ワンクリックで
file-writing
Governs how the agent writes files. Enforces chunked, sequential writes to prevent silent truncation from buffer/context limits.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Governs how the agent writes files. Enforces chunked, sequential writes to prevent silent truncation from buffer/context limits.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Guides you through learning programming and software engineering topics using a Socratic ladder. Never gives the exact answer or a matching example. Activates on explicit requests like "help me learn X", "explain Y", "I'm trying to understand Z". Gives links to online sources.
Distills the current session into a structured specification document (SPEC.md). Captures design decisions, canonical terms, edge cases, known weak points, open questions, and ADRs. Activate with "spec this", "create spec", "write the spec", "turn this into a spec", "specify this", or after a design-interview or premortem-this session concludes.
Runs a structured, relentless design interview to reach a shared understanding of a plan, feature, or system. Walks the design tree branch by branch, resolving decision dependencies one at a time. Activate when the user says "interview me", "let's design", "walk me through", "review my plan", or shares a spec/plan and asks for feedback.
Stress-tests any plan, project, or decision by imagining it has already failed and working backward to discover why. Surfaces hidden assumptions, weak decisions, and missing risks before resources are committed. Activate when the user says "premortem this", "stress test this plan", "what could go wrong", "poke holes in this", "what am I missing", or after a design-interview session to stress-test a clarified plan.
Governs the agent's internal reasoning process and communication style. Enforces hypothesis-driven, test-first, iterate-fast problem solving. Eliminates silent internal loops, premature abstraction, and invisible approach changes.
Proactively guides and enforces code quality practices: eliminating magic numbers and strings, constructor-based dependency injection, and thorough documentation. Flags violations as they are noticed, plans safe incremental refactors, and integrates with tdd-workflow, ddd, long-term-memory, and pair-mode when active.
| name | file-writing |
| description | Governs how the agent writes files. Enforces chunked, sequential writes to prevent silent truncation from buffer/context limits. |
This skill applies to every file write operation — new files and modifications alike. Never write an entire file in a single operation.
No more than 50 lines per write operation.
Even if a file appears short enough to write in one pass — if it exceeds 50 lines, split it into at least 2 chunks. No exceptions.
Write from the beginning. Append each subsequent chunk in order. Never write chunks out of sequence.
Briefly state:
Example:
"Wrote lines 1–50: file header and imports. Next: core logic, lines 51–100."
After the final chunk, verify the file is complete:
long-term-memoryWhen writing or updating MEMORY.md, apply chunked writes. Prefer
append-style operations over full-file rewrites where possible, per the
memory skill's own writing strategy.
This skill applies universally — docs, configs, code, skill files, and any other artifact the agent produces.