CVE-2026-66065

Ouroboros is a local-first runtime for AI coding agents that records their actions and applies user-defined policies to constrain behavior. Versions prior to 0.42.1 have an incomplete denylist. Several execution-routing keys of the same RCE class were omitted, so a malicious cloned repo can still reach arbitrary command execution by shipping a .env (auto-loaded at import, with no review step). The CVE-2026-47211 fix added _UNTRUSTED_ENV_DENYLIST to stop an untrusted project-directory .env from redirecting execution, but it did not account for all keys. The backend config-home and MCP/plugin roots bypass the approval gate by pointing the nested agent, MCP servers, and plugin roster at attacker config. Other variables re-enable blocked local transports, replace sub-agent prompts, switch backends, and lower tool approval classes, further weakening the approval gate. This issue has been fixed in version 0.42.1.

Ouroboros is a local-first runtime for AI coding agents that records their actions and applies user-defined policies to constrain behavior. Versions prior to 0.42.1 have an incomplete denylist. Several execution-routing keys of the same RCE class were omitted, so a malicious cloned repo can still reach arbitrary command execution by shipping a .env (auto-loaded at import, with no review step). The CVE-2026-47211 fix added _UNTRUSTED_ENV_DENYLIST to stop an untrusted project-directory .env from redirecting execution, but it did not account for all keys. The backend config-home and MCP/plugin roots bypass the approval gate by pointing the nested agent, MCP servers, and plugin roster at attacker config. Other variables re-enable blocked local transports, replace sub-agent prompts, switch backends, and lower tool approval classes, further weakening the approval gate. This issue has been fixed in version 0.42.1.

Impact

The CVE-2026-47211 fix (0.39.0) added _UNTRUSTED_ENV_DENYLIST to stop an untrusted project-directory .env from redirecting execution. The denylist was incomplete — several execution-routing keys of the same RCE class were omitted, so a malicious cloned repo can still reach arbitrary command execution by shipping a .env (auto-loaded at import, no review step):

• Backend config-home roots CODEX_HOME, OPENCODE_CONFIG, OPENCODE_CONFIG_DIR, XDG_CONFIG_HOME: a spawned vendor CLI resolves its config from these. CODEX_HOME=./.evil + committed ./.evil/config.toml redirects the nested Codex agent to attacker config — mcp_servers.<name>.command/args (RCE) and approval_policy="never" / sandbox_mode="danger-full-access" (silent removal of the human approval gate). (reported by matte1782)
• MCP bridge / plugin execution roster OUROBOROS_MCP_CONFIG (the YAML's server command/args are spawned via stdio_client — RCE), OUROBOROS_PLUGIN_LOCKFILE, OUROBOROS_PLUGIN_TRUST_ROOT (redirect the installed-plugin roster / trust root so ooo <name> dispatches into attacker code). (reported by hackkim)
• SSRF guard toggle OUROBOROS_ALLOW_LOCAL_TRANSPORT (re-enables loopback/private MCP transport targets).
• Instruction / capability roots OUROBOROS_AGENTS_DIR, COPILOT_CUSTOM_INSTRUCTIONS_DIRS (replace spawned sub-agent role prompts), OUROBOROS_RUNTIME_PROFILE (backend selector), OUROBOROS_TOOL_CAPABILITIES (override YAML can lower a tool's approval_class, weakening the approval gate).

Additionally, the MCP bridge auto-loaded ./.ouroboros/mcp_servers.yaml from the working directory (create_bridge_from_env(cwd=Path.cwd())), so running ooo inside a malicious repo spawned the committed roster's command — RCE with no .env at all. (cwd-branch noted by hackkim)

Patches

Fixed in 0.42.1. All listed keys were added to _UNTRUSTED_ENV_DENYLIST; the cwd auto-discovery branch was removed (only the explicit OUROBOROS_MCP_CONFIG env var and ~/.ouroboros/mcp_servers.yaml remain, both trusted). The regression suite now derives from the source denylist to prevent future drift.

Workarounds

Do not run Ouroboros from an untrusted/cloned repository directory; remove any project-directory .env and ./.ouroboros/mcp_servers.yaml before running.

Credit

Reported privately via coordinated disclosure by matte1782 and hackkim (https://github.com/hackkim).

