Imported from marcostomatti/template-agentic-research (.claude/skills/git-add-partial-failure/SKILL.md). Install upstream with npx skills add marcostomatti/template-agentic-research --skill git-add-partial-failure. Copyright stays with the author.

git add Aborts Entirely On Any Missing Pathspec

git add A B C treats the file list atomically. If any of A/B/C does not exist as a pathspec (typo, stale expectation, wrong lockfile extension), the whole git add fails and nothing gets staged — not even the files that DO exist. If you're piping stderr to /dev/null and chaining with a bare newline (not &&), the shell keeps going and the next git commit finds an empty index. The commit reports "no changes added to commit" and lists the still-modified files, which reads like the modifications weren't detected.

The specific bug

Ran in a bash loop iterating over multiple repos, adding + committing dep-pin changes:

for repo in ...; do
  cd "$repo"
  git status --short
  git add package.json bun.lock bun.lockb 2>/dev/null   # bun.lockb doesn't exist in modern bun
  git -c commit.gpgsign=false commit -m "..."
done

bun.lockb (legacy binary lockfile) does not exist post-migration to text bun.lock. Every iteration:

1. git add failed with fatal: pathspec 'bun.lockb' did not match any files (silenced by 2>/dev/null).
2. Nothing was staged.
3. git commit reported no changes added to commit with the modified files listed.

Looked like a mysterious "changes not being detected" symptom. Real cause: silent partial-add failure.

Fixes (any one is sufficient)

• Drop 2>/dev/null from git add. The stderr messages are the diagnostic.
• Add files individually, ignore per-file failures:

  for f in package.json bun.lock bun.lockb; do
    git add "$f" 2>/dev/null || true
  done

  This preserves the "add whatever exists" behavior without atomic abort.
• Use git add --ignore-errors A B C (best-of-both — atomic-ish behavior but skips non-existent paths).
• Use git add -A scoped to a directory if you actually want "everything modified here": git add -A . in the target dir. Loses the file-list intent but is bulletproof.
• Test each pathspec with [ -f "$f" ] before including it in the git add list — verbose but explicit.

Related failure mode

git rm A B C has the same behavior. If any pathspec doesn't match tracked files, the whole command fails. Same fixes apply.

When to Use

• A git commit in a scripted loop reports "no changes added to commit" for files that clearly show as modified in git status.
• Writing a shell loop that batches git add across multiple files, especially with 2>/dev/null or when file existence is uncertain (lockfile format transitions, optional dotfiles).
• Debugging why a chained git add && git commit produced an empty commit.

Anti-pattern

Assuming git add is a for-each-file operation. It isn't — it's a single atomic operation over the pathspec list. Modeling it as for-each leads to silent staging losses.