SkillForge is OpenCode-first. This guide explains how its workflow can be adapted to other coding agents.
| Agent | Works directly? | Mechanism | Status |
|---|---|---|---|
| OpenCode | Yes | .opencode/skills/skillforge/SKILL.md |
Primary target. Fully supported. |
| Claude Code | Partial | Claude Agent Skill (same SKILL.md format) | Workflow is portable. Not fully tested. |
| Codex | Partial | AGENTS.md / CODEX.md project instructions |
Workflow logic can be ported. Format differs. |
| Cursor | Partial | Cursor Rules / AGENTS.md |
Workflow logic can be ported. Format differs. |
| Gemini CLI | Partial | Skill activation via tool mapping | Workflow is adaptable. Not tested. |
| Other agents | Partial | Custom instruction files | Workflow steps are reusable as guidelines. |
The following aspects of SkillForge are agent-agnostic:
- The workflow (Intent Analysis → Composition Analysis → Community Discovery → Project Context → Questions → Skill Spec → Quality Audit → Instructions)
- The
templates/(skill-spec.md, skill-audit.md, skill-readme.md) - The
examples/(expected workflow outputs) - The quality audit rubric
- The safety boundaries and do-not rules
- The decision rubrics for existing skills, MCP, and community resources
| Feature | OpenCode | Other agents |
|---|---|---|
| Auto-activation | Uses .opencode/skills/ convention |
May require manual instruction-based activation |
| SKILL.md format | Native format | May need conversion to agent-specific instruction format |
| Tool references | OpenCode-native tools | Must be mapped to the target agent's tool set |
| Skill directory convention | .opencode/skills/ |
Each agent has its own convention for custom skills |
When adapting SkillForge to another agent, preserve:
- Workflow sequence — keep the 7 steps in order
- Output formats — keep the structured headings, tables, and code blocks
- Safety rules — keep all do-not rules and safety boundaries
- Decision rubrics — keep the fit-level rubric, skill-vs-MCP rubric, and community safety rules
- Audit rubric — keep the 9-criterion quality audit
Adapt:
- Activation triggers — rewrite for the target agent's trigger mechanism
- File paths — adjust
.opencode/skills/to the target agent's skill directory - Tool names — replace OpenCode tool names with equivalent tools in the target agent
- Not tested — SkillForge has not been fully tested on Claude Code, Codex, Cursor, Gemini CLI, or other non-OpenCode agents. The workflow is designed to be portable, but actual behavior depends on each agent's implementation.
- No native support claims — Do not claim "native support" for any agent unless verified through testing.
- Review third-party resources — Community skills, MCP servers, and public repositories referenced during Community Discovery should be reviewed for safety and relevance before use.