agent-skill-creator

name	agent-skill-creator
description	Create a new single-purpose Agent Skill folder that reaches review-ready with a clear trigger, risk-appropriate validation, concise positive and negative examples, and explicit roles for all local files. Use this when asked to draft or scaffold a new Agent Skill for this repository.
complexity	high
risk_profile	["ambiguity_sensitive","code_modification"]
use_when	["a recurring task deserves its own Agent Skill","an existing skill is too broad and should be split","the user asks for a new skill scaffold or starter folder"]
do_not_use_when	["the task only needs a small edit to an existing skill","the change belongs in a local example or checklist file"]

Purpose

Create a new skill that reaches review-ready.

Trigger / When to use

Use this skill when:

• a recurring task deserves its own Agent Skill
• an existing skill is too broad and should be split
• the user asks for a new skill, scaffold, or starter folder

Do not use this skill when:

• the task only needs a small edit to an existing skill
• the change belongs in a local example or checklist file

Inputs

• the skill's single responsibility
• the situations that should trigger the skill
• the skill's boundaries
• the reference material, if any
• any local assets the skill truly needs
• the complexity level, if it is already known
• whether the skill is lightweight, medium-complexity, or higher-risk
• any branching, external-tool, or downstream-impact signals that increase validation needs
• any risk_profile tags that apply

Process

1. If the responsibility, trigger, or boundaries are ambiguous, stop and ask the user before drafting.
2. Start from blueprint.md and folder-contract.md.
3. Decide the validation weight before drafting:
   • lightweight: low-risk, low-branching guidance can stay concise
   • medium-complexity: decision-heavy guidance may need clearer process detail
   • higher-risk: gatekeeping, release, tool-driven, or high-impact guidance needs stronger misuse prevention
4. Propose complexity in YAML frontmatter. Propose applicable risk_profile tags for medium and high complexity skills.
5. If a topic plan locks a creator/reviewer-first rollout, keep the work inside creator, reviewer, and the minimum supporting policy/template files. Defer downstream regular skills to later topics instead of broadening the current one.
6. Create .github/skills/<skill-name>/, where <skill-name> must use lowercase kebab-case.
7. Keep the skill focused on one job.
8. Write SKILL.md with an explicit Trigger / When to use section and concise positive and negative examples.
9. Add reference.md or examples.md.
10. Use references/ only as a split-reference supplement, not as a replacement for the required companion-file rule.
11. Split oversized reference material into references/ when one reference.md would exceed about 1,000 tokens or more than 3 logical topics.
12. If reference.md is the chosen companion file and becomes too broad, keep it focused or reduce it to a short overview while moving detailed topics into references/.
13. Add examples.md when the skill is high complexity or the concise examples are not enough for about 80% of routine usage.
14. When the skill is higher-risk or easy to misuse, add stronger validation signals in SKILL.md or local files, such as explicit verification guidance, red flags, rationalizations, or a checklist.
15. Do not force heavyweight validation onto a simple low-risk skill just because another skill needed it.
16. If you add optional files or subfolders, declare each role in Local references.
17. When the draft is review-ready, tell the user: This skill is review-ready. Please hand it to agent-skill-reviewer for review.

Examples

• Positive: Draft release-note-shortener with a clear trigger, brief positive and negative examples in SKILL.md, and stronger verification guidance only if the skill can materially affect release output.
• Negative: Draft a skill when the responsibility is still vague, mixes creation, review, and publishing, or forces release-grade validation onto a simple naming rule.

Outputs

• a new .github/skills/<skill-name>/ folder, using lowercase kebab-case
• SKILL.md with concise positive and negative examples
• examples.md for high-complexity skills, or reference.md for local detail
• references/ when local reference detail must be split by topic
• optional local additions with explicit roles
• validation weight aligned to the skill's risk and complexity
• a review-ready skill draft and explicit handoff message

Validation

Required Checks

• target skill folder does not exist before creation begins
• responsibility, trigger, and boundaries are unambiguous before drafting
• SKILL.md contains name, description, complexity, Purpose, Trigger / When to use, Inputs, Process, Examples, Outputs, Boundaries, and Local references
• at least one concise positive example and one concise incorrect example are present in SKILL.md

Quality Checks (best effort)

• risk_profile tags match the actual skill behavior
• examples.md is present when the skill is high complexity, involves branching or code modification, or the SKILL.md examples cover less than ~80% of routine usage
• stronger validation signals are present when risk warrants it
• Validation and Failure Handling sections are present for high complexity output

On Soft Fail

• mark status as INCOMPLETE
• continue with best-effort draft
• list missing information explicitly in the handoff message

Failure Handling

Missing Context

• mark output as INCOMPLETE
• list required additional inputs explicitly before proceeding

Ambiguous Requirement

• if blocking: stop and ask the user before drafting
• if non-blocking: proceed with stated assumptions, list them in the handoff message

Execution Limitation

• state the limitation explicitly in output
• do not fabricate content to fill gaps

Workflow State Contract

When participating in a multi-agent skill creation workflow, include:

• current_step:
• next_step:
• status: IN_PROGRESS | COMPLETE | INCOMPLETE | BLOCKED

Omit this section when the skill is created outside a multi-agent handoff flow.

Boundaries

• Do not draft when the responsibility, trigger, or boundaries are still ambiguous.
• Do not create multi-purpose skills.
• Do not leave oversized multi-topic reference material in one undifferentiated file.
• Do not rely on hidden context outside the skill folder.
• Do not expand a creator/reviewer-first topic into downstream regular-skill rollout; defer those changes to a later topic plan.
• Do not add heavyweight validation to a lightweight skill unless the risk truly warrants it.
• Do not claim approved or stable.
• Do not approve your own output.

Local references

• blueprint.md: starter shape and section skeleton for new skills
• folder-contract.md: required core and optional role rules
• examples.md: creation patterns, complexity triggers, and review-ready examples