Skill Creator
Purpose
Use this skill to build or improve reusable skills for this workspace without turning them into bloated promptware. A good skill should trigger cleanly, stay operational, keep its frontmatter lean, and reflect the house style: act first when the path is clear, keep warmth where it helps, and troubleshoot only when reality actually breaks.
Inputs expected
Required
- The workflow or problem the skill should handle.
- The target skill path or enough context to infer it.
Optional
- Existing
SKILL.md to improve.
- Reference docs, example prompts, or upstream skills to adapt.
- Whether the skill should stay minimal or include
references/ / scripts/.
Prerequisites
- Read
references/skills-reference-guide-for-agents.md before authoring or heavily restructuring a skill.
- Prefer exact paths and existing workspace conventions over invented structure.
Workflow
Confirm the task deserves a skill.
- A skill is warranted when the workflow recurs, has recognizable steps, and output quality improves when structure is reused.
- If the workflow is still mushy, tighten the use cases before writing anything.
Map the real trigger boundary.
- Define what the skill does, when it should load, and when it should stay quiet.
- Write the description from user intent, not internal architecture.
- Add negative scope when over-triggering is likely.
Design the smallest shape that will work.
- Default to one
SKILL.md.
- Add
references/ only when detailed material would bloat the main file.
- Add
scripts/ only when deterministic validation or transformation is genuinely better in code.
- Do not add folders just to look sophisticated.
Draft the frontmatter last, but keep it lean.
- Default to only
name and description unless extra fields are truly justified.
- Keep the description trigger-rich and compact.
- Put all trigger guidance in the frontmatter
description: what the skill does, when to use it, and when not to use it if overlap is likely.
- Do not add
When to use, Do not use when, Activation, Triggers, or similar trigger-boundary sections to the skill body. By the time the body is loaded, the agent should already know why this skill applies; if not, the description is wrong and must be improved.
- Avoid architecture-centred or product-marketing wording.
Write the operational body.
- Include the core job, boundaries, expected inputs, workflow, validation, error handling, output contract, and a few realistic examples.
- Keep body boundaries operational: prerequisites, safety limits, handoff rules, and workflow constraints are fine; trigger-selection guidance belongs in frontmatter only.
- Keep the step order explicit.
- Put critical rules near the top.
- Prefer exact file paths when pointing to supporting material.
Bake in house style deliberately.
- Assume availability first; do the normal thing before opening a troubleshooting box.
- Troubleshooting commands belong in a separate reference when possible, and should be read only when failure actually appears.
- If the path is clear and low-risk, act first instead of interrogating the user to death.
- Ask before destructive or high-blast-radius changes.
- Warm the prose when the skill is collaboration-facing or identity-adjacent, but keep procedural docs crisp.
When trigger quality feels off, correct it by feel first.
- If Igor says a skill does not trigger well enough, assume the description is too vague, too timid, or missing the actual user phrasing.
- Tighten or broaden the
description using the way Igor naturally asks for the thing.
- Add negative scope if the skill is barging in where it should stay quiet.
- Do not default to elaborate eval harnesses or benchmark theater unless the user explicitly wants that.
- Make the trigger sentence sharper, then let real use tell us if it still misses.
Validate the skill against the guide and the house.
- Trigger boundary is clear.
- Frontmatter is slim.
- Workflow is ordered.
- References are exact and only loaded when needed.
- It does not turn every task into diagnostics-first ritual.
- It does not become doctrine when a bounded tool would do.
Ship the skill cleanly.
- Create or update the files directly when the target path and intent are obvious.
- Summarize what was created or changed and why.
- If useful, call out the trigger sentence so it is easy to sanity-check.
Validation
- The skill solves a recurring workflow, not a one-off.
- The description states what it does, when to use it, and when not to use it if overlap is likely.
- Frontmatter is not bloated.
- The body does not contain
When to use, Do not use when, or equivalent trigger-selection sections.
SKILL.md stays operational; heavy detail is moved out.
- The workflow does the normal thing first and troubleshoots only on pain.
- If trigger quality was the complaint, the description was adjusted directly instead of disappearing into eval theater.
- Examples are concrete enough to teach the shape.
