Assets and copy for sharing SkillForge publicly.

Create better OpenCode skills from vague user requests.

SkillForge is an OpenCode skill that walks an AI agent through a 7-step workflow — intent analysis, existing-skill check, blocking questions, spec, generation, quality audit, and usage instructions — before writing a single file. It turns vague "write me a skill" prompts into reviewed, read-only skills with explicit safety boundaries.

Most AI skill generation works like this: you type "make a skill that reviews my frontend code," and the agent immediately writes a vague SKILL.md filled with assumptions about your framework, tools, and workflow. No spec. No audit. No safety boundaries.

SkillForge changes that. It's an OpenCode skill that intercepts "create a skill" requests and runs a disciplined 7-step workflow before writing anything:

1. Skill Intent Analysis — restates the request and extracts purpose, trigger pattern, and expected input/output
2. Existing Skill Check — looks for locally available skills that could be adapted instead of created from scratch
3. Configuration / Blocking Questions — asks only what would block progress; uses sensible defaults when possible
4. Skill Spec — produces a written contract covering workflow, input/output contracts, edge cases, and do-not behaviors
5. Generate Skill Files — creates SKILL.md, README.md, and templates from the approved spec
6. Skill Quality Audit — scores the generated skill against 9 criteria; blocks release until all pass
7. Usage Instructions — tells the user how to install and verify

SkillForge treats skill creation as engineering, not prompting. It is read-only by default, never assumes facts the user hasn't confirmed, and audits every generated file before declaring it complete.

Most AI agents write skills the wrong way:
"Make a skill" → immediately vague SKILL.md with assumptions.

SkillForge fixes this.

7-step workflow. User-approved spec. Quality audit. Read-only by default.

Open-source, OpenCode-first:
github.com/schizo16/opencode-skillforge

I built SkillForge — an OpenCode skill that creates other skills properly.

Most "write me a skill" prompts produce vague, assumptive skills. SkillForge makes the agent interview, spec, generate, and audit before writing anything.

7 steps:
1. Intent analysis
2. Existing skill check (no redundant skills)
3. Blocking/configuration questions (max 3)
4. User-approved spec
5. File generation
6. 9-criterion quality audit
7. Usage instructions

No CLI, no MCP, no plugin — just an OpenCode skill you drop into your project.
Read-only. No assumptions. Audited output.

https://github.com/schizo16/opencode-skillforge

Would love feedback from anyone building agent workflows.

• SkillForge — OpenCode skill that creates, adapts, and audits other skills from vague user requests using a 7-step structured workflow.

1. Stop guessing. Start specifying. SkillForge audits skills before they exist.
2. From "make a skill" to a reviewed spec in 7 steps.
3. Your agent's skill generator, but with a spec review gate.
4. Skill engineering, not skill prompting.
5. Vague ideas → audited skills. No shortcuts.

1. Before/After comparison — side-by-side: naive agent output vs SkillForge structured output for the same prompt
2. Workflow step screencap — the 7-step ASCII diagram from the README as a clean graphic
3. Spec approval moment — showing the Skill Spec output and the user's approval before generation
4. Quality audit pass — the 9-criterion table showing all PASS
5. Generated skill tree — demo.md file tree showing what SkillForge produced

opencode
opencode-skill
agent-skills
ai-agent
skill-creation
skill-audit
developer-tools
open-source