Skip to content
Back to search

skillstead

Practical, portable skills for agentic coding workflows — clearer diagrams, checked claims, safer GitHub releases, natural EN↔KO writing.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/kyungseo--skillstead/badge.svg)](/listings/kyungseo--skillstead)

HTML

<a href="/listings/kyungseo--skillstead"><img src="/api/listings/kyungseo--skillstead/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 244 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # Skillstead
  2. **English** · [한국어](./README.ko.md)
  3. Practical, portable skills for agentic work — create clearer artifacts, check public claims, guide safer GitHub
  4. releases, turn rough or translated text into natural, precise writing, and interpret portraits.
  5. > [!TIP]
  6. > **Skillstead = skill + homestead.** A small, durable place for skills that coding agents can carry into
  7. > real repositories. Each public support claim is tied to examples and runtime evidence.
  8. ## Start here
  9. 1. [Choose one skill](#choose-a-skill) for the task in front of you. Each skill works independently.
  10. 2. Follow the [installation guide](./docs/INSTALL.md) to copy that skill's complete folder at a pinned version.
  11. 3. Open the skill's README, copy a request that is close to your task, and adapt it to your material.
  12. 4. Check the skill's stated limits before relying on the result in a release or other consequential workflow.
  13. Looking for a particular guide? The [documentation index](./docs/README.md) separates user guides from
  14. maintainer references.
  15. ## Choose a skill
  16. | Skill | Best for | Version | Runtime support | Maturity |
  17. | --- | --- | --- | --- | --- |
  18. | [`svg-infographic`](./skills/svg-infographic) | Turning architecture notes, process flows, comparisons, and technical concepts into editable SVG + verified 2× PNG | `0.12.0` | Supported: Claude Code + Codex | Stable |
  19. | [`docs-claim-check`](./skills/docs-claim-check) | Checking whether public documentation claims are supported by supplied evidence | `0.11.0` | Claude Code | Stable |
  20. | [`github-release-guide`](./skills/github-release-guide) | Guiding a private repository's first public transition and every later version release, with separate approval before each change | `0.9.1` | Supported: Claude Code + Codex | Stable |
  21. | [`writing-quality-editor`](./skills/writing-quality-editor) | Composing and revising user-facing text, plus natural English↔Korean adaptation, without inventing or changing facts, intent, voice, or operational constraints | `0.19.0` | Supported: Claude Code + Codex | Stable |
  22. | [`street-portrait-artist`](./skills/street-portrait-artist) | Creating a caricature, pen-and-watercolor portrait, or drawn paper-cut figure from supplied photos | `0.2.0` | Supported: ChatGPT + Codex | Experimental |
  23. Each skill is self-contained and can be installed independently. You do not need to install the entire
  24. catalog—copy only the complete folder for the skill you want to use. See
  25. [`docs/INSTALL.md`](./docs/INSTALL.md) for global/project paths, pinned tags, clean updates, Windows commands,
  26. and the per-skill runtime matrix.
  27. The `Version` column above is per skill, not a catalog version. See
  28. [`docs/VERSIONING.md`](./docs/VERSIONING.md) for what it means and how it changes.
  29. For more examples—including natural requests, the `WQE` shorthand, and requests that involve more than one
  30. skill—see the repository-only
  31. [`intent and invocation contract`](./examples/intent-invocation-contract).
  32. GitHub's **Latest** badge identifies the most recently published individual skill release. It does not represent
  33. a catalog version.
  34. ## Highlights
  35. ### Explore the SVG gallery
  36. [![Six outputs from svg-infographic: a cloud topology, a branching swimlane, a decision matrix, nested trust rings, a before-and-after migration, and a sketch-treatment system map](./gallery/contact-sheet.en.png)](https://kyungseo.github.io/skillstead/gallery/)
  37. `svg-infographic` covers more than conventional architecture boxes. The six above were chosen for how
  38. differently they are shaped; they clear the lint, layout and typography gates, and they predate the TypePack
  39. receipts, so no receipt is claimed for them. The gallery behind the image carries the nine TypePacks with their
  40. receipts alongside these. The contact sheet intentionally keeps an English evidence frame around the locale-specific
  41. artifacts; the live gallery localizes the full page. [Open the live gallery](https://kyungseo.github.io/skillstead/gallery/), or
  42. [browse the legacy examples and the verified TypePack catalog](./examples/svg-infographic).