Ouroboros is a local-first runtime for AI coding agents that records their actions and applies user-defined policies to constrain behavior. Versions prior to 0.42.1 have an incomplete denylist. Several execution-routing keys of the same RCE class were omitted, so a malicious cloned repo can still reach arbitrary command execution by shipping a .env (auto-loaded at import, with no review step). The CVE-2026-47211 fix added _UNTRUSTED_ENV_DENYLIST to stop an untrusted project-directory .env from redirecting execution, but it did not account for all keys. The backend config-home and MCP/plugin roots bypass the approval gate by pointing the nested agent, MCP servers, and plugin roster at attacker config. Other variables re-enable blocked local transports, replace sub-agent prompts, switch backends, and lower tool approval classes, further weakening the approval gate. This issue has been fixed in version 0.42.1.

Ouroboros is a local-first runtime for AI coding agents that records their actions and applies user-defined policies to constrain behavior. Versions prior to 0.42.1 have an incomplete denylist. Several execution-routing keys of the same RCE class were omitted, so a malicious cloned repo can still reach arbitrary command execution by shipping a .env (auto-loaded at import, with no review step). The CVE-2026-47211 fix added _UNTRUSTED_ENV_DENYLIST to stop an untrusted project-directory .env from redirecting execution, but it did not account for all keys. The backend config-home and MCP/plugin roots bypass the approval gate by pointing the nested agent, MCP servers, and plugin roster at attacker config. Other variables re-enable blocked local transports, replace sub-agent prompts, switch backends, and lower tool approval classes, further weakening the approval gate. This issue has been fixed in version 0.42.1.

Impact

The CVE-2026-47211 fix (0.39.0) added _UNTRUSTED_ENV_DENYLIST to stop an untrusted project-directory .env from redirecting execution. The denylist was incomplete — several execution-routing keys of the same RCE class were omitted, so a malicious cloned repo can still reach arbitrary command execution by shipping a .env (auto-loaded at import, no review step):

• Backend config-home roots CODEX_HOME, OPENCODE_CONFIG, OPENCODE_CONFIG_DIR, XDG_CONFIG_HOME: a spawned vendor CLI resolves its config from these. CODEX_HOME=./.evil + committed ./.evil/config.toml redirects the nested Codex agent to attacker config — mcp_servers.<name>.command/args (RCE) and approval_policy="never" / sandbox_mode="danger-full-access" (silent removal of the human approval gate). (reported by matte1782)
• MCP bridge / plugin execution roster OUROBOROS_MCP_CONFIG (the YAML's server command/args are spawned via stdio_client — RCE), OUROBOROS_PLUGIN_LOCKFILE, OUROBOROS_PLUGIN_TRUST_ROOT (redirect the installed-plugin roster / trust root so ooo <name> dispatches into attacker code). (reported by hackkim)
• SSRF guard toggle OUROBOROS_ALLOW_LOCAL_TRANSPORT (re-enables loopback/private MCP transport targets).
• Instruction / capability roots OUROBOROS_AGENTS_DIR, COPILOT_CUSTOM_INSTRUCTIONS_DIRS (replace spawned sub-agent role prompts), OUROBOROS_RUNTIME_PROFILE (backend selector), OUROBOROS_TOOL_CAPABILITIES (override YAML can lower a tool's approval_class, weakening the approval gate).

Additionally, the MCP bridge auto-loaded ./.ouroboros/mcp_servers.yaml from the working directory (create_bridge_from_env(cwd=Path.cwd())), so running ooo inside a malicious repo spawned the committed roster's command — RCE with no .env at all. (cwd-branch noted by hackkim)

Patches

Fixed in 0.42.1. All listed keys were added to _UNTRUSTED_ENV_DENYLIST; the cwd auto-discovery branch was removed (only the explicit OUROBOROS_MCP_CONFIG env var and ~/.ouroboros/mcp_servers.yaml remain, both trusted). The regression suite now derives from the source denylist to prevent future drift.

Workarounds

Do not run Ouroboros from an untrusted/cloned repository directory; remove any project-directory .env and ./.ouroboros/mcp_servers.yaml before running.

Credit

Reported privately via coordinated disclosure by matte1782 and hackkim (https://github.com/hackkim).