Skip to content

AGENTS.md

Latest commit

 

History

History
111 lines (83 loc) · 5.17 KB

AGENTS.md

File metadata and controls

111 lines (83 loc) · 5.17 KB

SDLC Baseline (OpenCode)

This file is loaded automatically at the start of every OpenCode session in this project. It carries the same SDLC enforcement rules that the sdlc-wizard ecosystem applies across Claude Code, Codex, and OpenCode.

SDLC Baseline (every session)

  1. Plan before coding — state confidence level (HIGH / MEDIUM / LOW)
  2. TDD: Write failing test FIRST, then implement
  3. ALL tests must pass before commit — no exceptions
  4. Self-review before presenting — read back the diff, look for obvious issues
  5. LOW confidence → ASK USER before proceeding
  6. Failed twice → STOP and ASK USER — don't spin

Agent Skills Available

Skills are auto-discovered from .opencode/skills/ (and Claude-compatible .claude/skills/). The four canonical skills:

Skill Invocation Purpose
sdlc skill({ name: "sdlc" }) Full SDLC workflow guidance
setup-wizard skill({ name: "setup-wizard" }) Confidence-driven project setup
update-wizard skill({ name: "update-wizard" }) Smart update with drift detection
feedback skill({ name: "feedback" }) Privacy-first community feedback

Read each skill's SKILL.md for full guidance.

Workflow Phases

  1. Plan (research + outline) → present approach + confidence
  2. Transition (update docs) — feature docs current before commit
  3. Implementation — TDD RED → GREEN → PASS
  4. Self-review — read modified files, check for issues
  5. Cross-model review (high-stakes changes) — see SDLC skill

Hook Behavior (this project)

OpenCode does not have a Claude-style declarative hook config; the .opencode/plugins/sdlc-wizard.js plugin subscribes to events and shells out to bash hooks at .opencode/hooks/:

Event Hook Purpose
session.created instructions-loaded-check.sh + model-effort-check.sh Validate SDLC files exist; nudge model/effort upgrades
tool.execute.before (Write/Edit on source) tdd-pretool-check.sh TDD reminder before source edits
experimental.session.compacting precompact-seam-check.sh Block compact mid-review or in-flight rebase/merge

There is no UserPromptSubmit analog in OpenCode, so the per-prompt SDLC BASELINE reminder from the Claude/Codex siblings is replaced by this AGENTS.md content (loaded once per session). If you need a stronger nudge cadence, invoke skill({ name: "sdlc" }) explicitly.

Recommended Backend Configuration

OpenCode supports many backends via opencode.json's model field. This wizard ships a privacy-first picker that defaults to the strongest data-locality guarantee available on the machine.

Pick a backend in two steps:

# What's reachable on this machine?
bash .opencode/scripts/detect-backends.sh

# Or bias toward free providers (Cerebras / Groq / NVIDIA NIM / Google AI Studio quotas)
bash .opencode/scripts/detect-backends.sh --free-tier-first

# Configure the highest-privacy tier you can use
bash .opencode/scripts/configure-backend.sh \
     --tier private_local --provider ollama \
     --model qwen3-coder:30b

Tiers (privacy-first ordering):

Tier Where prompts travel Examples
private_local Stays on your machine Ollama, LM Studio, llama.cpp, vLLM, MLX (Apple Silicon)
enterprise Stays in your tenant Azure OpenAI, AWS Bedrock
hosted_oss Open weights, third-party host Together, Groq, OpenRouter, Cerebras, DeepSeek direct, NVIDIA NIM (nvidia_nim)
proprietary Vendor-bound Anthropic, OpenAI, Google AI Studio (google_aistudio — Gemini), Z.AI (zai — GLM Coding Plan, post-Anthropic-OAuth-ban migration target)
managed OpenCode-routed PAYG OpenCode Zen (opencode — 40+ models incl. free tier; official new-user entry point)
subscription Pre-paid sub via OAuth GitHub Copilot Pro+ (github-copilot), ChatGPT Plus/Pro (openai-codex / chatgpt), SuperGrok (grok / xai) — all via opencode /connect

See docs/cost-ladder.md for $0/$20/$200 monthly budget paths and a per-job picker (routine fix vs long-context refactor vs CI gate vs security audit).

The SDLC enforcement works on any backend that hits the capability floor (30B+ code-tuned models — Qwen-Coder, DeepSeek-Coder, Sonnet, Opus, GPT-5.x). Smaller models (7–13B) typically fail the full plan→TDD→self-review protocol; that's a capability result, not a wizard bug.

See PRIVACY.md for full tier walkthroughs, the Ollama private-path setup, and how to verify no prompts are leaving the machine.

Sibling Family

This wizard is one of four published siblings:

Package Agent Repo
agentic-sdlc-wizard Claude Code https://github.com/BaseInfinity/claude-sdlc-wizard
codex-sdlc-wizard Codex CLI https://github.com/BaseInfinity/codex-sdlc-wizard
claude-gdlc-wizard Claude Code (games) https://github.com/BaseInfinity/claude-gdlc-wizard
opencode-sdlc-wizard (this) OpenCode https://github.com/BaseInfinity/opencode-sdlc-wizard

Part of the XDLC ecosystem.