repo-support

Files (3)

• README.md
• SKILL.md
• skill.json

SKILL.md (first lines)

# Repo Support

Use this to orient a coding agent inside a customer or developer workspace with
the fewest possible tool round-trips, keeping the ChatGPT thread fast.

## Rules

- Start with one wide call, not many tiny ones.
- Do not re-read files you already read; read only the line range you need.
- For long output, store it locally and return only a compact summary/path.

## Steps

1. Map the repo in one call: `workspace_snapshot` (or `repo_map`).
2. Find code with `search_text` (ripgrep fast path) and `repo_symbols`.
3. Read narrowly with `read_file` line ranges or `read_many`.
4. Edit with `preview_patch` -> `validate_patch` -> `apply_patch`, then
   `run_changed_tests`.
5. Before handoff, call `session_report`. If the thread is long, call
   `checkpoint`, open a NEW ChatGPT chat, and call `resume` first.
6. For large logs/output, keep raw details in a local report/support file and
   share the path plus a compact summary instead of the raw text.

## Report Back

Return what the repo is, the key files touched, test status, and the local path
for any long output.

From LongNgn204/local-coding-agent, AGPL-3.0.

More from LongNgn204/local-coding-agent