GHSA-9GMC-JQMH-3RVM
Vulnerability from github – Published: 2026-08-19 19:16 – Updated: 2026-08-19 19:16Copier: trust-prefix bypass via path traversal runs tasks unprompted
Summary
In copier >= 9.5.0, <= 9.15.1, the trust setting's prefix match
(copier/_settings.py) compares the template URL against a trusted prefix with
a raw str.startswith and no path normalization, while the URL is
normalized when the template is actually fetched (Path.resolve() for local
paths; libcurl dot-segment removal for https). A template reference that
textually starts with a trusted prefix but contains ..
(e.g. https://github.com/trusted-org/../attacker-org/repo.git) is therefore
granted trust yet resolves to a different, attacker-controlled template, whose
tasks / migrations / jinja_extensions then run without the --trust
prompt — arbitrary command execution. Likely CWE-22 (Improper Limitation of
a Pathname) in the trust check leading to CWE-94 (code execution).
Details
trust lets users mark template locations as trusted so copier skips the
unsafe-feature gate. A trailing / makes an entry a prefix match
(docs/settings.md: "Locations ending with / will be matched as prefixes,
trusting all templates from that location").
copier/_settings.py:141-146 (tag v9.15.1):
return any(
repository.startswith(_normalize(t))
if t.endswith("/")
else repository == _normalize(t)
for t in trust
)
_normalize only expands ~; it does not touch .. or collapse segments —
copier/_settings.py:149-152 (tag v9.15.1):
def _normalize(url: str) -> str:
if url.startswith("~"): # Only expand on str to avoid messing with URLs
url = expanduser(url) # noqa: PTH111
return url
This decision gates code execution — copier/_main.py:293 (tag v9.15.1):
if self.unsafe or is_trusted_repository(self.settings.trust, self.template.url):
return # skip the unsafe-feature check entirely
The chain: the trust comparison sees the raw URL, so
"https://github.com/safeorg/../evilorg/t.git".startswith("https://github.com/safeorg/")
is True; but the value copier hands to git/pathlib is normalized, so
the template actually loaded is evilorg/t (a different, attacker-owned org).
Trust is granted to a location the user never trusted, and _check_unsafe
returns early, so the malicious template's tasks execute with no prompt.
This is most acute on copier update, which reads _src_path from the
project's .copier-answers.yml (copier/_subproject.py) — i.e. an attacker who
hands you a project controls the URL that the trust check is applied to.
In-repo asymmetry that confirms the omission: copier consistently resolves
paths everywhere else it makes a security decision — Path.resolve() plus
is_relative_to(...) guards in _render_template, template_copy_root, and
_external_data — but not in the trust comparison.
PoC
Self-contained standalone script; runs against a clean, pinned PyPI install via
the real copier CLI only. Static by default (copier copy --pretend
reaches the trust decision but does not execute tasks); --prove-exec is an
opt-in supplementary run that fires an inert marker (echo + touch). The full
poc.py accompanies this report.
Build and run:
python -m venv venv && . venv/bin/activate
pip install "copier==9.15.1"
python poc.py # static proof (default)
python poc.py --prove-exec # also fire the inert marker
Observed output (copier 9.15.1):
== version proof ==
copier == 9.15.1
module : .../site-packages/copier/__init__.py
== inputs ==
trusted prefix (settings.yml): /tmp/copier_trust_poc_XXXX/trusted_templates/
control src (canonical) : /tmp/copier_trust_poc_XXXX/attacker/evil_template
exploit src (traversal) : /tmp/copier_trust_poc_XXXX/trusted_templates/../attacker/evil_template
both resolve to the SAME dir : True
exploit startswith trusted/ : True
minimal delta : exploit = '/tmp/copier_trust_poc_XXXX/trusted_templates/..' + '/attacker/evil_template'
== static proof (copier copy --pretend; payload NOT executed) ==
control (canonical, untrusted): exit=4 -> BLOCKED (UnsafeTemplateError)
exploit (trusted-prefix /..) : exit=0 -> TRUSTED, task reached
marker on disk after --pretend: False (expected False: --pretend does not run tasks)
--- copier's own output for the exploit (note the task it WOULD run) ---
| Copying from template version None
| create hello.txt
| > Running task 1 of 1: echo COPIER-TRUST-BYPASS-RCE-MARKER && touch COPIER_RCE_PROOF
== VERDICT ==
BYPASS CONFIRMED: identical template is refused by canonical path
(exit 4) yet granted trust via '<trusted>/..' traversal (exit 0),
so its tasks run with no --trust prompt.
