Back to skill

Security audit

元阁 yotta-skills

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious because scanners report serious weaknesses in how it handles untrusted packages and a risk from repeatedly recommending an unpinned installer; the review is incomplete, so it does not establish that the artifact is safe.

Consider the suspicious verdict and incomplete inspection before installing. The strongest concerns are the unpinned installer examples and the reported risk of executing untrusted code during inspection. Do not treat the disputed alerts as confirmed, but do not treat missing or partial coverage as evidence of safety.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
bin/yotta-skills.js
Finding

Doctor executes untrusted package code even when identity validation fails

Content
View full analysis

Impact

An attacker-provided skill can execute code as the inspecting user, including accessing files and environment credentials or modifying user-owned configuration. A later failure result cannot undo this execution.

T09 · Insecure Skill Coding Practices

Warning
Location
lib/hub-adopt.js
Finding

Unvalidated external skill names escape the Hub adoption directory

Content
View full analysis

Impact

For example, with the default Hub location, name='../injected' can create a skill-controlled sibling directory under ~/.yottaskills when that destination is absent. Existing real directories require force mode before replacement; in that mode they are moved to trash and replaced. This is an out-of-scope directory write, not unrestricted overwrite of arbitrary existing files.

Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (283)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 107)May include surrounding context.

md
6. **判断标准**:技能目录有对应 `SKILL.md` 即视为已装;拿不准 → 视为未装并给提示。

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 395)May include surrounding context.

md
`SKILL.md` frontmatter,生成/更新本地注册表 `~/.yottaskills/registry.json`,数据不出本机。

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 423)May include surrounding context.

md
(对 npm `dist-tags.latest`),按本地 `SKILL.md` 版本对比——无论技能源自哪个安装渠道

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 445)May include surrounding context.

md
`doctor` 是只读自检,不会修改技能目录或注册表。它检查目标技能目录、`SKILL.md`、

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 448)May include surrounding context.

md
后者会按 `SKILL.md` / `skill-manifest.json` 自动识别技能身份。`--json` 返回稳定字段,

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 580)May include surrounding context.

md
包内 `SKILL.md` 与 `skill-manifest.json` 身份一致(slug / package / trust / 版本)、

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · bin/yotta-skills.js (reported line 69)May include surrounding context.

js
ect.keys(state.selected).filter(function (k) { return state.selected[k]; }).length;\n    $('#adoptCount').textContent = String(n);\n    $('#adoptApply').disabled = n === 0;\n  }\n  $('#adoptSearch').addEventListener('input', function (e) { state.adoptFilter = e.target.value; renderAdopt(); });\n  $('#adoptList').addEventListener('change', function (e) {\n    var box = e.target.closest('input[data-slug]');\n    if (!box) return;\n    if (box.checked) state.selected[box.dataset.slug] = true; else delete state.selected[box.dataset.slug];\n    updateAdoptCount();\n  });\n  $('#adoptAll').onclick = function () {\n    ((state.adopt && state.adopt.candidates) || []).forEach(function (c) { if (!c.inHub) state.selected[c.slug] = true; });\n    renderAdopt();\n  };\n  $('#adoptNone').onclick = function () { state.selected = {}; renderAdopt(); };\n  $('#adoptApply').onclick = function () {\n    var slugs = Object.keys(state.selected).filter(function (k) { return state.selected[k]; });\n    if (!s

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · lib/hub.js (reported line 1461)May include surrounding context.

js
// ── 4. ledger(定向清台账;不调 syncHubState) ──────────────────────────
  const state = readHubState(hubDir);
  if (state.skills && state.skills[slug]) {
    delete state.skills[slug];
    writeHubState(hubDir, state);
  }
  const linkState = readLinkState(hubDir);

Unvalidated Output Injection

High
Category
Output Handling
Confidence
90% confidence
Finding

Model output is used without validation or sanitization. Unvalidated output injected into downstream contexts (SQL, shell, HTML) enables injection attacks and arbitrary code execution.

