pull-changes-resolve-conflicts — Claude Code skill: merge without
A skill that walks the agent through branch merges: it reports conflicts and asks first, edits later. Useful on projects with several contributors.
npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a claude-codeWhat pull-changes-resolve-conflicts does
The skill sets the order of operations for git merge and git rebase so the current branch's work does not disappear quietly. The agent runs git fetch, then git merge origin/main (or the named branch); if a merge is already underway it runs git status and lists every unmerged file. Then it stops and reports: current branch, target branch, merge or rebase, and the paths.
Before touching conflict markers, the agent examines each area using conversation history, repo documents such as PRD.md, recent commits on the branch, and the hunks themselves via git show :2:path and git show :3:path. Changes get sorted into orthogonal, overlapping, and corruption-risk cases where blocks are duplicated. Recommendations come ordered by impact: structural and product decisions first, then behavior, then layout and styling. Ambiguous spots turn into explicit questions before any file is edited.
The hard rule is blunt: no conflict is resolved without user agreement. Removing <<<<<<< markers or running git add waits for a reply, or for an explicit "follow your recommendations". Without that reply the merge just sits there. You can back out with git merge --abort or git rebase --abort, though the in-progress integration state is lost.
What it helps with
- merge main into a feature branch without dropping your own edits
- get a full list of unmerged files after a merge or rebase
- tell orthogonal changes apart from overlapping ones
- abort a merge when the conflicts turn out to be a mess
- separate product-level conflicts from styling and spacing tweaks
Example prompts for the agent
- merge origin/main into this branch and show me the conflicts
- a merge is in progress, list the unmerged paths and propose a plan
- resolve the conflicts following your recommendations
How to install pull-changes-resolve-conflicts
1Skills CLI (Vercel)
npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a claude-code2Manually: put the files in place
~/.claude/skills/pull-changes-resolve-conflicts/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/pull-changes-resolve-conflicts/
1Skills CLI (Vercel)
npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a codex2Manually: put the files in place
.agents/skills/pull-changes-resolve-conflicts/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a cursor2Manually: put the files in place
.cursor/skills/pull-changes-resolve-conflicts/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install cognitedata/builder-skills pull-changes-resolve-conflicts2Manually: put the files in place
.github/skills/pull-changes-resolve-conflicts/SKILL.md1Skills CLI (Vercel)
npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a gemini-cli2Platform CLI
gemini skills install https://github.com/cognitedata/builder-skills --path skills/pull-changes-resolve-conflictsManually: put the files in place
.devin/skills/pull-changes-resolve-conflicts/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about pull-changes-resolve-conflicts
How do I install pull-changes-resolve-conflicts in Claude?
Run npx skills add cognitedata/builder-skills --skill pull-changes-resolve-conflicts -a claude-code. The skill file lands at ~/.claude/skills/pull-changes-resolve-conflicts/SKILL.md.
What does pull-changes-resolve-conflicts do?
It gives the agent a merge workflow: fetch, merge, report conflicts, analyze ours versus theirs, present prioritized recommendations, and ask the user. Edits to files start only after an answer.
Can I just tell the agent to take main?
The skill forbids picking one side wholesale without analysis and approval. You have to say so explicitly, otherwise the agent stops at the conflict report.