== --prove-exec: running the exploit for real (inert marker) ==
| COPIER-TRUST-BYPASS-RCE-MARKER
| > Running task 1 of 1: echo COPIER-TRUST-BYPASS-RCE-MARKER && touch COPIER_RCE_PROOF
exit=0 marker file 'COPIER_RCE_PROOF' created: True
-> ARBITRARY COMMAND EXECUTED via a 'trusted' template, no --trust
The exploit is the same template as the control plus the minimal delta
<trusted_prefix>/... Deterministic: same input → same result. The PoC uses a
local trusted prefix for a self-contained, network-free run; the https case is
identical because git normalizes .. before the request — e.g.
git ls-remote "https://github.com/copier-org/../pallets/flask.git" emits
warning: redirecting to https://github.com/pallets/flask.git/ and returns
pallets/flask's refs, a different org than the trusted copier-org/.
Impact
A user who has configured a trusted prefix (a trailing-/ entry in
trust, a documented feature) no longer gets the unsafe-feature prompt for a
template that merely appears to live under that prefix. Any party who can
influence the template URL — most realistically the author of a project the
victim runs copier update on, since _src_path comes from that project's
.copier-answers.yml — can host the real template under a different
org/location reached via .. and have its tasks/migrations/
jinja_extensions execute arbitrary commands with no prompt. It fires on a
default, modern git for both local paths and https.
Proposed severity: High, comparable to the project's prior unsafe-template
advisory (GHSA-3xw7-v6cj-5q8h). Proposed CVSS v4 vector (maintainer to finalize;
AT:P reflects the required trusted-prefix configuration):
CVSS:4.0/AV:L/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H. Conservative
variant if you scope impact to the user account only (no host escape claim):
drop SC/SI/SA to N.
Recommended fix
Normalize both sides before comparing, instead of raw startswith. For
local entries, compare resolved absolute paths (Path(t).resolve() vs
Path(repository).resolve()) using segment containment / is_relative_to, the
pattern already used in _render_template and template_copy_root. For URL
entries, parse the URL and reject or collapse .././empty path segments
before the prefix test. As defense-in-depth, reject any _src_path read from an
answers file that contains .. segments after the scheme/host, since
legitimate template URLs never need them.
References
- CWE-22 — https://cwe.mitre.org/data/definitions/22.html
- CWE-94 — https://cwe.mitre.org/data/definitions/94.html
- Affected source (tag
v9.15.1):copier/_settings.py:141-146(prefix match),copier/_settings.py:149-152(_normalize),copier/_main.py:293(trust gate). - Documented prefix behavior:
docs/settings.md("Locations ending with/will be matched as prefixes"). https..normalization: libcurl removes dot segments by default (CURLOPT_PATH_AS_ISdefaults to off) — https://curl.se/libcurl/c/CURLOPT_PATH_AS_IS.html- Novelty: distinct from copier's published advisories, which concern filesystem
read/write traversal in rendered output; this is an authorization bypass in
the
trustsetting's URL matching. The flawed match is identical between releasedv9.15.1and currentmasterHEAD, and unchanged since the trust-prefix feature was introduced inv9.5.0(originallycopier/settings.py, commit71358ed; renamed tocopier/_settings.pyin the v9.12.0 refactor).
