update-config

CHANGE = $ARGUMENTS - if that reads as a literal placeholder instead of a description, use the description I gave when invoking this skill. If I gave none, ask what config change I want before doing anything else.

Agent config on this machine is managed by the agentrc repo. What looks like ~/.claude/settings.json is a symlink into ~/agentrc/claude/, and the same goes for the other agent config files. Two consequences:

• File-editing tools refuse to write through symlinks. Always resolve the real path first with readlink -f <file> and edit that (e.g. ~/agentrc/claude/settings.json).
• A config edit is a repo change. It goes through the git workflow in AGENTS.md, never a silent in-place edit.

Your task

1. Resolve the target. Run readlink -f on the config file to get its repo path. If the file is NOT a symlink into ~/agentrc, stop and ask whether it should be added to the repo first.
2. Branch per AGENTS.md: switch to main (stash or commit unrelated WIP, never carry it along), pull, then branch off the fresh main.
3. Make the edit in the repo file. Read the file first and merge with what is there; never regenerate the whole file.
4. Permission rules must use forms the permission checker actually matches. File rules use Edit(path) - the Edit(...) form covers all file-editing tools (Edit, Write, NotebookEdit). Never write Write(path) rules; they are silently ignored and cause startup warnings.
5. Validate JSON files with jq . <file>; if jq is missing on the machine, fall back to python3 -m json.tool <file>. For settings semantics (hook events, schema fields), defer to the harness's own bundled config documentation rather than guessing.
6. Commit with a Conventional Commit message, then STOP. Show me the diff and wait - per AGENTS.md, never push or open a PR without my explicit go-ahead. New skill folders also need scripts/fetch-skills.sh re-run to link them; mention it when that applies.