  43. ### Use one skill—or combine them across a release
  44. [![Independent Skillstead skills connected across a practical project and release path](./examples/catalog-overview.en.png)](./examples/catalog-overview.en.svg)
  45. Each skill installs and works independently. When a project needs a wider path, use
  46. `writing-quality-editor` for clear prose, `svg-infographic` for visual explanation, `docs-claim-check` for
  47. evidence-bounded public claims, and `github-release-guide` for approval-gated release decisions. This is not a
  48. required pipeline: start with the skill you need, skip the others, and recheck an earlier artifact when it changes.
  49. `street-portrait-artist` is a separate creative workflow for character portraits from supplied references; it does
  50. not need to participate in that release path.
  51. ## Skill details
  52. ### svg-infographic
  53. Technical diagrams often begin as prose and end as hard-to-edit screenshots. `svg-infographic` computes a
  54. layout before drawing, authors structured SVG, checks the source, then exports a dimension-verified 2× PNG.
  55. Use it for architecture and cloud topology, process or approval flows, before/after migrations, roadmaps,
  56. layer models, qualitative matrices, and Korean/CJK-ready technical one-pagers.
  57. - Friendly guide: [`svg-infographic` README](./skills/svg-infographic/README.md)
  58. - Live gallery: [featured outputs and the verified TypePack catalog](https://kyungseo.github.io/skillstead/gallery/)
  59. - Examples: [legacy examples and the verified TypePack catalog](./examples/svg-infographic)
  60. - Name the skill: `Use svg-infographic to turn this migration plan into an editable technical infographic.`
  61. - Or ask naturally: `Turn this migration plan into an editable technical SVG and verified 2× PNG. Show the output path before creating files.`
  62. ### docs-claim-check
  63. Release-facing docs can sound certain even when their evidence is partial or stale. `docs-claim-check` splits
  64. objective statements into atomic claims and labels each one `verified`, `unsupported`, `stale-suspected`, or
  65. `needs-human` within an explicit reviewed-input scope.
  66. Use it before publishing a README, install guide, release note, or announcement. It is advisory only: the
  67. contract runs no commands during assessment and does not generate fixes, code review, or security verdicts.
  68. - Friendly guide: [`docs-claim-check` README](./skills/docs-claim-check/README.md)
  69. - Validation material: [synthetic AcmeTask fixture and worked assessment](./examples/docs-claim-check)
  70. - Name the skill: `Use docs-claim-check to assess these release-note claims against the supplied tag and CI evidence.`
  71. - Or ask naturally: `Check whether these README claims are supported by the evidence below. Report findings only; do not rewrite the document.`
  72. ### github-release-guide
  73. GitHub releases combine documentation work with changes to visibility, branches, tags, settings, and GitHub
  74. Releases that can be difficult to undo. `github-release-guide` first checks readiness without changing the
  75. repository. It then shows each proposed change, checks the current state again, and asks for approval in two
  76. distinct scopes — running a repository command, and changing the repository — before verifying the result and
  77. moving on.
  78. V1 can be used at two points: when an existing private github.com repository becomes public for the first time,
  79. and whenever that public repository publishes a new version afterward. It does not bootstrap repositories,
  80. publish packages, sign binaries, deploy cloud services, claim a security audit, force-push, or rewrite history.
  81. **First, pick the mode.** Assess inspects without changing anything; Guided starts only after Assess is
  82. complete, release-critical blockers are resolved, and you explicitly choose to switch.
  83. [![Assess checks the repository without changing it and returns Ready, Needs attention or Blocked; Guided starts only when all three entry conditions are met; either mode works on the first-public or version-release profile](./examples/github-release-guide/mode-profile-map/mode-profile-map.en.png)](./examples/github-release-guide/mode-profile-map/mode-profile-map.en.svg)
  84. **Then, in Guided, one change at a time.** Approval is asked in two distinct scopes: running a command is
  85. approved separately from changing the repository, and neither implies the other.
