app-feature-craft

Use when implementing or improving a user-facing app workflow. Guides product-grade UX, frontend, backend/API behavior, AI/provider or long-running task states, tests, and end-to-end verification from rough requirements to a shippable feature.

Jump to install

Repository: sidiangongyuan/codex-skills-library
Last source activity: July 12, 2026 at 11:41
Detected SKILL.md language: English
Stars: 5
Forks: 0

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/sidiangongyuan/codex-skills-library --skill app-feature-craft

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer

5 files

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

paper-refinement-skills

Use when refining research-paper prose or section logic, including abstracts, introductions, related work, methods, captions, conclusions, and rebuttals. Improves clarity, concision, transitions, terminology, notation, and venue-appropriate style without inventing evidence.

2026-08-12

experiment-planner

Use when exploring a deep-learning or computer-science research idea before implementation or paper writing. Converts claims into pilot-first experiment matrices covering ablations, diagnostics, robustness, failure analysis, resource coordination, and paper-story viability.

2026-08-10

feishu-paper-reading

Search, screen, deeply read, synthesize, and publish recent research papers for a user-specified topic, window, count, quality/attention balance, and language. Use for evidence-anchored literature digests, cross-paper comparisons, original-language excerpts, Chinese explanations, guided Feishu setup, or Feishu research reports.

2026-07-29

View all from this repository

App Feature Craft

Use this skill to turn an app request into a feature users can actually rely on. Default to shipping the smallest complete behavior, not a technical stub.

Workflow

Ground in the product before designing.

Inspect routes, screens, state stores, API clients, backend handlers, data models, tests, packaging scripts, and existing copy.
Resolve discoverable facts from the repo before asking the user.
If visual behavior matters, inspect screenshots or run the app.

Define the user-facing outcome.

State what the user can do after the change.
Identify the audience, success criteria, in/out of scope, data safety constraints, and where the behavior must appear.
Separate optional convenience from behavior required for a complete feature.

Design the whole path.

Cover UI states: empty, loading, success, failure, stale data, first-run, and already-configured.
Keep controls near the user's task context. Avoid bolted-on panels, rigid form-heavy surfaces, or result views that disappear behind an ambiguous "back" action.
Prefer reading-first UI with focused edit affordances for dense knowledge or writing surfaces.
Cover data/API behavior: persistence, validation, compatibility, migration, idempotency, and failure modes.
Cover lifecycle behavior for desktop apps: startup, restart, sidecars, installer state, and local paths when relevant.
For long-running AI or provider work, design background progress, reopening/recovery, cancellation semantics, and honest stage labels before implementing the request.

Implement narrowly.

Follow existing local patterns instead of inventing a new framework.
Keep user data safe by default; never delete, overwrite, or migrate destructively unless the user explicitly asked for it.
Do not expose raw backend/browser errors in public UI.
Do not report a fake success for external integrations. Mocked tests prove code paths, not real provider availability.
Run only the user-selected AI/profile/configuration set. Do not silently add a default provider after the user made an explicit choice.
Keep separate provider profiles and locally saved keys isolated by owner; never let one profile's credential overwrite another profile's source.
Make AI-generated drafts, candidate imports, and writeback actions review-first unless the user explicitly requested automatic mutation.

Verify from the user's seat.

Run focused unit/API tests for changed behavior.
Run frontend tests for state, copy, and interactions.
Use browser or app screenshots for UI/visual changes.
For provider, installer, or filesystem features, run at least one real local smoke test when credentials and environment are available.
For background jobs, verify close/reopen, status reconnect, cancellation, completion, and failure states without resending the provider request.

Report the result.

Summarize user-visible changes, important files, tests run, and any residual risk.
Say clearly when a test was mocked, skipped, or only verified locally.

Reference Use

Read references/app-quality-principles.md before implementing complex app features, AI/provider integrations, long-running jobs, first-run experiences, data-path behavior, or desktop lifecycle work.

Read references/living-to-tell-casebook.md when the task resembles one of the captured cases: startup feedback, Not Found cleanup, AI provider transport, data location safety, installation/version mismatch, async selection races, or motif/excerpt deduplication, AI result recovery, profile key isolation, or release asset verification.

Ask Only When Needed

Ask the user only when the decision is product intent and cannot be discovered from the codebase. Prefer one concrete tradeoff question over a broad design question.