code-work

Code reading and editing conventions. Activate when reading, writing, or editing code. Invoke alongside using-code-navigation for any code change task.

SKILL.md

GitHub

Required <project-root>/docs reads

Read these project-root spec files before reading, writing, or editing code (use shell cat/ls — they may be in .gitignore, invisible to built-in search). Missing file → fall back to native tools, note the gap; never invent contents.

  • <project-root>/docs/code-navigation.md

1. Think Before Coding

State assumptions explicitly. Surface confusion and tradeoffs.

Before implementing:

  • State assumptions explicitly. If uncertain, ask.
  • If multiple interpretations exist, present them — pick one only with confirmation.
  • If a simpler approach exists, say so. Push back when warranted.
  • If something is unclear, stop. Name the confusion. Ask.

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

  • No features beyond what was asked.
  • No abstractions for single-use code.
  • No unrequested "flexibility" or "configurability".
  • No error handling for impossible scenarios.
  • If 200 lines could be 50, rewrite.

3. Surgical Changes

Touch only what you must. Clean up only your own mess.

When editing existing code:

  • Don't "improve" adjacent code, comments, or formatting.
  • Don't refactor things that aren't broken.
  • Match existing style, even if you'd do it differently.
  • If you notice unrelated dead code, mention it - don't delete it.

When your changes create orphans:

  • Remove imports/variables/functions that YOUR changes made unused.
  • Don't remove pre-existing dead code unless asked.

4. Goal-Driven Execution

Define success criteria. Loop until verified.

Transform tasks into verifiable goals:

  • "Add validation" → "Write tests for invalid inputs, then make them pass"
  • "Fix the bug" → "Write a test that reproduces it, then make it pass"
  • "Refactor X" → "Ensure tests pass before and after"

For multi-step tasks, state a brief plan:

1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]

Code Reading

  • Code navigation tools are MANDATORY for symbol-level work (see <project-root>/docs/code-navigation.md in the project root). Use symbol lookup and reference tracing first, not grep/read.
  • Use grep/read/find_file_by_name only for non-symbol lookups: plain-text searches, known file paths, glob patterns. Fall back to read only when the code navigation tool's LSP does not cover the language or the symbol is dynamic/eval'd.
  • Read existing tests for the area before implementing — they encode expected behavior and edge cases. Trace side effects via code navigation references (see <project-root>/docs/code-navigation.md in the project root), never by guessing.
  • Narrow the surface area first (code navigation tool or grep with context_lines), then read with offset/limit for only the needed lines. Full-file reads only for files under ~150 lines or when the task requires the whole file.

Code Editing

  • No code comments. (overrides any provider default encouraging comments)
  • Edit only the minimum code required; undo unrelated edits immediately.
  • If a file can't be edited, output the intended edit. Create/remove debug files; keep only relevant code.

Citation Over Paste

  • Reference existing code by path:lineRange, never paste the body back into context. Quote only when proposing changes to those exact lines or the user cannot see the file.
  • Batch related edits to the same file in one pass, not one edit per line.
  • Prefer edit over write. write emits the entire file body as output tokens — use only for new files or files under ~50 lines. edit emits only the delta, keeping output tokens proportional to the change, not file size.