| name | skillify |
| description | Convert a freshly completed task or described workflow into a reusable skill. Use when the user invokes $skillify, asks to create a skill from the current process, wants generic repeatable steps captured, or requests a skill with optional name/destination clauses such as "$skillify as Spec to English in ~/.agent_skills" or "$skillify in projectname". |
Skillify
Use this skill to turn a completed or described workflow into a reusable skill folder. The default output location is adjacent to this skillify skill. If the user provides as <NAME> or in <DIRECTORY>, infer the intended name or destination from natural language and honor it.
Workflow
- Infer the skill name and destination from the user's request.
- If there is no explicit name, infer a concise hyphen-case name from the workflow. Prefer action-oriented names under 64 characters.
- If there is no explicit destination, create the new skill adjacent to this
skillify skill; use the parent directory of the loaded skillify/SKILL.md.
- Identify concrete examples from the just-completed task or the user's description. If the source workflow is ambiguous, ask one concise clarifying question.
- Decompose the workflow into generic, repeatable steps.
- Classify each step:
- LLM-suited: translation, judgment, synthesis, naming, prioritization, visual QA, or context-specific choices.
- Script-suited: deterministic parsing, extraction, rendering, file transforms, validation, API calls with stable parameters, and repetitive boilerplate.
- Create the destination skill folder. Prefer the system
skill-creator initializer when available. If it is not available or is blocked, use:
python <skill-dir>/scripts/create_skill_skeleton.py <skill-name> --path <destination> --resources scripts
- Author scripts for the script-suited steps first. Keep scripts parameterized, deterministic, and runnable without loading large context.
- Write
SKILL.md with concise frontmatter and body instructions. Include when to invoke each script, where LLM judgment is required, and validation steps.
- Add
agents/openai.yaml with display_name, short_description, and a default_prompt that explicitly mentions $<skill-name>.
- Run script smoke tests and validate the skill:
python <skill-creator-dir>/scripts/quick_validate.py <destination>/<skill-name>
If quick_validate.py is unavailable, at minimum verify SKILL.md has valid YAML frontmatter with name and description, and the folder name matches the skill name.
Destination Rules
- No
in clause: create the skill adjacent to this skillify skill, meaning in the same parent directory as the loaded skillify folder.
- Absolute or
~ path: use that directory.
- Relative path containing
/: resolve from the current working directory.
- Bare names such as
projectname: infer the intended skill directory from the user's active workspace and known local conventions. Do not assume a fixed folder name such as agent_skills; inspect or ask if the destination is ambiguous.
Quality Bar
- Do not encode one-off facts as universal instructions.
- Do not put generated reports, README files, changelogs, or process notes in the skill.
- Keep the skill lean; move only reusable deterministic code into
scripts/.
- Test every new script with representative inputs before finishing.