PraisonAI Dynamic Context history and terminal tools read files outside configured storage via path traversal
- When
- Where
- Global (internet)
- Category
- cyber_advisory · pip
# PraisonAI Dynamic Context history and terminal tools read files outside configured storage via path traversal ## Summary PraisonAI's Dynamic Context module provides filesystem-backed history and terminal-log storage. The SDK reference describes the module as providing: - artifact storage for tool outputs, history, and terminal logs; - history persistence with search; and - terminal session logging. The module also exports agent-callable tool factories: - `create_history_tools()` returns `history_search`, `history_tail`, and `history_get`. - `create_terminal_tools()` returns `terminal_tail`, `terminal_grep`, and `terminal_commands`. Those tools accept `run_id` and `agent_id` arguments from the tool caller. The underlying stores join those values into filesystem paths without rejecting absolute paths or `..` traversal: ```python history_dir = self.base_dir / run_id / "history" return history_dir / f"{agent_id}.jsonl" ``` ```python terminal_dir = self.base_dir / run_id / "terminal" return terminal_dir / f"{agent_id}.log" ``` Because `run_id` can be an absolute path and `agent_id` can contain traversal, a lower-trust prompt/user that can call these tools can read `.jsonl` and `.log` files outside the configured Dynamic Context base directory. ## Affected Product - Repository: `MervinPraison/PraisonAI` - Ecosystem: `pip` - Package: `praisonai` - Component: Dynamic Context history and terminal tools - Current source paths: - `src/praisonai/praisonai/context/history_store.py` - `src/praisonai/praisonai/context/terminal_logger.py` - Latest PyPI version validated: `4.6.58` - Current `origin/main` validated: `1ad58ca02975ff1398efeda694ea2ab78f20cf3e` - Current `origin/main` tag validated: `v4.6.58` Suggested affected range: ```text pip:praisonai >= 3.8.1, <= 4.6.58 ``` Representative local sweep: - `3.8.1`: vulnerable - `4.0.0`: vulnerable - `4.5.113`: vulnerable - `4.6.33`: vulnerable - `4.6.34`: vulnerable - `4.6.40`: vulnerable - `4.6.50`: vulnerable - `4.6.58`: vulnerable ## Root Cause `HistoryStore._get_history_path()` and `TerminalLogger._get_log_path()` treat logical identifiers as path segments, but never validate that the resolved path stays under `base_dir`. History path construction: ```python def _get_history_path(self, run_id: str, agent_id: str) -> Path: history_dir = self.base_dir / run_id / "history" history_dir.mkdir(parents=True, exist_ok=True) return history_dir / f"{agent_id}.jsonl" ``` Terminal path construction: ```python def _get_log_path(self, run_id: str, agent_id: str) -> Path: terminal_dir = self.base_dir / run_id / "terminal" terminal_dir.mkdir(parents=True, exist_ok=True) return terminal_dir / f"{agent_id}.log" ``` The agent tools pass caller-controlled `run_id` and `agent_id` directly into these helpers: ```python def history_tail(agent_id: str = "default", run_id: str = "default", count: int = 10) -> str: messages = history_store.get_last_messages(agent_id=agent_id, run_id=run_id, count=count) ``` ```python def terminal_tail(agent_id: str = "default", run_id: str = "default", lines: int = 50) -> str: return term_logger.tail_session(agent_id=agent_id, run_id=run_id, lines=lines) ``` There is no check equivalent to: ```python resolved = candidate.resolve() base = self.base_dir.resolve() resolved.relative_to(base) ``` There is also no identifier allowlist preventing `/`, `\`, or `..` in `run_id` or `agent_id`. ## Local PoV Run against the latest PyPI package: ```bash uv run --with 'praisonai==4.6.58' \ python poc/pov_prai_cand_027_history_terminal_tools_path_traversal.py --json ``` The PoV: 1. Creates a temporary Dynamic Context base directory. 2. Creates a separate outside directory containing `secret.jsonl` and `secret.log`. 3. Creates legitimate in-base history and terminal log controls. 4. Calls `history_tail()` and `history_get()` with `run_id=<outside-dir>` and `agent_id=../secret`. 5. Calls `terminal_tail()` and `terminal_grep()` with the same traversal. 