Back to skill

Security audit

Multi-Agent Orca

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a legitimate orchestration skill, but it grants very broad worker, terminal, Git, SSH, and background-execution authority that users should review carefully before installing.

Install only if you intentionally want an agent to manage multiple autonomous coding workers and you trust it with your repositories, terminals, provider sessions, and configured remote nodes. Review the auto-accept, SSH remote dispatch, Git/PR cleanup, and Autopilot/night-watch features before use; keep real tokens and personal routing files out of the published package and prefer explicit per-task authorization.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (396)

Dangerous chain: exec() wrapping compile

Critical
Category
Dangerous Code Execution
Confidence
95% confidence
Finding

The code dynamically executes Python source loaded from disk via exec(compile(...)). Although it applies some trust checks (fixed filenames, non-symlink, same UID, not group/world writable), those checks do not eliminate code-execution risk if an attacker can modify those files as the same user, poison the local checkout, or influence the skill package contents. In a worker-orchestration skill, executing helper modules from disk expands the trusted computing base and turns file integrity failure into arbitrary code execution inside the agent context.

Content

Scanner excerpt · scripts/dispatch-profile-adapter.py (reported line 41)May include surrounding context.

python
module = types.ModuleType(name)
    module.__file__ = str(path)
    sys.modules[name] = module
    exec(compile(raw, str(path), "exec"), module.__dict__)
    return module, {"path": str(path), "sha256": hashlib.sha256(raw).hexdigest()}

Dangerous chain: exec() wrapping compile

Critical
Category
Dangerous Code Execution
Confidence
95% confidence
Finding

A dangerous execution chain combines code execution (exec/eval) with a dynamic source (network, encoded data, dynamic import), creating a high-confidence attack vector.

Content

Scanner excerpt · scripts/zcode-orca-launcher.py (reported line 198)May include surrounding context.

python
require(sha(helper_buffer)==expected[0].decode(),"borrowed_helper_changed")
        import types
        module=types.ModuleType("mao_borrowed_worktree");module.__file__=str(helper_path)
        exec(compile(helper_buffer,str(helper_path),"exec"),module.__dict__)
        try:
            module.validate(binding["contract_file"],m["project"],str(wt),m["branch"],req["orca_bin"],
                            req["session"],binding["lock_owner_pid"],os.environ.get("ORCA_TERMINAL_HANDLE", ""),

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · CHANGELOG.md (reported line 730)May include surrounding context.

md
### 技术优化

- 公开技能对 zcode 额度保持"中立合同消费方"定位:适配器只消费 zcode-quota 的输出(stdin/日志/CLI),不直接调用官方接口、不读取或解密 `~/.zcode/v2/credentials.json`。

### 验证

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · CHANGELOG.md (reported line 841)May include surrounding context.

md
### 待办事项

- 真实 GitHub 仓库中“短 Worker PR → 长期集成分支 → 自动清理 Worker head”的外部链路仍为 `NOT_VERIFIED`;当前证据来自 fake-gh 与真实临时 Git 仓。
- Skill Lint 安全扫描为 0 critical / 0 high;Harness 静态审查仍因受精确交付证据约束的 `git push --delete` 报通用 `HFA-011`,并命中 4 个本轮未改旧测试的 `HRA-001`,因此不声明全 Skill Harness 已验证,也不以命令变形规避扫描。候选缺少正式约束追踪合同与签名多轮证据,Instruction Stability 继续为 `NOT_VERIFIED`。

## [2.16.1] - 2026-09-05

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · CHANGELOG.md (reported line 841)May include surrounding context.

md
### 待办事项

- 真实 GitHub 仓库中“短 Worker PR → 长期集成分支 → 自动清理 Worker head”的外部链路仍为 `NOT_VERIFIED`;当前证据来自 fake-gh 与真实临时 Git 仓。
- Skill Lint 安全扫描为 0 critical / 0 high;Harness 静态审查仍因受精确交付证据约束的 `git push --delete` 报通用 `HFA-011`,并命中 4 个本轮未改旧测试的 `HRA-001`,因此不声明全 Skill Harness 已验证,也不以命令变形规避扫描。候选缺少正式约束追踪合同与签名多轮证据,Instruction Stability 继续为 `NOT_VERIFIED`。

## [2.16.1] - 2026-09-05

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · CHANGELOG.md (reported line 1259)May include surrounding context.

md
- PR #84 review BLOCKER 1:liveness gate 不再用 `.result.workerSession`(该字段在真 Orca 响应里不存在 → 永远 DEAD 等于无门槛,会误杀活 worker)。
- PR #84 review BLOCKER 2:不再删 METADATA(导致 dispatch/run_id 丢失,`worker-abandon` 变 unrecoverable leak),保留 METADATA 给 PM 后续清理。
- PR #84 review MAJOR 3:--destroy 路径先 unlink `node_modules` 软链(`[ -L ] && rm -f` 无尾斜杠),不跟随软链误删主仓。
- PR #84 review MAJOR 4:provider-lease release fail-loud(禁止 `>/dev/null 2>&1` 吞错),失败 exit 非 0。
- PR #84 review MAJOR 5:--destroy 路径含 dirty 检查 + git worktree remove,与 `clean-worktree.sh` 同步。
- PR #84 review MAJOR 6:references/13-orca-cli-worker.md §9 + references/14-pm-orchestrate.md 命令清单同步更新(之前文档与 v1 行为矛盾)。

Chaining Abuse

High
Category
Tool Misuse
Confidence
75% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · CHANGELOG.md (reported line 1306)May include surrounding context.

md
- PR #84 review BLOCKER 1:liveness gate 不再用 `.result.workerSession`(该字段在真 Orca 响应里不存在 → 永远 DEAD 等于无门槛,会误杀活 worker)。
- PR #84 review BLOCKER 2:不再删 METADATA(导致 dispatch/run_id 丢失,`worker-abandon` 变 unrecoverable leak),保留 METADATA 给 PM 后续清理。
- PR #84 review MAJOR 3:--destroy 路径先 unlink `node_modules` 软链(`[ -L ] && rm -f` 无尾斜杠),不跟随软链误删主仓。
- PR #84 review MAJOR 4:provider-lease release fail-loud(禁止 `>/dev/null 2>&1` 吞错),失败 exit 非 0。
- PR #84 review MAJOR 5:--destroy 路径含 dirty 检查 + git worktree remove,与 `clean-worktree.sh` 同步。
- PR #84 review MAJOR 6:references/13-orca-cli-worker.md §9 + references/14-pm-orchestrate.md 命令清单同步更新(之前文档与 v1 行为矛盾)。

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill presents itself as an orchestration skill, but the content also authorizes many side-effecting behaviors beyond that simple description, including configuration mutation, authentication flows, remote dispatch, terminal injection, cleanup/destructive actions, and quota or lease management. Even if some individual static-analysis examples appear over-broad, the core issue is real: the declared purpose materially under-describes the operational authority and side effects, which can mislead users into running a much more privileged workflow than expected.

Content

No source excerpt is available for this finding.