CVE-2026-50024 is a medium-severity (CVSS 5.3) Path Traversal vulnerability in githacker. A fix is available for githacker — see the affected versions and patch details below.
GitHacker: Path traversal in ref/hash parsing enables existence oracle and hex-fragment exfiltration via a malicious .git server
Exploitation Status
No confirmed exploitation observed yet
- CISA assesses this as automatable — exploitation doesn’t require manual, per-target effort, which raises the odds of mass scanning and opportunistic attacks.
- CISA’s own triage has not observed active exploitation or public proof-of-concept code for this CVE as of its last assessment.
Exploitation and automatability from CISA’s SSVC triage for CVE-2026-50024.
EPSS Exploitation Probability
Probability of exploitation in the next 30 days, from FIRST.org EPSS.
How urgent is this, really
CVE-2026-50024 by exploitation likelihood (EPSS) against impact (CVSS). Outside the shaded patch-first corner.
Where this sits among everything scored
Of 395,249 CVEs with a current EPSS score, this one falls in the < 10% band (highlighted). Counts from FIRST.org, log-scaled.
Real-World Exposure
githackerReal-time download stats are indexed for npm and PyPI packages. This vulnerability affects PyPI packages — download data is not available via public APIs for these ecosystems.
Description
Summary
GitHacker through 1.1.7 did not validate path segments parsed from attacker-controlled .git/HEAD before joining them onto its output directory. A malicious server could coerce GitHacker into reading arbitrary local files. Contents do not stream back wholesale, but the recovery loop turns any 40-character hex substring into an outbound HTTP GET — an existence oracle for arbitrary paths plus hex-fragment exfiltration of file contents.
Details
Vulnerability
GitHacker rebuilds a remote .git/ by fetching files into temp_dst. Two functions derived filesystem paths from server-controlled content:
add_head_file_tasksreads the downloaded.git/HEAD, parsesref: <ref-path>, and joins the raw ref-path ontotemp_dst/.git/logs/before reading the resulting file.add_hashes_parsedscans any file it reads for 40-character hex substrings and emitsGET .git/objects/<sha[0:2]>/<sha[2:]>for each one — onto the attacker's server and into the local output tree.
Pre-fix, add_head_file_tasks did not validate the ref segments. A malicious .git/HEAD of
ref: ../../../../../../etc/passwd
caused add_head_file_tasks to traverse out of temp_dst and read /etc/passwd. The bytes flowed into add_hashes_parsed, which emitted one outbound HTTP request per 40-char-hex match — observable on the attacker's logs.
Impact
PR #65 originally classified this as arbitrary local file read. Joint analysis during coordinated disclosure narrowed the primitive: file contents do not stream back wholesale because the only egress channel is the 40-char-hex regex. In practice an attacker can:
- Existence oracle for any path on the GitHacker host (
/etc/shadow,/root/.ssh/id_rsa,/home/<user>/.git-credentials, build artifacts under/tmp/). - Hex-fragment exfiltration when the targeted file contains 40-char hex sequences: other git repos' refs / pack filenames, password hashes, HMAC-SHA1 outputs, some session tokens.
Not exploitable without victim action: the attacker must persuade the victim to run GitHacker against a URL they control. Project guidance has always been to run GitHacker inside a disposable container.
Scoping note: no write-side primitive in 1.1.7
A working write-side primitive (attacker drops content outside temp_dst via add_folder / add_task) does not reproduce against the shipped 1.1.7 source. Empirical testing by the reporter against GitHacker-1.1.7.tar.gz (25 traversal-style payloads including ....//, %2e%2e%2f, layered foo/../../, NUL bytes, backslash variants, absolute paths) yielded 0/25 escapes. Two structural reasons:
add_folderanchors every derived path onself.url + '.git/', so the first path component after theurl_lengthstrip is always.git.os.path.join's absolute-path short-circuit never fires.- Python's
str.replace("..", "")is greedy non-overlapping;....//collapses to//,......//to///, etc. No literal..survives intoadd_task.
5f2a8ba is still the correct fix for the read-side primitive and additionally hardens add_task as defense in depth against future regressions — for example, if a later caller removes the .git/ anchor in add_folder or wires a new server-controlled segment source into add_task.
Fix
Commit 5f2a8ba introduces _is_safe_path_segment as a single trust boundary: every segment about to be joined onto temp_dst or appended to an outgoing URL is validated against an allowlist before add_task accepts it. Empty / . / .. / separators / NUL / control characters are rejected; the brittle replace("..", "") filter is removed.
The fix also tightens adjacent surfaces preemptively:
add_folderswitches tourlparse-based scheme + netloc + path comparison.construct_url_from_path_componentspercent-encodes every segment.
PR #65's two-layer defense (allowlist regex + os.path.realpath() confinement) was consolidated onto the allowlist applied at queue-time, removing the TOCTOU window an after-the-fact realpath() check leaves open and the per-call-site drift risk. PR #65 was closed in favour of the broader fix.
Regression tests in tests/test_ref_validation.py (commit 16fcd81) pin the PoC and six bypass variants (extra-depth, mid-path, NUL, absolute path, leading-dot, .lock-suffix).
Credit
Reported and patched-prototyped by Zac Wang (@7a6163) in #65. Zac refined the impact framing from "arbitrary file read" to "existence oracle + hex-fragment exfiltration" and verified the absence of a write-side primitive against the shipped 1.1.7 sdist with a 25-payload harness.
Patches
Patched in 1.1.8 (commit 5f2a8ba; tests 16fcd81).
Workarounds
Run GitHacker inside a disposable container. Do not point GitHacker at any URL whose contents are not under your control.
Resources
- https://github.com/WangYihang/GitHacker/pull/65
- https://github.com/WangYihang/GitHacker/commit/5f2a8ba
- https://github.com/WangYihang/GitHacker/blob/main/tests/test_ref_validation.py
- https://githacker.pages.dev/security
- https://github.com/justinsteven/advisories
- https://drivertom.blogspot.com/2021/08/git.html
Affected Packages
| Ecosystem | Package | Vulnerable range | Fix |
|---|---|---|---|
| 🐍PyPI | githacker | all versions | 1.1.8pip install --upgrade 'githacker==1.1.8' |
Detection & mitigation playbook
Open-source dependencyDetect
Scan your dependency tree (package-lock.json, pnpm-lock.yaml, requirements.txt, go.sum, etc.) for githacker, including transitive dependencies — a direct dependency you never call can still pull in a vulnerable version.
Fix
Update githacker to 1.1.8 or later, then make sure no transitive (indirect) dependency still pins the vulnerable range — O3 confirms CVE-2026-50024 is resolved across your whole dependency graph.
Workarounds
Resolve every user-supplied path to its canonical form and reject anything that escapes the intended directory, and run the component under an account that has no read or write access outside the directory it legitimately serves.
Frequently Asked Questions
Is CVE-2026-50024 in your dependencies?
Find it across PyPI, including transitive dependencies.