my-opencode-config

Quick-reference card for all caveman modes, skills, and commands. One-shot display, not a persistent mode. Trigger: /caveman-help, "caveman help", "what caveman commands", "how do I use caveman".

Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

Review upstream changes to the local my-opencode reviewer swarm before selectively applying them. Use whenever the user asks to pull, sync, inspect, compare, or adopt reviewer or swarm changes from ~/Documents/yo/my-opencode. Never use it to install the upstream configuration wholesale.

Quick-reference card for all ponytail modes, skills, and commands. One-shot display, not a persistent mode. Trigger: /ponytail-help, "ponytail help", "what ponytail commands", "how do I use ponytail".

Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports intensity levels: lite, full (default), ultra. Use whenever the user says "ponytail", "be lazy", "lazy mode", "simplest solution", "minimal solution", "yagni", "do less", or "shortest path", and whenever they complain about over-engineering, bloat, boilerplate, or unnecessary dependencies.

Use whenever the task involves writing, drafting, revising, or reviewing a system prompt for an AI agent — autonomous or semi-autonomous LLM systems that use tools in a loop, including operator agents, research/sub-agents, tool-using assistants, MCP servers, multi-agent orchestrators, and coding agents. Trigger even when the user says "instructions", "system message", "agent persona", or "rules for the model", not just the word "prompt". Do NOT use for one-shot prompts (classification, single generation, single Q&A) — those need lighter prompt techniques, not the full agent skeleton.

Source-first SDK and framework research with btca. Use when a task depends on current behavior or recommended usage of AI SDK, Cloudflare, Effect, PostHog, or adjacent integrations, and you want the agent to consult btca before relying on memory or generic web search.

Post-implementation cleanup gate for non-trivial or visibly AI-generated diffs. Use before final verification to remove unnecessary type declarations, redundant code, and obsolete paths without broad refactoring.

Use when reviewing code changes — pull requests, git diffs, uncommitted work, patch sets, or any "is this code OK?" question. Trigger phrases include "review this", "check my changes", "is this ready", "does this look right", "PR feedback", "code review", "look at my diff", "audit this code", "did I break anything". Also use when an agent is asked to verify whether a change solves a stated problem, follows codebase patterns, or is regression-safe. Do NOT use for greenfield code-writing requests — those need different guidance.

CodeGraph MCP server setup — install, index, and configure semantic code search. Use when setting up CodeGraph in a new repo, re-indexing, or troubleshooting codegraph MCP.

Groups git changes by logical purpose and commits them sequentially. Use when user runs /commit, asks to commit changes, or wants to review, group, and commit git changes.

Stage changes, create a conventional commit, push the branch, and create or refresh PR metadata for shipped work. Use when the user asks to ship current work to GitHub.

Diagnose hard bugs and performance regressions with a reproducible feedback loop. Use when a failure is intermittent, poorly understood, repeatedly misdiagnosed, or needs root-cause evidence before a fix. Skip for obvious one-line errors with a proven cause.

Turn a rough task brief, vague ambition, or draft Codex /goal into a compact, evidence-driven goal for long-running autonomous work. Use when the user asks to write, improve, sharpen, stress-test, or activate a Codex goal, thread goal, durable goal, or /goal prompt.

Compact the current session into a redacted handoff for a fresh agent. Use when the user asks for a handoff, continuation document, session transfer, or a durable summary before starting a new chat.

Inspect this Mac's current hardware, OS, shell, and developer toolchain when host-specific facts are relevant. Use live commands instead of relying on stored version or free-space values.

Prepare PRs for review by cleaning noisy history, improving PR descriptions, and adding reviewer guidance without changing code behavior. Use for "make this easy to review", "tidy this PR", "clean up commits", or "annotate the diff".

Python project setup and management using the MiquelGomezCorral template. Use for virtual environments (conda), dependency management (uv), editable installs, Jupyter kernels, and project scaffolding.

Two-phase correctness and convention gate. Use when the user explicitly asks for a staff review, quality gate, convention audit, or `full`/`quick`/`deep` review mode. Use `code-review` alone for ordinary code review.

Security scanner for AI agent skills. Scan skills, repositories, or directories for vulnerabilities before installing. Trigger on "scan skill", "check skill safety", "audit skill", "skill security", "skillspector", "is this skill safe".

Pressure-test a plan, design, architecture, implementation approach, or product decision through a direct decision-tree interview. Use when the user asks to be grilled, stress-tested, challenged, interviewed, or pressure-tested before execution.

General multi-agent orchestration for OpenCode. Use proactively whenever a non-trivial task can benefit from independent parallel research, investigation, review, testing, debugging, comparison, verification, or disjoint implementation. Also trigger when the user asks for a swarm, multiple agents, parallel agents, subagents, independent opinions, or broader coverage. Keep trivial, tightly coupled, or single-writer work in the main thread.

Use ONLY when editing or creating opencode's own config: opencode.json, skills, commands, agents, plugins, MCP servers, instructions, permissions, security rules. Not for application code. Trigger on 'configure opencode', 'opencode setup', 'add skill to opencode', 'update opencode config', 'opencode plugin'.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Explain or use automatic OpenCode subagent model recovery. Use when selecting a model for a subagent, diagnosing fallback behavior, or handling a final task failure after all configured models were attempted.

Persistent project memory bootstrap and maintenance. Use in any repository with (or needing) an `AGENTS.md` + `./agents/` system. **Triggers:**\n* **Initialization:** Starting in a new repo, running `/init`, or requests to "set up agents memory."\n* **Development:** Planning features, proposing architecture (→ decisions), writing code (→ conventions), or navigating the codebase (→ architecture).\n* **Debugging:** Fixing complex bugs, investigating issues, or encountering recurring mistakes (→ known-errors).\n* **Memory Management:** Fetching or proposing workflows, glossary terms, conventions, decisions, or known errors.

Operate the btca CLI for local resources and source-first answers. Use when setting up btca in a project, connecting a provider, adding or managing resources, and asking questions via btca commands. Invoke this skill when the user says "use btca" or needs to do more detailed research on a specific library or framework.

Guidelines for technical writing, formal paper formatting, and LaTeX document structures. Use when writing academic papers, technical documents, LaTeX formatting, bibliography, or formal writing.

Use for Obsidian vault tasks via vault-context and filesystem tools. Trigger on words like "obsidian", "vault", "note", "daily note", "frontmatter", "append note", "move note".

Turn the current session's work into an incremental teaching loop. Use when the user wants to deeply understand work just completed — asks to be taught, quizzed, or walked through what was done and why ("teach me", "quiz me", "make sure I understand this").

Use this skill for any task that needs web search, current information, external docs, source discovery, fact checking, URL reading, news, product/pricing/status lookup, citations, or anything beyond local files and memory. Load it even if the user does not explicitly say "search the web" when the answer depends on non-local or up-to-date information. Do not use when the task is purely local repo/vault search or the user says not to use the web.