Reported issues for GitHub

Pod holds 7 of 7 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to GitHub.

Most discussed

get_review_comments overflows on GitHub review comment IDs above int32

Summary

pull_request_read(method: "get_review_comments") fails for current GitHub review-comment database IDs that exceed signed 32-bit range.

Reproduction

Repository: stokaro/ptah.

Calls such as:

{"owner":"stokaro","repo":"ptah","pullNumber":1405,"method":"get_review_comments","perPage":100}

fail with:

failed to get pull request review threads: json: cannot unmarshal number 3757339736 into Go value of type githubv4.Int

The same failure occurs for other

Read the thread · 2026-08-11 · open · outside contributor · 0 comments

Missing rebase-state recovery tools: git rebase --abort / --continue / --skip

What is missing

There is no tool to manage an in-progress rebase. Notably, git_pull performs an automatic rebase — so the server itself can put a repository into a conflicted rebase in progress state that none of its own tools can then exit.

How it happened (real session)

1. A feature branch had been squash-merged into master on GitHub.
2. While still on the local feature branch, git_pull (branch: master) fetched and started rebasing the branch onto the new master.
3. The

Read the thread · 2026-07-11 · open · outside contributor · 1 comment

Proposal: ship a Claude Code skill ("omnigit-mcp-only") to enforce MCP-first usage

Scenario

Claude Code agents working in repos with omnigit-mcp configured tend to drift back to Bash: git ... / Bash: gh ... for "small" operations even when MCP equivalents exist. Each time this happens you lose the structural advantages of the MCP (typed I/O, no shell-quoting bugs in commit messages, no PATH or alias surprises) and you mix two different audit/auth surfaces.

A Claude Code skill is the right shape to fix this — once installed, the skill auto-loads whenever the user

Read the thread · 2026-04-26 · closed · outside contributor · 0 comments

Missing: force-push, commit --amend, and reset --hard (gated as opt-in)

Problem

Several common git operations have no MCP wrapper today, forcing AI agents to fall back to Bash: git ... despite a strict "MCP only" policy. Specifically:

1. git push --force / git push --force-with-lease

CLI used:

git push --force-with-lease origin <branch>

What I needed: Overwrite a remote branch after amending or rewriting history locally. Standard part of an iterative PR workflow (fix typo → amend → force-push), and force-with-lease is the safe varian

Read the thread · 2026-04-26 · open · outside contributor · 1 comment

Missing: path filtering in git_log, git_show --stat, and cross-branch log (--all)

Problem

When investigating schema drift in a repository, I needed several git operations that the current MCP tools don't support, which forced me to fall back to CLI commands. Specifically:

1. git log with path filter (-- <path>)

CLI used:

git log --oneline -20 -- go/models/
git log --oneline -20 -- go/schema/migrations/_sqldata/

What I needed: See which commits touched a specific file or directory. This is essential for tracing when a file was last modified.

**C

Read the thread · 2026-03-02 · open · outside contributor · 0 comments

git_worktree_add reports success but working directory is empty (files not checked out)

Problem

git_worktree_add returns a success message and the worktree appears correctly in git_worktree_list, but the worktree directory is completely empty — no files are checked out.

Reproduction steps:

1. Call git_worktree_add with repo_path, worktree_path, new_branch, and commitish:

   {
     "repo_path": "/path/to/repo",
     "worktree_path": "/path/to/repo/.claude/worktrees/my-feature",
     "new_branch": "my-feature-branch",
     "commitish": "master"
   }
   

Read the thread · 2026-02-26 · closed · outside contributor · 2 comments

git_worktree_add fails with "invalid reference" for freshly-created remote branches

Problem

git_worktree_add fails with fatal: invalid reference when the target branch was just created on the remote (e.g. via create_branch_github_git_mcp) but has not yet been fetched into the local repo.

Reproduction steps:

1. Create a branch on GitHub using create_branch_github_git_mcp (or any remote operation).
2. Immediately call git_worktree_add with commitish set to that branch name.
3. The tool fails:

   Failed to add worktree: failed to add worktree: git comm
   

Read the thread · 2026-02-26 · open · outside contributor · 0 comments

Most recent

The remaining reports are on the project's issue tracker.