  86. [![In Guided each change is previewed and rechecked, then takes execution approval only when a command must run, always takes mutation approval, applies only what was previewed, and verifies the observed result before continuing or stopping](./examples/github-release-guide/approval-safety-loop/approval-safety-loop.en.png)](./examples/github-release-guide/approval-safety-loop/approval-safety-loop.en.svg)
  87. - Friendly guide: [`github-release-guide` README](./skills/github-release-guide/README.md)
  88. - Validation material and diagrams: [synthetic scenarios, answer key, and worked outputs](./examples/github-release-guide)
  89. - Natural readiness check: `Check whether this public GitHub repository is ready for its next version release. Inspect only and do not change it.`
  90. - Assess example: `Use github-release-guide in Assess mode for this public repository's upcoming version release.`
  91. - Guided example: `Use github-release-guide in Guided mode to prepare this private repository for first publication. Start with Assess, then show only the first proposed change. Do not change the repository until I approve that exact step.`
  92. - Safety boundary: immediately before a repository becomes public, the guide explains what cannot be undone and
  93. asks the user to approve that visibility change separately. The release decision remains with the user.
  94. ### writing-quality-editor
  95. `writing-quality-editor` writes new text from notes and evidence or makes existing prose easier to understand.
  96. Its editing rules require preserving facts, intent, author voice, commands, conditions, limitations, risks, and
  97. next actions. Those rules do not make every result natural or accurate; important writing still needs a final review.
  98. For wording and naturalness requests, it now starts with the smallest complete span that has a concrete reader
  99. problem and leaves the surrounding text unchanged. Ambiguous discretion, approval, or notification wording stays unchanged
  100. for human judgment while other safe edits continue.
  101. Its `Adapt` mode rewrites between English and Korean for the target-language reader instead of copying the source
  102. sentence structure. It may change information order, sentence rhythm, idioms, and explanation density, but it
  103. does not invent claims or hide ambiguity. AI-detector gaming and provenance concealment are explicit non-goals.
  104. - Friendly guide: [`writing-quality-editor` README](./skills/writing-quality-editor/README.md)
  105. - Validation material: [36 scenarios and a separate answer key](./examples/writing-quality-editor)
  106. - Name the skill: `Use writing-quality-editor to make the document below read naturally. Preserve its core facts, conditions, and requirements.`
  107. - Use the shorthand: `Use WQE to review this onboarding guide. Identify problems, but do not revise it yet.`
  108. - Or ask naturally: `Review this README. Do not revise the prose yet.` · `Write a new README using only information supported by the material below.` · `Rewrite this English release note so it reads naturally to Korean readers. Preserve its meaning and conditions.`
  109. - Specify a mode only when needed: `Use writing-quality-editor in Assess mode to review this release note. Do not draft revisions.`
  110. The versions listed above include a conditional WQE paragraph-flow check and portrait consistency, Editorial
  111. Watercolor, and Paper-Cut Illustration. See the respective skill guides for development evidence and limitations.
  112. ### street-portrait-artist
  113. `street-portrait-artist` studies face shape, feature spacing, expression, and the main distinguishing feature before
  114. drawing. It uses those observations for a kind, near-monochrome `Street Caricature`, a pen-and-watercolor
  115. `Romance Watercolor` portrait, or a flat-color `Paper-Cut Illustration` in the supplied scene. All modes are
  116. designed to retain the same person’s recognizable features.
  117. Provide one or more clear portrait references. Supplied portraits, analysis, and outputs remain task-scoped unless the
  118. user separately asks to publish or retain them.
  119. - Friendly guide: [`street-portrait-artist` README](./skills/street-portrait-artist/README.md)
  120. - Public-safe gallery: [two synthetic Twin Portrait examples](./examples/street-portrait-artist)
  121. - Name the skill: `Use street-portrait-artist to make a kind Street Caricature from my attached portrait.`
  122. - Or ask naturally: `Turn this portrait into a delicate pen-and-watercolor character portrait. Keep the jaw, hairline, and expression recognizable.`
  123. ## Playbooks (maintainer reference)
  124. For changes to Skillstead itself, start with the [maintainer workflow](./docs/MAINTAINER-WORKFLOW.md):
  125. document alignment, root changelog, adopted personal-skill synchronization with commit and push, version
  126. announcements, blog follow-up, and optional social drafts.
  127. [`playbooks/public-release`](./playbooks/public-release/README.md) contains the canonical public-release playbook:
  128. generic checklists and templates for taking a private repository public and verifying it afterward. These are
  129. reference documents for maintainers — not installable skills, and installing any skill never requires them.