- File paths in references are exact.
Error handling
Error: workflow is too vague
Action: reduce it to 2-3 concrete use cases before drafting the skill.
Error: over-triggering or under-triggering risk
Action: tighten or broaden the description with explicit examples and exclusions.
Error: main file is getting bloated
Action: move factual detail or edge cases into references/.
Error: deterministic check is too fuzzy in prose
Action: move that check into scripts/ only if the repeatable logic really benefits from code.
Error: copied external skill is preachy or tool-religious
Action: keep the useful workflow, delete the church.
Output contract
A completed skill-creation pass should leave:
- a valid
SKILL.md in the right folder
- any needed
references/ or scripts/ files
- a clear trigger description
- explicit boundaries and workflow
- a concise summary of what changed and why
Examples
Example 1
User says: "Make me a skill for turning recurring repo spelunking into a reusable workflow."
Expected behaviour:
- Confirm the workflow is recurring.
- Read the skills guide.
- Draft a minimal skill with a sharp trigger boundary.
- Add references only if the main file starts getting fat.
Example 2
User says: "This skill is too robotic and keeps running diagnostics first. Fix it."
Expected behaviour:
- Read the existing skill.
- Identify the diagnostics-first smell.
- Split troubleshooting into a separate reference if needed.
- Rework the default workflow so it tries the normal path first.
Example 3
User says: "Port this upstream skill, but make it fit our house."
Expected behaviour:
- Extract the real workflow from the upstream material.
- Keep the useful patterns.
- Remove doctrinal, bloated, or tool-religious framing.
- Rebuild it with our trigger style, boundaries, and house posture.
1---2name: skill-creator-33description: Designs, writes, refactors, and packages agent skills for this workspace. Use when the user asks to create a new skill, improve an existing `SKILL.md`, tighten trigger descriptions, split heavy detail into `references/` or `scripts/`, add examples or validation, or port a workflow into a reusable skill. Do not use for one-off prompt tweaks, strict deployment runbooks, or vague workflows that are not repeatable yet.4metadata:5 source: https://skillmd.com/skills/igorwarzocha/skill-creator-36---78# Skill Creator910## Purpose1112Use this skill to build or improve reusable skills for this workspace without turning them into bloated promptware. A good skill should trigger cleanly, stay operational, keep its frontmatter lean, and reflect the house style: act first when the path is clear, keep warmth where it helps, and troubleshoot only when reality actually breaks.1314## Inputs expected1516### Required1718- The workflow or problem the skill should handle.19- The target skill path or enough context to infer it.2021### Optional2223- Existing `SKILL.md` to improve.24- Reference docs, example prompts, or upstream skills to adapt.25- Whether the skill should stay minimal or include `references/` / `scripts/`.2627## Prerequisites2829- Read `references/skills-reference-guide-for-agents.md` before authoring or heavily restructuring a skill.30- Prefer exact paths and existing workspace conventions over invented structure.3132## Workflow33341. **Confirm the task deserves a skill.**35 - A skill is warranted when the workflow recurs, has recognizable steps, and output quality improves when structure is reused.36 - If the workflow is still mushy, tighten the use cases before writing anything.37382. **Map the real trigger boundary.**39 - Define what the skill does, when it should load, and when it should stay quiet.40 - Write the description from user intent, not internal architecture.41 - Add negative scope when over-triggering is likely.42433. **Design the smallest shape that will work.**44 - Default to one `SKILL.md`.45 - Add `references/` only when detailed material would bloat the main file.46 - Add `scripts/` only when deterministic validation or transformation is genuinely better in code.47 - Do not add folders just to look sophisticated.48494. **Draft the frontmatter last, but keep it lean.**50 - Default to only `name` and `description` unless extra fields are truly justified.51 - Keep the description trigger-rich and compact.52 - Put all trigger guidance in the frontmatter `description`: what the skill does, when to use it, and when not to use it if overlap is likely.53 - Do not add `When to use`, `Do not use when`, `Activation`, `Triggers`, or similar trigger-boundary sections to the skill body. By the time the body is loaded, the agent should already know why this skill applies; if not, the `description` is wrong and must be improved.54 - Avoid architecture-centred or product-marketing wording.55565. **Write the operational body.