{
"affected": [
{
"database_specific": {
"last_known_affected_version_range": "\u003c= 9.15.1"
},
"package": {
"ecosystem": "PyPI",
"name": "copier"
},
"ranges": [
{
"events": [
{
"introduced": "9.5.0"
},
{
"fixed": "9.15.2"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-53951"
],
"database_specific": {
"cwe_ids": [
"CWE-22",
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2026-08-19T19:16:54Z",
"nvd_published_at": "2026-07-08T16:16:30Z",
"severity": "HIGH"
},
"details": "# Copier: trust-prefix bypass via path traversal runs tasks unprompted\n\n### Summary\n\nIn copier `\u003e= 9.5.0, \u003c= 9.15.1`, the `trust` setting\u0027s prefix match\n(`copier/_settings.py`) compares the template URL against a trusted prefix with\na raw `str.startswith` and **no path normalization**, while the URL *is*\nnormalized when the template is actually fetched (`Path.resolve()` for local\npaths; libcurl dot-segment removal for `https`). A template reference that\ntextually starts with a trusted prefix but contains `..`\n(e.g. `https://github.com/trusted-org/../attacker-org/repo.git`) is therefore\ngranted trust yet resolves to a different, attacker-controlled template, whose\n`tasks` / `migrations` / `jinja_extensions` then run **without the `--trust`\nprompt** \u2014 arbitrary command execution. Likely **CWE-22 (Improper Limitation of\na Pathname)** in the trust check leading to **CWE-94 (code execution)**.\n\n### Details\n\n`trust` lets users mark template locations as trusted so copier skips the\nunsafe-feature gate. A trailing `/` makes an entry a **prefix** match\n(`docs/settings.md`: *\"Locations ending with `/` will be matched as prefixes,\ntrusting all templates from that location\"*).\n\n`copier/_settings.py:141-146` (tag `v9.15.1`):\n\n```python\n return any(\n repository.startswith(_normalize(t))\n if t.endswith(\"/\")\n else repository == _normalize(t)\n for t in trust\n )\n```\n\n`_normalize` only expands `~`; it does **not** touch `..` or collapse segments \u2014\n`copier/_settings.py:149-152` (tag `v9.15.1`):\n\n```python\ndef _normalize(url: str) -\u003e str:\n if url.startswith(\"~\"): # Only expand on str to avoid messing with URLs\n url = expanduser(url) # noqa: PTH111\n return url\n```\n\nThis decision gates code execution \u2014 `copier/_main.py:293` (tag `v9.15.1`):\n\n```python\n if self.unsafe or is_trusted_repository(self.settings.trust, self.template.url):\n return # skip the unsafe-feature check entirely\n```\n\nThe chain: the trust comparison sees the **raw** URL, so\n`\"https://github.com/safeorg/../evilorg/t.git\".startswith(\"https://github.com/safeorg/\")`\nis `True`; but the value copier hands to `git`/`pathlib` is **normalized**, so\nthe template actually loaded is `evilorg/t` (a different, attacker-owned org).\nTrust is granted to a location the user never trusted, and `_check_unsafe`\nreturns early, so the malicious template\u0027s tasks execute with no prompt.\n\nThis is most acute on `copier update`, which reads `_src_path` from the\nproject\u0027s `.copier-answers.yml` (`copier/_subproject.py`) \u2014 i.e. an attacker who\nhands you a project controls the URL that the trust check is applied to.\n\nIn-repo asymmetry that confirms the omission: copier consistently resolves\npaths *everywhere else* it makes a security decision \u2014 `Path.resolve()` plus\n`is_relative_to(...)` guards in `_render_template`, `template_copy_root`, and\n`_external_data` \u2014 but not in the trust comparison.\n\n### PoC\n\nSelf-contained standalone script; runs against a clean, pinned PyPI install via\nthe real `copier` CLI only. **Static by default** (`copier copy --pretend`\nreaches the trust decision but does not execute tasks); `--prove-exec` is an\nopt-in supplementary run that fires an inert marker (`echo` + `touch`). The full\n`poc.py` accompanies this report.\n\nBuild and run:\n\n```bash\npython -m venv venv \u0026\u0026 . venv/bin/activate\npip install \"copier==9.15.1\"\npython poc.py # static proof (default)\npython poc.py --prove-exec # also fire the inert marker\n```\n\nObserved output (`copier 9.15.1`):\n\n```\n== version proof ==\n copier == 9.15.1\n module : .../site-packages/copier/__init__.py\n\n== inputs ==\n trusted prefix (settings.yml): /tmp/copier_trust_poc_XXXX/trusted_templates/\n control src (canonical) : /tmp/copier_trust_poc_XXXX/attacker/evil_template\n exploit src (traversal) : /tmp/copier_trust_poc_XXXX/trusted_templates/../attacker/evil_template\n both resolve to the SAME dir : True\n exploit startswith trusted/ : True\n minimal delta : exploit = \u0027/tmp/copier_trust_poc_XXXX/trusted_templates/..