  130. The `github-release-guide` skill mirrors the playbook's rules inside its own self-contained package. English is
  131. canonical, and each document has a Korean mirror with the `.ko.md` suffix. Update both languages in the same pull
  132. request whenever the meaning changes.
  133. [`playbooks/skill-development`](./playbooks/skill-development/README.md) defines the repository standard for
  134. designing, validating, independently reviewing, releasing, and retiring a skill. Its package template is
  135. deliberately non-installable until the reserved `sample-skill` identity is replaced. The tracked retirement and
  136. major-transition evidence enforced by the release gates is specified in
  137. [`docs/VALIDATION.md`](./docs/VALIDATION.md).
  138. ## Quality and evidence bar
  139. Every public skill must have:
  140. - a clear description of what it does and does not do,
  141. - synthetic, non-client validation material,
  142. - runtime support and maturity labels limited to what was actually tested,
  143. - public paths free of credentials, private provenance, and host-specific data,
  144. - and a repeatable validation path appropriate to its output.
  145. Runtime support is per skill, not catalog-wide.
  146. `svg-infographic` has passed the same three frozen, fresh-context briefs on Claude Code and Codex. The checks
  147. covered project-local discovery, source lint before rendering, exact 2× browser export, Korean/CJK text,
  148. containment, connector semantics, and fail-closed recovery when browser launch crossed a sandbox boundary.
  149. Codex evidence includes macOS Codex CLI and a fresh Codex App task in a Windows 11 ARM64 VM. It is
  150. `Supported` for Claude Code and Codex within that recorded evidence scope.
  151. `github-release-guide` has passed clean Claude Code/Codex material-parity checks, the disposable first-public
  152. live E2E, pinned `v0.5.0` project installation and discovery smoke, and the final strict claim audit. It is
  153. `Supported` for Claude Code and Codex within that recorded evidence scope.
  154. `writing-quality-editor` is Stable as of `0.18.0`, following bounded validation and explicit owner acceptance.
  155. Its existing `Supported` labels for Claude Code and Codex remain within the recorded evidence scope. The fixed
  156. candidate was checked through eight regression outputs and forty final outputs on the two CLIs, including
  157. actual package reads and same-conversation follow-ups. These checks do not establish universal discovery,
  158. perfect preservation, general superiority, or new runtime support. Historical failures and remaining wording
  159. limitations are retained in the [WQE guide](./skills/writing-quality-editor/README.md).
  160. `street-portrait-artist` has two public-safe synthetic Twin Portrait cases that document its intended visual direction
  161. and privacy boundary. Fresh installations of the published `0.1.0` package were then discovered and invoked in ChatGPT and
  162. Codex with synthetic portrait input, reference-image generation, fail-visible size fallback, and output delivery. Both
  163. runtimes are `Supported` within that recorded scope. Maturity remains Experimental because likeness and the refined
  164. visual behavior have not passed a broad cross-subject regression set.
  165. ## Current limitations
  166. - `svg-infographic` browser rendering is verified on macOS and in the recorded Windows 11 ARM64 VM.
  167. This does not claim every Windows machine or filesystem; Linux rendering remains documented but unverified.
  168. - `docs-claim-check` is advisory and evidence-bound; it does not execute verification commands.
  169. - `github-release-guide` v1 is github.com-only. It covers the one-time private-to-public transition and each
  170. version release after the repository is public.
  171. - For `github-release-guide`, a clean assessment only means that no issue was found within the inspected scope;
  172. it does not prove that the repository being assessed has no secrets, private information, or security risks.
  173. - `writing-quality-editor` is designed as locale-neutral, but its initial localization fixtures cover only
  174. English↔Korean (`ko-KR` for Korean output). Other locale pairs are not yet claimed as behaviorally validated.
  175. - `street-portrait-artist` likeness and image quality remain non-deterministic, exact pixel export depends on the
  176. active image surface, and final artwork requires human review. The recorded ChatGPT and Codex support does not
  177. promise exact dimensions or consistent visual quality across different image-generation runs.
  178. - Agent output is non-deterministic. `writing-quality-editor` defaults to local-first edits and holds ambiguous
  179. wording for review, but it cannot guarantee that every run avoids preference-driven changes outside the marked
  180. spans. Review the final delta before publishing or relying on an important document.
  181. ## License
  182. [Apache-2.0](./LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.