Back to skill

Security audit

AI Native Bootstrap

Security checks for vulnerabilities and agentic risk

Overview

The skill is marked suspicious because scanners report instructions that could authorize repository changes by default and evaluation failures that may be recorded as passes; one referenced path was missing, so the evidence is incomplete.

Consider the suspicious verdict and the two high-confidence instruction and evaluation concerns before installing. The scan covered all 26 listed components but is marked partial because two referenced paths were missing, so the evidence is incomplete; do not treat the clean persistence rating or other expected-for-purpose matches as proof of safety.

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
  • Rogue AgentSelf-Modification, Session Persistence
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (14)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
65% 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 · references/implementation-standard.md (reported line 21)May include surrounding context.

md
2. **保持现状**:不要修改任何与接入无关的业务逻辑;不得自行升级依赖、迁移数据库、重构架构或改动产品文案。
3. **先做审计再改文件**:完成阶段 0,并在项目内 ai-sdlc 文档目录的审计文件中留下事实、证据与风险;如没有阻断条件,继续进行阶段 1–4。审计不等于审批,但低风险的本地接入已由本任务授权。
4. **以可执行为准**:不允许“创建了 check.sh 但里面都是 TODO / echo success”“添加占位测试而没有真实断言”“把未运行写成 PASS”。现有测试条件不足时如实标记 `BLOCKED` 或 `NOT RUN`,不能假造通过。
5. **默认在独立分支或 worktree 完成**:保留用户未提交改动,禁止 `git reset --hard`、`git clean -fd`、强制推送、覆盖已有文件或未经许可的 stash。不能安全隔离时,停止写操作并报告阻断原因。
6. **权限分离**:可以在本地生成变更、运行已有安全测试、写出交付报告;默认不得推送远程、合并 PR、修改生产配置、执行生产部署或操作真实生产数据。若仓库环境/已有审批明确允许推送新分支和创建 PR,可按仓库政策执行,但**不能自行批准和合并**。
7. **完成后提供可审核证据**:最终报告必须给出真实修改文件列表、实际执行命令、退出码、失败项、遗留风险、回滚指令,以及建议的下一步。

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · scripts/_core.py (reported line 137)May include surrounding context.

python
def write_new(base: Path, relative: str, contents: bytes, created_files: list, made_dirs: list):
    dest = safe_destination(base, relative)
    if dest.exists():
        raise Blocked('refusing to overwrite existing file: '+relative)
    missing = []
    p = dest.parent
    while not p.exists() and p != base:

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · tests/test_bootstrap.py (reported line 41)May include surrounding context.

python
self.assertEqual(list(self.root.iterdir()),[])
        self.assertIn('docs/ai-sdlc/audit.md',r.stdout)
    def test_read_only_audit_never_leaks_env_values(self):
        (self.root/'.env').write_text('SENSITIVE_DONT_PRINT=topsecretmarker')
        r=call(BOOT,'audit','--project',self.root)
        self.assertEqual(r.returncode,0,r.stderr)
        self.assertNotIn('topsecretmarker',r.stdout+r.stderr)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · tests/test_bootstrap.py (reported line 45)May include surrounding context.

python
r=call(BOOT,'audit','--project',self.root)
        self.assertEqual(r.returncode,0,r.stderr)
        self.assertNotIn('topsecretmarker',r.stdout+r.stderr)
        self.assertEqual(set(x.name for x in self.root.iterdir()),{'.env'})
    def test_init_non_git_empty_and_verify(self):
        r=call(BOOT,'init','--project',self.root,'--apply')
        self.assertEqual(r.returncode,0,r.stderr)

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The execution instructions enable repository writes by default despite the manifest's default-off write policy.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The reference asserts that the user has already authorized local changes and directs the agent to continue through implementation without further approval. Because it is presented as an authoritative execution specification, this can turn an audit-only request into unauthorized repository or CI modifications despite the skill's explicit opt-in policy; production exclusions limit but do not eliminate the risk.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/_core.py (reported line 65)May include surrounding context.