**57 - Include the core job, boundaries, expected inputs, workflow, validation, error handling, output contract, and a few realistic examples.58 - Keep body boundaries operational: prerequisites, safety limits, handoff rules, and workflow constraints are fine; trigger-selection guidance belongs in frontmatter only.59 - Keep the step order explicit.60 - Put critical rules near the top.61 - Prefer exact file paths when pointing to supporting material.62636. **Bake in house style deliberately.**64 - Assume availability first; do the normal thing before opening a troubleshooting box.65 - Troubleshooting commands belong in a separate reference when possible, and should be read only when failure actually appears.66 - If the path is clear and low-risk, act first instead of interrogating the user to death.67 - Ask before destructive or high-blast-radius changes.68 - Warm the prose when the skill is collaboration-facing or identity-adjacent, but keep procedural docs crisp.69707. **When trigger quality feels off, correct it by feel first.**71 - If Igor says a skill does not trigger well enough, assume the description is too vague, too timid, or missing the actual user phrasing.72 - Tighten or broaden the `description` using the way Igor naturally asks for the thing.73 - Add negative scope if the skill is barging in where it should stay quiet.74 - Do not default to elaborate eval harnesses or benchmark theater unless the user explicitly wants that.75 - Make the trigger sentence sharper, then let real use tell us if it still misses.76778. **Validate the skill against the guide and the house.**78 - Trigger boundary is clear.79 - Frontmatter is slim.80 - Workflow is ordered.81 - References are exact and only loaded when needed.82 - It does not turn every task into diagnostics-first ritual.83 - It does not become doctrine when a bounded tool would do.84859. **Ship the skill cleanly.**86 - Create or update the files directly when the target path and intent are obvious.87 - Summarize what was created or changed and why.88 - If useful, call out the trigger sentence so it is easy to sanity-check.8990## Validation9192- The skill solves a recurring workflow, not a one-off.93- The description states what it does, when to use it, and when not to use it if overlap is likely.94- Frontmatter is not bloated.95- The body does not contain `When to use`, `Do not use when`, or equivalent trigger-selection sections.96- `SKILL.md` stays operational; heavy detail is moved out.97- The workflow does the normal thing first and troubleshoots only on pain.98- If trigger quality was the complaint, the description was adjusted directly instead of disappearing into eval theater.99- Examples are concrete enough to teach the shape.100- File paths in references are exact.101102## Error handling103104### Error: workflow is too vague105106Action: reduce it to 2-3 concrete use cases before drafting the skill.107108### Error: over-triggering or under-triggering risk109110Action: tighten or broaden the description with explicit examples and exclusions.111112### Error: main file is getting bloated113114Action: move factual detail or edge cases into `references/`.115116### Error: deterministic check is too fuzzy in prose117118Action: move that check into `scripts/` only if the repeatable logic really benefits from code.119120### Error: copied external skill is preachy or tool-religious121122Action: keep the useful workflow, delete the church.123124## Output contract125126A completed skill-creation pass should leave:127128- a valid `SKILL.md` in the right folder129- any needed `references/` or `scripts/` files130- a clear trigger description131- explicit boundaries and workflow132- a concise summary of what changed and why133134## Examples135136### Example 1137138User says: "Make me a skill for turning recurring repo spelunking into a reusable workflow."139140Expected behaviour:1411421. Confirm the workflow is recurring.1432. Read the skills guide.1443. Draft a minimal skill with a sharp trigger boundary.1454. Add references only if the main file starts getting fat.146147### Example 2148149User says: "This skill is too robotic and keeps running diagnostics first. Fix it."150151Expected behaviour:1521531. Read the existing skill.1542. Identify the diagnostics-first smell.1553. Split troubleshooting into a separate reference if needed.1564. Rework the default workflow so it tries the normal path first.157158### Example 3159160User says: "Port this upstream skill, but make it fit our house."161162Expected behaviour:1631641. Extract the real workflow from the upstream material.1652. Keep the useful patterns.1663. Remove doctrinal, bloated, or tool-religious framing.1674. Rebuild it with our trigger style, boundaries, and house posture.