\u0027 + \u0027/attacker/evil_template\u0027\n\n== static proof (copier copy --pretend; payload NOT executed) ==\n control (canonical, untrusted): exit=4 -\u003e BLOCKED (UnsafeTemplateError)\n exploit (trusted-prefix /..) : exit=0 -\u003e TRUSTED, task reached\n marker on disk after --pretend: False (expected False: --pretend does not run tasks)\n\n --- copier\u0027s own output for the exploit (note the task it WOULD run) ---\n | Copying from template version None\n | create hello.txt\n | \u003e Running task 1 of 1: echo COPIER-TRUST-BYPASS-RCE-MARKER \u0026\u0026 touch COPIER_RCE_PROOF\n\n== VERDICT ==\n BYPASS CONFIRMED: identical template is refused by canonical path\n (exit 4) yet granted trust via \u0027\u003ctrusted\u003e/..\u0027 traversal (exit 0),\n so its tasks run with no --trust prompt.\n\n== --prove-exec: running the exploit for real (inert marker) ==\n | COPIER-TRUST-BYPASS-RCE-MARKER\n | \u003e Running task 1 of 1: echo COPIER-TRUST-BYPASS-RCE-MARKER \u0026\u0026 touch COPIER_RCE_PROOF\n exit=0 marker file \u0027COPIER_RCE_PROOF\u0027 created: True\n -\u003e ARBITRARY COMMAND EXECUTED via a \u0027trusted\u0027 template, no --trust\n```\n\nThe exploit is the same template as the control plus the minimal delta\n`\u003ctrusted_prefix\u003e/..`. Deterministic: same input \u2192 same result. The PoC uses a\nlocal trusted prefix for a self-contained, network-free run; the `https` case is\nidentical because git normalizes `..` before the request \u2014 e.g.\n`git ls-remote \"https://github.com/copier-org/../pallets/flask.git\"` emits\n`warning: redirecting to https://github.com/pallets/flask.git/` and returns\n`pallets/flask`\u0027s refs, a different org than the trusted `copier-org/`.\n\n### Impact\n\nA user who has configured a trusted **prefix** (a trailing-`/` entry in\n`trust`, a documented feature) no longer gets the unsafe-feature prompt for a\ntemplate that merely *appears* to live under that prefix. Any party who can\ninfluence the template URL \u2014 most realistically the author of a project the\nvictim runs `copier update` on, since `_src_path` comes from that project\u0027s\n`.copier-answers.yml` \u2014 can host the real template under a different\norg/location reached via `..` and have its `tasks`/`migrations`/\n`jinja_extensions` execute arbitrary commands with no prompt. It fires on a\ndefault, modern git for both local paths and `https`.\n\nProposed severity: **High**, comparable to the project\u0027s prior unsafe-template\nadvisory (GHSA-3xw7-v6cj-5q8h). Proposed CVSS v4 vector (maintainer to finalize;\n`AT:P` reflects the required trusted-prefix configuration):\n`CVSS:4.0/AV:L/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H`. Conservative\nvariant if you scope impact to the user account only (no host escape claim):\ndrop `SC/SI/SA` to `N`.\n\n### Recommended fix\n\nNormalize **both** sides before comparing, instead of raw `startswith`. For\nlocal entries, compare resolved absolute paths (`Path(t).resolve()` vs\n`Path(repository).resolve()`) using segment containment / `is_relative_to`, the\npattern already used in `_render_template` and `template_copy_root`. For URL\nentries, parse the URL and reject or collapse `..`/`.`/empty path segments\nbefore the prefix test. As defense-in-depth, reject any `_src_path` read from an\nanswers file that contains `..` segments after the scheme/host, since\nlegitimate template URLs never need them.\n\n### References\n\n- CWE-22 \u2014 https://cwe.mitre.org/data/definitions/22.html\n- CWE-94 \u2014 https://cwe.mitre.org/data/definitions/94.html\n- Affected source (tag `v9.15.1`): `copier/_settings.py:141-146` (prefix match),\n `copier/_settings.py:149-152` (`_normalize`), `copier/_main.py:293` (trust gate).\n- Documented prefix behavior: `docs/settings.md` (\"Locations ending with `/`\n will be matched as prefixes\").\n- `https` `..` normalization: libcurl removes dot segments by default\n (`CURLOPT_PATH_AS_IS` defaults to off) \u2014 https://curl.se/libcurl/c/CURLOPT_PATH_AS_IS.html\n- Novelty: distinct from copier\u0027s published advisories, which concern filesystem\n read/write traversal in rendered output; this is an authorization bypass in\n the `trust` setting\u0027s URL matching. The flawed match is identical between\n released `v9.15.1` and current `master` HEAD, and unchanged since the\n trust-prefix feature was introduced in `v9.5.0` (originally `copier/settings.py`,\n commit `71358ed`; renamed to `copier/_settings.py` in the v9.12.0 refactor).",
"id": "GHSA-9gmc-jqmh-3rvm",
"modified": "2026-08-19T19:16:54Z",
"published": "2026-08-19T19:16:54Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/copier-org/copier/security/advisories/GHSA-9gmc-jqmh-3rvm"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-53951"
},
{
"type": "PACKAGE",
"url": "https://github.com/copier-org/copier"
},
{
"type": "WEB",
"url": "https://github.com/copier-org/copier/releases/tag/v9.15.2"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:L/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H",
"type": "CVSS_V4"
}
],
"summary": "Copier has a trust-prefix bypass via path traversal that runs tasks unprompted"
}
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.