6. Confirms the traversal paths resolve to files outside the configured base. Observed output summary from `evidence/pov-pypi-4.6.58.json`: ```json { "package": "praisonai", "package_version": "4.6.58", "controls": { "valid_history_read_works": true, "valid_terminal_read_works": true, "outside_history_file_outside_base_dir": true, "outside_terminal_file_outside_base_dir": true, "traversal_history_path_resolves_to_outside_file": true, "traversal_terminal_path_resolves_to_outside_file": true }, "outside_history_tail": "Last 1 messages:\\n\\n[system]: PRAI-CAND-027-HISTORY-SECRET", "outside_terminal_tail": "PRAI-CAND-027-TERMINAL-SECRET\\nsecond line\\n", "outside_terminal_grep": "Found 1 matches:\\n\\n--- Line 1 ---\\n> PRAI-CAND-027-TERMINAL-SECRET\\n second line", "vulnerable": true } ``` The PoV is local-only. It does not start a server, contact a third-party target, or use real credentials. ## Why This Is Not Intended Behavior This report does not claim that history and terminal helpers should be unable to read legitimate history or terminal logs. The issue is narrower: logical `run_id` and `agent_id` values can escape the configured Dynamic Context base directory. The controls show the intended boundary: - legitimate in-base history remains readable; - legitimate in-base terminal logs remain readable; - the outside `.jsonl` and `.log` files are not under the configured `base_dir`; and - the tools still disclose those outside files through traversal identifiers. The official context reference describes history persistence and terminal logging as filesystem-backed Dynamic Context features. The context security documentation also treats absolute paths, path traversal, and sensitive files as privacy/security risks. Reading files outside the configured context store conflicts with that documented boundary. ## Impact If a PraisonAI application exposes these Dynamic Context tools to untrusted or lower-trust prompts, the lower-trust caller can read files outside the configured context storage when the target file can be reached with the tool-imposed suffix: - `history_*` tools can disclose reachable `.jsonl` files; - `terminal_*` tools can disclose reachable `.log` files; and - cross-run or cross-agent context/history/logs can be disclosed if their path is known or guessable. This can expose conversation history, prompts, terminal output, command logs, tokens, API keys, cloud credentials, operational data, or other secrets stored in JSONL/log files readable by the PraisonAI process. The impact is confidentiality-only in the tested surface. Integrity and availability are not claimed for this report. ## Severity Suggested severity: High. Rationale: - `AV`: applies when an application exposes an agent with these tools over a network chat/API surface. - `AC`: the traversal needs only chosen `run_id` and `agent_id` values. - `PR`: an unauthenticated or public-facing agent endpoint can be exploited without an account. Deployments that require authenticated chat/API access may score this as `PR:L`. - `UI`: the attacker directly supplies the prompt/tool argument to the exposed agent surface. - `C`: conversation history and terminal logs can contain secrets and private operational data. - `I:N/A`: this report demonstrates read-only disclosure. ## Remediation Treat `run_id` and `agent_id` as logical identifiers, not path components. Recommended fixes: 1. Reject absolute paths, path separators, and traversal components in `run_id` and `agent_id`. 2. Build candidate paths, call `.resolve()`, and reject any path that is not under `self.base_dir.resolve()`. 3. Apply the same containment helper to history append/read/search/clear/export and terminal log/read/search/clear/export paths. 4. Prefer opaque server-generated run and agent IDs in tool schemas. 5. Add regression tests for absolute `run_id`, `../` in `run_id`, and `../`
Sources
- GitHub Advisory Database ↗ · first seen 2026-06-18 13:52 UTC
Defaxon links out to the original reporting and never republishes article text.
Correlated events
Computed by the Defaxon correlation engine — linked by shared actors, co-location, and temporal proximity. Scored hypotheses, never causal claims.
No correlated events found in the current window. As more events arrive, connections form automatically.