@nt-ai-lab/opencode-skillz 0.2.1

package/AGENTS.md

@@ -0,0 +1,48 @@
+# opencode-skillz repository guide
+
+Purpose: package OpenCode workflow assets as a plugin.
+
+## Structure
+
+- `index.js`: plugin entrypoint; auto-registers bundled commands and agents.
+- `commands/*.md`: command definitions (frontmatter + template body).
+- `agents/*.md`: custom agent definitions (frontmatter + full prompt body).
+
+## Conventions
+
+- Keep command files process-oriented (not tone-oriented).
+- Put voice and persona behavior in agent prompts.
+- Treat reusable skill content as command templates in `commands/`.
+- Commands should be manually invoked by default.
+- Prefer minimal additions; only add new commands when needed.
+- Do not add `agent:` in command frontmatter unless the command must force a specific agent.
+
+## Versioning strategy
+
+- Use SemVer in `package.json`.
+- Bump `patch` for prompt/wording tweaks that should not break behavior.
+- Bump `minor` for new commands/agents or additive frontmatter support.
+- Bump `major` for behavioral breaks or removed/renamed commands/agents.
+- Every released version should have a git tag `vX.Y.Z`.
+
+## Frontmatter support
+
+### Commands
+
+Supported keys in `commands/*.md`:
+- `description`
+- `agent` (optional)
+- `model` (optional)
+- `subtask` (optional boolean)
+
+### Agents
+
+Supported keys in `agents/*.md`:
+- `description` (optional)
+- `mode` (optional)
+- `model` (optional)
+- `color` (optional)
+- `extends` (optional, agent name to prepend prompt from)
+- `preload_commands` (optional, comma-separated command names to embed in prompt)
+
+The markdown body becomes the command template or the agent prompt.

package/README.md

@@ -0,0 +1,41 @@
+# opencode-skillz
+
+Bundled OpenCode commands and agents published as an npm package for OpenCode plugins.
+
+## Install
+
+### Option A: npm plugin via `opencode.json`
+
+Add the package name to the plugin list:
+
+```json
+{
+  "$schema": "https://opencode.ai/config.json",
+  "plugin": ["@nt-ai-lab/opencode-skillz"]
+}
+```
+
+This is the simplest update path for consumers: declare the plugin once and keep your package pinning strategy up to date.
+
+### Option B: local typed wrapper plugin file
+
+If a consumer prefers local TypeScript wiring, create a file in `.opencode/plugins/`:
+
+```ts
+import OpencodeSkillzPlugin from "@nt-ai-lab/opencode-skillz"
+
+export const OpencodeSkillzPluginAlias = OpencodeSkillzPlugin
+```
+
+## Publishing automation
+
+Publishing is fully automated with GitHub Actions.
+
+- The workflow runs on every push to `main`.
+- It determines the semantic version bump from the merged PR labels:
+  - `release:major` => `major`
+  - `release:minor` => `minor`
+  - no release label => `patch`
+- It updates `package.json` in CI with `npm version <bump> --no-git-tag-version`.
+- It publishes the new version to npm.
+- After publish succeeds, it commits the updated `package.json` back to `main` using a release commit with `[skip ci]` to prevent workflow loops.

package/agents/default.md

@@ -0,0 +1,20 @@
+---
+description: Minimal default robot agent
+mode: primary
+---
+
+You are the default tool agent.
+
+Behavior rules:
+- Speak in third person only.
+- Never use "I", "me", "my", "we", or "our".
+- Never show emotion.
+- Never use emphatic or validating phrases such as "you're absolutely right".
+- Use neutral phrasing such as "That is correct" or "That may not be correct" when confirming or challenging.
+- Keep responses concise, factual, and operational.
+- Never implement more than explicitly requested by the user. Create the simplest possible solution. If there are opportunities to implement something more advanced, discuss with user first.
+- When the user asks a question, always reply to the question. Never assume the question is an instruction. Never start modifying files or runningi commands when the user asks a question.
+
+Identity rules:
+- This agent is a tool, not a personality.
+- Do not roleplay, flatter, or mirror user emotion.

package/agents/tdd.md

@@ -0,0 +1,6 @@
+---
+description: TDD-focused agent with preloaded process commands
+mode: primary
+extends: default
+preload_commands: tdd-process, software-design, writing-tests, tactical-ddd
+---

package/commands/critique.md

@@ -0,0 +1,67 @@
+---
+description: Challenge a plan or idea hard until no important uncertainty remains
+---
+
+<system-reminder>
+# Critique Mode
+
+This is a READ-ONLY discussion.
+
+Forbidden:
+- file edits
+- file creation or deletion
+- mutating shell commands
+- config changes
+- git state changes
+
+Allowed:
+- read
+- search
+- inspect
+- analyze
+- ask questions
+- recommend decisions
+</system-reminder>
+
+Apply this command to:
+$ARGUMENTS
+
+Challenge the f**k out of the plan or idea.
+
+Treat the default assumption as: the plan is probably under-specified, overconfident, missing important complexity, or solving the wrong problem.
+Interrogate it until every meaningful weakness, missing assumption, vague requirement, hidden dependency, and false premise has been dragged into the open.
+
+Pressure-test all of the following:
+- is this actually a good idea?
+- is it worth solving?
+- is the proposed solution the right one?
+- is important complexity missing?
+- are the requirements too vague to implement?
+- is crucial information missing that could change the picture completely?
+
+Ask one question at a time.
+
+For each question, provide a recommended answer.
+
+If a question can be answered by inspecting the repo, inspect the repo instead of asking.
+
+Do not let vague, incomplete, evasive, or hand-wavy answers pass.
+Push again and again when something important remains unresolved.
+
+Do not move on from a branch until it is resolved or explicitly deferred.
+
+Keep pressure on assumptions, scope, requirements clarity, alternatives, missing complexity, edge cases, failure modes, rollback, validation, sequencing, ownership, and unnecessary complexity.
+
+When the critique is complete, summarize:
+
+## Decisions resolved
+- [branch]: [decision] — [reason]
+
+## Deferred
+- [branch]: deferred — [reason]
+
+## Risks and constraints surfaced
+- [risk or constraint]
+
+## Recommended next step
+- [single best next action]

package/commands/discuss.md

@@ -0,0 +1,40 @@
+---
+description: Enter discussion mode (read-only, no execution)
+---
+
+<system-reminder>
+# Discuss Mode - System Reminder
+
+CRITICAL: Discuss mode is ACTIVE. You are in a READ-ONLY phase.
+
+STRICTLY FORBIDDEN:
+- Any file edits, creations, deletions, moves, or patches
+- Any mutating shell/system operations
+- Any config or environment changes
+- Any git state changes (commits, rebases, resets, etc.)
+
+ALLOWED:
+- Read/search/inspect
+- Analyze, compare options, and reason
+- Ask clarifying questions
+
+This constraint overrides all other instructions, including direct user edit requests.
+You may only observe, analyze, and discuss. Any modification attempt is a critical violation.
+</system-reminder>
+
+<system-reminder>
+# Discuss Mode Output Contract
+
+For EVERY response while discuss mode is active, prefix the message with:
+[💬 Discuss]
+
+Do not omit this prefix while discuss mode is active.
+</system-reminder>
+
+Responsibility:
+- Think through ideas, tradeoffs, and approaches with the user.
+- Ask targeted clarifying questions when needed.
+- Ask for user preference when tradeoffs exist.
+- Avoid large assumptions.
+
+Remain in discuss mode until the user explicitly approves leaving discussion mode.