docs(ai): git apply on diff (#377)

https://claude.ai/code/session_013WwSdSkpqSyDZ7WhJWn2f1

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: 3 people

.github/copilot-instructions.md

@@ -19,3 +19,4 @@ Read and follow the CONTRIBUTING.md file in this repository for all code style c
 - If you change CLI arguments, help text, or anything that affects command-line output, run `./generate-completions.sh` to regenerate the shell completion files, help text files, and `USAGE.md`. **Do not attempt to regenerate these files manually** — always use the script.
 - Run `FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh` to validate changes. If a test fails with a hint about `TEST_SKIP`, follow the hint and rerun with the suggested variable. If a sync test fails, read its error message carefully and run the exact command it tells you to run.
 - **ALWAYS run the full test suite** (`FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh`) before committing, regardless of how trivial the change seems — this includes documentation-only changes, comment edits, config changes, and refactors. The test suite checks formatting, linting, building, tests, and docs across multiple feature combinations; any type of change can break any of these checks.
+- When the user provides a diff and you need to update the files, don't manually interpret each hunk (that'd be slow); apply it with `git apply` instead. If the user provides a diff for context or discussion rather than as a change to apply, respond accordingly instead.

AGENTS.md

@@ -19,3 +19,4 @@ Read and follow the CONTRIBUTING.md file in this repository for all code style c
 - If you change CLI arguments, help text, or anything that affects command-line output, run `./generate-completions.sh` to regenerate the shell completion files, help text files, and `USAGE.md`. **Do not attempt to regenerate these files manually** — always use the script.
 - Run `FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh` to validate changes. If a test fails with a hint about `TEST_SKIP`, follow the hint and rerun with the suggested variable. If a sync test fails, read its error message carefully and run the exact command it tells you to run.
 - **ALWAYS run the full test suite** (`FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh`) before committing, regardless of how trivial the change seems — this includes documentation-only changes, comment edits, config changes, and refactors. The test suite checks formatting, linting, building, tests, and docs across multiple feature combinations; any type of change can break any of these checks.
+- When the user provides a diff and you need to update the files, don't manually interpret each hunk (that'd be slow); apply it with `git apply` instead. If the user provides a diff for context or discussion rather than as a change to apply, respond accordingly instead.

CLAUDE.md

@@ -19,4 +19,5 @@ Read and follow the CONTRIBUTING.md file in this repository for all code style c
 - If you change CLI arguments, help text, or anything that affects command-line output, run `./generate-completions.sh` to regenerate the shell completion files, help text files, and `USAGE.md`. **Do not attempt to regenerate these files manually** — always use the script.
 - Run `FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh` to validate changes. If a test fails with a hint about `TEST_SKIP`, follow the hint and rerun with the suggested variable. If a sync test fails, read its error message carefully and run the exact command it tells you to run.
 - **ALWAYS run the full test suite** (`FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh`) before committing, regardless of how trivial the change seems — this includes documentation-only changes, comment edits, config changes, and refactors. The test suite checks formatting, linting, building, tests, and docs across multiple feature combinations; any type of change can break any of these checks.
+- When the user provides a diff and you need to update the files, don't manually interpret each hunk (that'd be slow); apply it with `git apply` instead. If the user provides a diff for context or discussion rather than as a change to apply, respond accordingly instead.
 - `gh` (GitHub CLI) is not installed — do not attempt to use it

template/ai-instructions/shared.md

@@ -19,3 +19,4 @@ Read and follow the CONTRIBUTING.md file in this repository for all code style c
 - If you change CLI arguments, help text, or anything that affects command-line output, run `./generate-completions.sh` to regenerate the shell completion files, help text files, and `USAGE.md`. **Do not attempt to regenerate these files manually** — always use the script.
 - Run `FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh` to validate changes. If a test fails with a hint about `TEST_SKIP`, follow the hint and rerun with the suggested variable. If a sync test fails, read its error message carefully and run the exact command it tells you to run.
 - **ALWAYS run the full test suite** (`FMT=true LINT=true BUILD=true TEST=true DOC=true ./test.sh`) before committing, regardless of how trivial the change seems — this includes documentation-only changes, comment edits, config changes, and refactors. The test suite checks formatting, linting, building, tests, and docs across multiple feature combinations; any type of change can break any of these checks.
+- When the user provides a diff and you need to update the files, don't manually interpret each hunk (that'd be slow); apply it with `git apply` instead. If the user provides a diff for context or discussion rather than as a change to apply, respond accordingly instead.