Source text
SKILL.md · Apache-2.0
--- name: pull-changes-resolve-conflicts description: >- Standard workflow for pulling updates from main or other branches on multi-contributor projects (including Flows apps) without silently discarding work. Guides fetching/merging, requires listing merge conflicts explicitly, analyzing ours vs theirs using conversation history and repo context, presenting prioritized recommendations, and obtaining user answers before editing conflict markers or completing the merge. Triggers: pull main, merge main, merge origin, rebase, merge conflict, unmerged paths, both modified, integrate branch, sync with main, git merge abort, resolve conflicts, UU status, theirs vs ours, feat branch update. allowed-tools: Read, Glob, Grep, Shell, Write metadata: argument-hint: "" --- # Pull changes & resolve merge conflicts Use this skill whenever integrating another branch (usually `main`) into the current feature branch, or when `git status` shows unmerged paths after a merge or rebase. Applies to any Git-based team workflow; Flows/React apps are a common case where conflicts cluster in app shells and shared libraries. ## Goals - Preserve intentional work on the current branch; **do not assume** “main wins” or “ours wins” without analysis. - Make trade-offs **visible** to the user **before** any conflict resolution edits. - Order discussion by **impact**: structural / feature / API / data-model changes before styling, copy, or spacing. ## Hard rules 1. **No silent resolution** — Do not remove `<<<<<<<` / `=======` / `>>>>>>>` or run `git add` on conflicted files until the user has agreed to the plan (or explicitly says “use your recommendations”). 2. **Stop at conflicts** — If a merge or rebase introduces conflicts, **pause** and report; do not bulldoze through large files by picking one side wholesale unless the user explicitly requests that. 3. **Prioritize impact** — When presenting conflicts, group and order roughly as: - **P0 — Structural / product:** removed routes, deleted modules, dropped features, changed data or API contracts, SDK or schema changes, auth or routing shell. - **P1 — Behavior:** logic, hooks, queries, filters, error handling, loading states. - **P2 — UI structure:** layout regions, new or removed sections, navigation. - **P3 — Presentation:** tokens, spacing, class names, copy tweaks. ## Workflow ### 1. Fetch and integrate (or diagnose) - Prefer `git fetch` then `git merge origin/main` (or the named branch) unless the user asked for rebase. - If merge is already in progress, run `git status` and list **every** unmerged file. ### 2. Report conflicts to the user (explicit) Output a clear list: - **Branch state:** current branch, target branch (e.g. `origin/main`), merge vs rebase. - **Unmerged files:** paths only, then optionally `git diff --name-only --diff-filter=U`. - **Per file (short):** one line on *what* diverged (e.g. “`AlertsPage` — layout + new data scope”) if inferable from paths and `git diff` **without** resolving. ### 3. Analyze before editing Use **all** of: - **Conversation history** — What was the user or team trying to ship on this branch? - **Repo signals** — Product or architecture docs if present (e.g. `PRD.md`), recent commits on the current branch, file ownership (e.g. large feature module vs shared `lib/`). - **Conflict hunks** — `git show :2:path` (ours) vs `git show :3:path` (theirs) during merge, or read conflict markers; identify duplicated vs orthogonal changes. Classify each conflicted area as: - **Orthogonal** — safe to combine (e.g. import sort + new prop). - **Overlapping** — must choose or manually merge (same lines). - **Corruption risk** — duplicated blocks (common after bad merges); flag and recommend reconstructing from one side then re-applying the other side’s intent manually. ### 4. Recommendations + questions (required) Present to the user: 1. **Summary table or bullets** — file → recommended side *or* “manual merge” → one-line **why**. 2. **Ordered by P0 → P3** — call out anything that **removes** a feature or **changes** public behavior first. 3. **Explicit questions** — anything ambiguous (e.g. “Keep main’s global behavior or the branch’s scoped variant?”). 4. **Ask for direction** — e.g. “Reply with: (a) follow recommendations, (b) keep branch for file X, (c) keep main for file Y, (d) abort merge.” **Only after** the user confirms (or gives a precise mapping), apply resolutions: - Prefer small, surgical edits; preserve both sides’ intent when possible. - Re-run `git status`; ensure no conflict markers remain; run tests or lint the user cares about for touched areas. ### 5. If the user wants to abort - `git merge --abort` or `git rebase --abort` as appropriate; confirm they lose in-progress integration state for that operation. ## Anti-patterns (do not do) - Picking `--ours` / `--theirs` on the whole repo without user approval. - “Resolving” by deleting a feature branch’s work because main touched the same file. - Hiding conflict lists inside a long code dump without a short executive summary. - Fixing low-impact style conflicts first while leaving P0 decisions implicit. ## Quick reference - **Ours vs theirs (merge):** stage 2 = current branch (`HEAD`), stage 3 = incoming (`MERGE_HEAD`). Verify with `git checkout --conflict=merge <file>` if needed. - **Typical high-touch paths** in full-stack or Flows apps: root app shell, top navigation, route modules, and shared `lib/` or `hooks/`. For optional command snippets and a merge message template, see [reference.md](reference.md).
© cognitedata · Apache-2.0 · cognitedata/builder-skills