle | FREE | `@pickle-implement` |
 
-Agent configs live in [`opencode/aig_agents/`](opencode/aig_agents/). Rules live in [`skills/`](skills/) (symlinked into [`.claude/rules/`](.claude/rules/)) and [`.cursor/rules/`](.cursor/rules/).
+`planning-agent` now includes plan review mode, `qa` now handles requirements-driven / BDD test generation, and `pickle-think` / `pickle-implement` provide a free first-pass lane on Big Pickle. The default OpenCode build lane in [`opencode.json`](opencode.json) now uses `MiniMax M2.5 Free`; `@engineer` remains the premium escalation path on Gemini 3.1 Pro. Agent configs live in [`opencode/aig_agents/`](opencode/aig_agents/). Rules live in [`skills/`](skills/) (symlinked into [`.claude/rules/`](.claude/rules/)) and [`.cursor/rules/`](.cursor/rules/).
 
 For frontend design skills (`/audit`, `/polish`, `/critique`, `/animate`, `/frontend-design`, etc.), install [Impeccable](https://impeccable.style/):
 

opencode.json

@@ -1,10 +1,10 @@
 {
   "$schema": "https://opencode.ai/config.json",
-  "model": "opencode/gpt-5.4",
-  "small_model": "opencode/minimax-m2.5-free",
+  "model": "opencode/minimax-m2.5-free",
+  "small_model": "opencode/big-pickle",
   "mode": {
     "build": {
-      "model": "opencode/gpt-5.4",
+      "model": "opencode/minimax-m2.5-free",
       "tools": {
         "write": true,
         "edit": true,