skills

A personal collection of agent Skills authored or maintained by mtk177a, plus two designated unmodified Gist mirrors.

The Skills follow the Agent Skills specification.
Whether a Skill works also depends on the client, and on what the agent can access and do.
Check the runtime requirements before installation.

No warranty: The owner maintains this personal repository on a best-effort basis.
The owner can change or remove Skills without notice.
Use at your own risk.

Install

There are two installation methods, and both install the same Skills from skills/:

Plugin ZIP: the only method for ChatGPT.
A plugin uploaded to ChatGPT is also available in Codex.
APM: for Claude Code, Codex, Cursor, and the other clients that APM supports.

To choose a method and install, see Install and update the Skills.

Skills

Choose a Skill by task.
Then read the linked Skill to see when to use it and what it produces.
To use several Skills on one piece of work, see Choose the next Skill.

Clarify and choose an approach

Task	Skill
Clarify an ambiguous request	`clarify-request`
Define what ambiguous terms refer to	`define-referents`
Compare different problem frames and solutions	`explore-decision-space`
Choose an AI execution environment for the task	`choose-ai-execution-setup`
Preserve understanding and decision ownership while using AI	`calibrate-learning-support`

Design, implement, and investigate

Task	Skill
Plan scope, risks, and verification before implementation	`design-changes`
Implement an authorized change and verify it	`implement-changes`
Investigate unexpected behavior and failures	`investigate-failure`
Assess safety, recovery, and authorization before a consequential action	`assess-risky-change-readiness`
Stop repeated failed attempts and choose the next diagnostic step	`break-failure-loop`

Review and communicate changes

Task	Skill
Review a new diff, or re-review the whole diff when explicitly requested	`review-changes`
Assess findings and decide how to respond	`triage-review-feedback`
Recheck identified fixes against their original findings	`validate-fix`
Draft PR comments from supplied review decisions	`draft-review-comments`
Draft an Issue without posting it	`draft-issue`
Plan commit units and draft their messages	`draft-commit`
Summarize a diff or commit range for its readers	`summarize-changes`
Record evidence for resuming work in another session	`record-session-handoff`

Research and write

Task	Skill
Research the Web with traceable evidence	`research-web-safely`
Select and organize information for the reader	`structure-what-you-write`
Explain an answer and the reasoning needed to understand it	`make-it-click`
Write and revise natural Japanese without changing meaning	`write-natural-japanese`
Apply Japanese technical-writing rules	`japanese-tech-writing`
Design pacing in explanatory Japanese prose	`cognitive-rhythm-writing`
Have an authorized independent subagent revise a document	`revise-docs-fresh-eyes`

Maintain Skills and agent instructions

Task	Skill
Diagnose gaps between agent guidance and behavior	`audit-agent-guidance`
Design a Skill’s responsibility and boundaries	`design-skill`
Design instruction files and client loading relationships	`design-agent-instructions`

japanese-tech-writing and cognitive-rhythm-writing are unmodified copies of upstream Gists.
Their source revisions, licenses, and synchronization records are in each Skill's UPSTREAM.md and THIRD_PARTY_NOTICES.md.

Maintain

Task	Guide
Verify changes and prepare Issues, commits, and pull requests	Contributing
Author a Skill	Skill authoring guide
Build the plugin ZIP	Build the plugin ZIP
Review third-party material and executable capabilities	Skill security review
Choose the language and maintain Japanese reference translations	Languages and Japanese reference translations
Check the terms used for repository concepts	Glossary

External contributions are not normally expected.

License

Original code and content are licensed under the MIT License.
Third-party material retains the terms listed in THIRD_PARTY_NOTICES.md.