Content

Scanner excerpt · bin/yotta-skills.js (reported line 69)May include surrounding context.

js
\"mono\">' + esc(preview) + '</div>', '开始收编', false, async function () {\n      var r = await api('/api/adopt/apply', { body: { include: slugs } });\n      if (r.error) { toast(r.error); return; }\n      closeModal();\n      var order = { imported: 0, blocked: 1, conflict: 2, skip: 3, error: 4 };\n      var results = (r.results || []).slice().sort(function (a, b) { return (order[a.status] || 9) - (order[b.status] || 9); });\n      $('#adoptResultsWrap').hidden = false;\n      $('#adoptResults').innerHTML = results.map(function (x) {\n        var tone = x.status === 'imported' ? 'ok' : x.status === 'error' || x.status === 'blocked' ? 'danger' : 'warn';\n        return '<div class=\"result\"><span class=\"badge ' + tone + '\">' + esc(x.status) + '</span><b>' + esc(x.slug) + '</b><span>' + esc(x.note || '') + '</span>' + (x.scanVerdict ? '<span class=\"faint\">verdict: ' + esc(x.scanVerdict) + '</span>' : '') + '</div>';\n      }).join('');\n      toast('收编完成:导入 ' + (r.summary ? r.summary

The documented loopback-only server permits binding to non-loopback interfaces.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The server accepts non-loopback bind addresses, and its listen method also permits overriding the address. If exposed to a network, a remote client can retrieve the HTML containing the session token and then invoke destructive APIs; Host, Origin, and Fetch Metadata checks do not authenticate non-browser clients.

Content

No source excerpt is available for this finding.

Scanner-less exceptions can suppress findings despite documentation requiring explicit yotta-verify scope.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

Missing, malformed, or empty scanner lists are treated as unrestricted rather than rejected, so exceptions without explicit yotta-verify scope can remove findings and improve the installation verdict. Version and tree-hash binding constrain affected packages, but they do not enforce the required scanner authorization.

Content

No source excerpt is available for this finding.

verifyEngine can report success without the installation record required by its documented trust model.

High
Category
Excessive Agency
Confidence
97% confidence
Finding

L011-L013 and L128 promise fail-closed rejection when the trusted installation record is missing or its digest does not match. But L134 returns a successful package inspection whenever requireRecord is false, bypassing both record checks; inspectPackage explicitly performs no trust determination according to L074.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 42)May include surrounding context.

md
the registry afterwards, and <code>--reindex</code> re-scans on demand. An optional <code>yotta-skills</code> MCP (on-demand, not resident) exposes <code>list_installed_skills</code> / <code>describe_skill</code> / <code>reindex</code> / <code>route_request</code> / <code>decide_memory</code>; see <code>SKILL.md</code> for the config.
- **No silent writes** - 元阁 only suggests. Installing skills, adding a client <code>mcpServers</code> entry, and writing an agent's global memory each require explicit user confirmation beforehand; declining never blocks the plain CLI workflow.

Boundaries: it only downloads, places, gates and summarizes — it does **not** develop skill content, does **not** bundle any skill body, does **not** use <code>-g</code> global installs, and does **not** silently write host configuration or agent global memory. Besides the target directory, it keeps its registry, snapshots, install evidence and update-check cache under <code>~/.yottaskills</code>.

## Usage cases

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The documented npx command executes an unpinned installer, allowing a compromised future release to run with the user's privileges. This is a CLI supply-chain risk rather than evidence of an unpinned MCP configuration; pinning downloaded skills does not pin the installer.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This npx example does not constrain the executable package version, so a compromised future release could execute with user privileges. A read-only subcommand does not protect against malicious code in the CLI itself.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The command executes a floating version of the installer rather than a reviewed release. A compromised release could execute arbitrary code and distribute malicious skills through the Hub.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The npx invocation leaves the executable version unconstrained, exposing execution to a compromised future package release. The cross-host linking workflow could amplify the consequences, although no malicious implementation is shown.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The panel command starts code from an unpinned npm package. A compromised future release could execute with user privileges regardless of the panel's intended management functions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The self-install example executes an unconstrained npm release before placing a persistent local copy. A compromised release could therefore both run immediately and remain available for later invocation.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Choosing a custom destination does not constrain the package that npx executes. A compromised future installer release could run arbitrary code and place a compromised copy at that destination.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Even this informational command executes an unpinned npm package. Its intended read-only behavior cannot prevent arbitrary execution if the selected package release is compromised.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The list operation still requires npx to execute an unpinned package. The claim that listing requires no network or changes describes the CLI operation, not a guarantee about npx fetching or executing that package.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This installation command executes an unpinned installer with access to the user's agent skill directory. A compromised future release could run arbitrary code or introduce malicious skills, independently of the documented downstream scan.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The arbitrary-directory install example executes a floating installer release. A compromised release would run with the user's filesystem permissions, not merely the permissions implied by the selected destination.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Selecting only specific skills does not pin the executable installer. A compromised future CLI release could run arbitrary code before any selected skill is scanned or installed.

Content

No source excerpt is available for this finding.