python
def git(root: Path, *args):
    try:
        p = subprocess.run(['git', '-C', str(root), *args], capture_output=True, text=True, timeout=10, check=False)
        return p.returncode, p.stdout.strip()
    except (FileNotFoundError, subprocess.TimeoutExpired):
        return 127, ''

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/check_runner.py (reported line 65)May include surrounding context.

python
start=time.monotonic()
        try:
            # Output captured, never replayed automatically; may contain credentials.
            proc=subprocess.run(argv,cwd=str(root),stdout=subprocess.DEVNULL,stderr=subprocess.DEVNULL,timeout=args.timeout,check=False)
            status='PASS' if proc.returncode==0 else 'FAIL'
            rc=proc.returncode
            if rc: failures.append(name)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/eval_runner.py (reported line 51)May include surrounding context.

python
for c in enabled:
        begin=time.monotonic()
        try:
            proc=subprocess.run(c['argv'],cwd=str(root),stdout=subprocess.DEVNULL,stderr=subprocess.DEVNULL,timeout=a.timeout,check=False)
            actual=proc.returncode
        except subprocess.TimeoutExpired:actual=124
        except (FileNotFoundError,PermissionError):actual=127

Runner failures can be reported as passing evaluations despite the promise never to fabricate results.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Timeouts and launch failures are converted to numeric results 124 and 127, which are then compared with the case's permitted expected exit code. A missing or non-executable program can therefore produce PASS without running the regression, undermining the evidence-based gate and potentially concealing an untested change.

Content

No source excerpt is available for this finding.

compile() call detected

Medium
Category
Dangerous Code Execution
Confidence
65% confidence
Finding

compile() creates code objects from strings. When combined with exec()/eval(), it enables obfuscated code execution.

Content

Scanner excerpt · scripts/validate_bundle.py (reported line 29)May include surrounding context.

python
for p in ROOT.rglob('*'):
        if p.is_symlink():problems.append('symlink: '+str(p.relative_to(ROOT)))
    for p in (ROOT/'scripts').glob('*.py'):
        try:compile(p.read_bytes(),str(p),'exec')
        except (SyntaxError,UnicodeError) as e:problems.append(f'Python syntax: {p.name}: {e}')
    try:
        skill=(ROOT/'SKILL.md').read_text(encoding='utf-8')

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_bootstrap.py (reported line 16)May include surrounding context.

python
PY=sys.executable

def call(script,*args):
    return subprocess.run([PY,str(script),*map(str,args)],stdout=subprocess.PIPE,stderr=subprocess.PIPE,text=True,timeout=25)

def git(folder,*args):
    return subprocess.run(['git','-C',str(folder),*args],capture_output=True,text=True,timeout=10)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_bootstrap.py (reported line 19)May include surrounding context.

python
return subprocess.run([PY,str(script),*map(str,args)],stdout=subprocess.PIPE,stderr=subprocess.PIPE,text=True,timeout=25)

def git(folder,*args):
    return subprocess.run(['git','-C',str(folder),*args],capture_output=True,text=True,timeout=10)

class BootstrapTests(unittest.TestCase):
    def setUp(self):

Excessive Permissions

Low
Category
Privilege Escalation
Confidence
80% confidence
Finding

Skill requests more permissions than appear necessary for its stated functionality. Review if elevated access is justified.

Content

Scanner excerpt · ACCEPTANCE.md (reported line 36)May include surrounding context.

md
- **Live Agent skill discovery:** NOT RUN; paths follow documented conventions, but must be tested in actual Codex/Claude/OpenCode sessions.
- **WorkBuddy host behaviour (T1/T2/T3):** NOT RUN; only static simulation was performed, with `behavior_verification: not_run`.
- **Platform review:** NOT RUN; only a real upload can confirm. Local validators cannot prove platform acceptance.
- **Remote branch protection, production gate and deploy permissions:** UNVERIFIED; the Skill does not provision or approve any of them.
- **LLM-quality Evals:** NOT RUN; the generic deterministic runner is included but intentionally starts with zero cases, so it fails BLOCKED until genuine cases are approved.
- **Python compatibility:** source targets Python 3.9+; exercised here with Python 3.13, not independently matrix-tested across all minor versions.