Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/Abd0r/porcupineai --skill project-hygiene명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Write code that does not look like AI output. Reject low-evidence and low-signal TypeScript patterns (chained assertions, unknown/object widening, Reflect, runtime typeof, module mocks) in favor of typed, boundary-checked code.
Control a Home Assistant smart home through its MCP server. Query sensor and entity states read-only, then act on devices (lights, switches, climate) via call_service with explicit user approval. Use for "turn off the living room lights", "what is the thermostat set to", "set bedroom temp", or any Home Assistant entity/device task.
Produce clear, accessible diagrams and charts (architecture, flow, sequence, ER, state, org, Gantt, timeline) as Mermaid or polished self-contained SVG. Covers picking the right diagram type, a consistent visual system, and accessibility. Use whenever you need to explain a system, a flow, data, or a process in a README, doc, benchmark report, plan, or launch graphic.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | project-hygiene |
| description | Create a durable, evidence-backed project workspace. |
| stack | meta |
Use this skill to give a substantial project one canonical home for its purpose, verified state, decisions, evidence, and next work. The canonical location is Project/<project-name>/ relative to the current repository or explicitly chosen workspace.
The goal is continuity, not paperwork. Keep documents short, factual, and current. Do not create empty ceremony files for a trivial task.
Do not use this for a one-file fix, a disposable experiment, or a task whose existing repository documentation is already the canonical project home.
Create this minimum structure:
Project/
└── <project-name>/
├── README.md
└── STATUS.md
Add these files only when their contents exist and will remain useful:
├── PLAN.md # Multi-step implementation or research plan.
├── TASKS.md # Small actionable checklist across sessions.
├── DECISIONS.md # Non-obvious choices, alternatives, and rationale.
├── EVIDENCE.md # Measurements, sources, experiments, and validation.
└── CHANGELOG.md # Meaningful project milestones, not every edit.
Use a lowercase, hyphenated directory name such as Project/sparse-attention/. Preserve the human-readable project title inside README.md.
read and grep to inspect existing README files, plans, experiment logs, task notes, and source entry points.Project/<project-name>/ unless the user explicitly selects another root.Completion criterion: the project has one unambiguous directory and no competing status document is introduced.
Create README.md with:
# <Project Title>
## Objective
One measurable purpose.
## Scope
- In scope:
- Out of scope:
## Current approach
The current technical or research direction.
## Success criteria
- Observable gate 1
- Observable gate 2
## Workspace map
- `path/`: why it matters
## Related material
- Links to repository paths, papers, issues, or external sources.
Create STATUS.md with:
# Status: <Project Title>
## State
`planning` | `active` | `blocked` | `paused` | `complete`
## Last verified
- Date or commit:
- What was executed or inspected:
- Result:
## Current position
A short factual snapshot. No aspirational claims.
## Blockers and risks
- Blocker or `None`.
## Next verified action
One concrete action, including the command, test, or artifact that will prove it.
Completion criterion: README.md explains why the project exists, and STATUS.md gives another session enough context to take the next action without guessing.
PLAN.md when the work has multiple dependent steps. Each step names its files, verification command, and completion gate.TASKS.md for small, movable tasks. Keep only current work there; completed tasks belong in CHANGELOG.md only if they changed a milestone.DECISIONS.md when a choice would otherwise be rediscovered. Record date, decision, alternatives, evidence, and consequence.EVIDENCE.md for measurements, reproducible experiment results, source citations, dataset provenance, or evaluation output. Separate observations from interpretations.CHANGELOG.md only for milestones such as a validated prototype, completed dataset version, accepted design, or released artifact.Completion criterion: every supporting document answers a real future question. Delete or avoid placeholder documents that do not.
After a meaningful project action:
STATUS.md with the actual result, blocker, and one next verified action.WORKFLOW.md when a procedure changes (a new flag, a different gate, a new recovery step) so the recorded workflow never silently diverges from reality.Do not put credentials, tokens, private session transcripts, speculative benchmark numbers, or copied tool output dumps in project documents.
Completion criterion: STATUS.md never claims a milestone without its evidence being available in the repository, test output, or EVIDENCE.md.
Project/ for every minor code edit.STATUS.md as a marketing update instead of a handoff record.Before finishing a project-hygiene pass, verify:
Project/<project-name>/README.md and STATUS.md exist.planning-and-task-breakdown for implementation plans with file-level verification.source-driven-development before documenting behavior that depends on existing code.test-loop when a project milestone requires a repeatable test or build gate.memory-hygiene for durable user preferences rather than project-progress notes.