sm

modules/_opencode/command/supermemory-init.md

@@ -0,0 +1,150 @@
+---
+description: Initialize Supermemory with comprehensive codebase knowledge
+---
+
+# Initializing Supermemory
+
+You are initializing persistent memory for this codebase. This is not just data collection - you're building context that will make you significantly more effective across all future sessions.
+
+## Understanding Context
+
+You are a **stateful** coding agent. Users expect to work with you over extended periods - potentially the entire lifecycle of a project. Your memory is how you get better over time and maintain continuity.
+
+## What to Remember
+
+### 1. Procedures (Rules & Workflows)
+Explicit rules that should always be followed:
+- "Never commit directly to main - always use feature branches"
+- "Always run lint before tests"
+- "Use conventional commits format"
+
+### 2. Preferences (Style & Conventions)
+Project and user coding style:
+- "Prefer functional components over class components"
+- "Use early returns instead of nested conditionals"
+- "Always add JSDoc to exported functions"
+
+### 3. Architecture & Context
+How the codebase works and why:
+- "Auth system was refactored in v2.0 - old patterns deprecated"
+- "The monorepo used to have 3 modules before consolidation"
+- "This pagination bug was fixed before - similar to PR #234"
+
+## Memory Scopes
+
+**Project-scoped** (\`scope: "project"\`):
+- Build/test/lint commands
+- Architecture and key directories
+- Team conventions specific to this codebase
+- Technology stack and framework choices
+- Known issues and their solutions
+
+**User-scoped** (\`scope: "user"\`):
+- Personal coding preferences across all projects
+- Communication style preferences
+- General workflow habits
+
+## Research Approach
+
+This is a **deep research** initialization. Take your time and be thorough (~50+ tool calls). The goal is to genuinely understand the project, not just collect surface-level facts.
+
+**What to uncover:**
+- Tech stack and dependencies (explicit and implicit)
+- Project structure and architecture
+- Build/test/deploy commands and workflows
+- Contributors & team dynamics (who works on what?)
+- Commit conventions and branching strategy
+- Code evolution (major refactors, architecture changes)
+- Pain points (areas with lots of bug fixes)
+- Implicit conventions not documented anywhere
+
+## Research Techniques
+
+### File-based
+- README.md, CONTRIBUTING.md, AGENTS.md, CLAUDE.md
+- Package manifests (package.json, Cargo.toml, pyproject.toml, go.mod)
+- Config files (.eslintrc, tsconfig.json, .prettierrc)
+- CI/CD configs (.github/workflows/)
+
+### Git-based
+- \`git log --oneline -20\` - Recent history
+- \`git branch -a\` - Branching strategy
+- \`git log --format="%s" -50\` - Commit conventions
+- \`git shortlog -sn --all | head -10\` - Main contributors
+
+### Explore Agent
+Fire parallel explore queries for broad understanding:
+\`\`\`
+Task(explore, "What is the tech stack and key dependencies?")
+Task(explore, "What is the project structure? Key directories?")
+Task(explore, "How do you build, test, and run this project?")
+Task(explore, "What are the main architectural patterns?")
+Task(explore, "What conventions or patterns are used?")
+\`\`\`
+
+## How to Do Thorough Research
+
+**Don't just collect data - analyze and cross-reference.**
+
+Bad (shallow):
+- Run commands, copy output
+- List facts without understanding
+
+Good (thorough):
+- Cross-reference findings (if inconsistent, dig deeper)
+- Resolve ambiguities (don't leave questions unanswered)
+- Read actual file content, not just names
+- Look for patterns (what do commits tell you about workflow?)
+- Think like a new team member - what would you want to know?
+
+## Saving Memories
+
+Use the \`supermemory\` tool for each distinct insight:
+
+\`\`\`
+supermemory(mode: "add", content: "...", type: "...", scope: "project")
+\`\`\`
+
+**Types:**
+- \`project-config\` - tech stack, commands, tooling
+- \`architecture\` - codebase structure, key components, data flow
+- \`learned-pattern\` - conventions specific to this codebase
+- \`error-solution\` - known issues and their fixes
+- \`preference\` - coding style preferences (use with user scope)
+
+**Guidelines:**
+- Save each distinct insight as a separate memory
+- Be concise but include enough context to be useful
+- Include the "why" not just the "what" when relevant
+- Update memories incrementally as you research (don't wait until the end)
+
+**Good memories:**
+- "Uses Bun runtime and package manager. Commands: bun install, bun run dev, bun test"
+- "API routes in src/routes/, handlers in src/handlers/. Hono framework."
+- "Auth uses Redis sessions, not JWT. Implementation in src/lib/auth.ts"
+- "Never use \`any\` type - strict TypeScript. Use \`unknown\` and narrow."
+- "Database migrations must be backward compatible - we do rolling deploys"
+
+## Upfront Questions
+
+Before diving in, ask:
+1. "Any specific rules I should always follow?"
+2. "Preferences for how I communicate? (terse/detailed)"
+
+## Reflection Phase
+
+Before finishing, reflect:
+1. **Completeness**: Did you cover commands, architecture, conventions, gotchas?
+2. **Quality**: Are memories concise and searchable?
+3. **Scope**: Did you correctly separate project vs user knowledge?
+
+Then ask: "I've initialized memory with X insights. Want me to continue refining, or is this good?"
+
+## Your Task
+
+1. Ask upfront questions (research depth, rules, preferences)
+2. Check existing memories: \`supermemory(mode: "list", scope: "project")\`
+3. Research based on chosen depth
+4. Save memories incrementally as you discover insights
+5. Reflect and verify completeness
+6. Summarize what was learned and ask if user wants refinement

modules/ai-tools.nix

@@ -33,6 +33,7 @@ in {
 				small_model = "openai/gpt-5.1-codex-mini";
 				plugin = [
 					"opencode-claude-auth"
+					"opencode-supermemory"
 				];
 				permission = {
 					external_directory = {