Back to skill

Security audit

skill-author

Security checks for vulnerabilities and agentic risk

Overview

The verdict is benign with medium confidence: the reported alerts appear largely tied to security guidance and examples, but scanning was partial, so the evidence is incomplete.

The scanners returned a benign verdict, but confidence is medium and coverage is partial: two referenced files were not fully inspected, and some instruction text and expressions exceeded scanner limits. Consider the verdict a favorable but incomplete assessment; review the referenced guidance, especially shell examples, before following them.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (17)

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 · SKILL.md (reported line 3)May include surrounding context.

md
---
name: skill-author
description: 'Author, harden, and publish agent skills for ClawHub — write skills that conform to ClawHub conventions (frontmatter, structure, naming, versioning, licensing), run pre-publish hardening, and remediate clawhub.ai security-audit findings. Use when creating a skill for ClawHub, preparing a new version to publish, or when given a clawhub.ai/<owner>/<slug>/security-audit page.'
version: 1.2.3
license: BSD-3-Clause
description_zh: "撰写、加固并发布符合 ClawHub 规范与惯例的 agent skill:按 frontmatter/结构/命名/版本/许可约定写作 skill,发布前加固,整改 clawhub.ai 安全审计发现。制作要发布的 skill、准备新版本、或拿到 security-audit 页面时使用。"

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 78)May include surrounding context.

md
Run the checklist in `references/publish-checklist.md`. The ten points:

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 125)May include surrounding context.

md
es a `skill-card.md`; never add one to source. Full commands and polling loop: `references/publish-checklist.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 130)May include surrounding context.

md
- `references/publish-checklist.md` — completeness / publish / verification checklist with exact CLI commands (registry diff, moderation polling, repo liveness

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 98)May include surrounding context.

md
llib.request.urlopen` once so all call sites inherit it. Defense-in-depth; see `audit-patterns.md` §9. |

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 101)May include surrounding context.

md
Full pattern → fix mapping with real examples: `references/audit-patterns.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 108)May include surrounding context.

md
2. **Fix hard findings one by one** — locate, apply the matching pattern from `references/audit-patterns.md`, verify (compile/parse the file, run the skill's ow

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 131)May include surrounding context.

md
- `references/audit-patterns.md` — audit pattern → fix mapping (command injection, data egress, **Bearer credential cross-origin redirect**, **residual server-s

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 · references/audit-patterns.md (reported line 91)May include surrounding context.

md
| Flagged content | Why it's a false positive | Handling |
|---|---|---|
| `sanitize_filename` docstring mentioning `../../../etc/passwd` | Documents **defense against** path traversal; read as "credential file access" | Keep it; note in the report that it is protective code |
| Reading the skill's own workflow JSON | Read as "memory poisoning / context stuffing" | Keep it; normal file loading |
| Docs describing error categories | Read as "abnormal behavior" | Keep it |
| Documented shell snippets (`clawhub ...`, `diff`, `mktemp -d`, `trap ... EXIT`) in a reference file | The package contains no executable code; these are **documented examples the operator runs by hand**. In `SKILL.md` frontmatter, `declared-permissions` factually separates what the *package ships* (docs only) from what an *operator who follows it* will actually do (run `clawhub`, edit files). Read as operator-run examples, not package behavior | State, in `declared-permissions`, exactly what the package ships (Markd

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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/audit-patterns.md (reported line 141)May include surrounding context.

md
### Case: skill-author@1.2.0 (LLM `suspicious`, A.I.G 0 findings)

- **Only one LLM concern**, dimension `environment_proportionality` (file:line `publish-checklist.md`) — the pre-release diff step used `rm -rf /tmp/<slug>-old && mkdir -p /tmp/<slug>-old`, a **predictable fixed temp path**: if pre-existing files lived there, the `rm -rf` would delete them unconditionally.
- A.I.G static layer: **0 findings**. The other `note`-rated dimension (`instruction_scope`) was weak evidence the reviewers already rejected (license wording + security-doc wording tripping scope alerts). Overall verdict `suspicious` only because of the temp-path line.
- **Fix (matching the scanner's own advice):** replace the fixed path with `STALE_DIR=$(mktemp -d "${TMPDIR:-/tmp}/clawhub-diff.XXXXXX")` + `trap 'rm -rf "$STALE_DIR"' EXIT`. `mktemp -d` guarantees a brand-new path each run, so the delete can never hit pre-existing files.
- Lesson: **documented shell snippets are scanned too** — even a harmless "pull-

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% 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/audit-patterns.md (reported line 141)May include surrounding context.

md
### Case: skill-author@1.2.0 (LLM `suspicious`, A.I.G 0 findings)

- **Only one LLM concern**, dimension `environment_proportionality` (file:line `publish-checklist.md`) — the pre-release diff step used `rm -rf /tmp/<slug>-old && mkdir -p /tmp/<slug>-old`, a **predictable fixed temp path**: if pre-existing files lived there, the `rm -rf` would delete them unconditionally.
- A.I.G static layer: **0 findings**. The other `note`-rated dimension (`instruction_scope`) was weak evidence the reviewers already rejected (license wording + security-doc wording tripping scope alerts). Overall verdict `suspicious` only because of the temp-path line.
- **Fix (matching the scanner's own advice):** replace the fixed path with `STALE_DIR=$(mktemp -d "${TMPDIR:-/tmp}/clawhub-diff.XXXXXX")` + `trap 'rm -rf "$STALE_DIR"' EXIT`. `mktemp -d` guarantees a brand-new path each run, so the delete can never hit pre-existing files.
- Lesson: **documented shell snippets are scanned too** — even a harmless "pull-

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% 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/audit-patterns.md (reported line 141)May include surrounding context.

md
### Case: skill-author@1.2.0 (LLM `suspicious`, A.I.G 0 findings)

- **Only one LLM concern**, dimension `environment_proportionality` (file:line `publish-checklist.md`) — the pre-release diff step used `rm -rf /tmp/<slug>-old && mkdir -p /tmp/<slug>-old`, a **predictable fixed temp path**: if pre-existing files lived there, the `rm -rf` would delete them unconditionally.
- A.I.G static layer: **0 findings**. The other `note`-rated dimension (`instruction_scope`) was weak evidence the reviewers already rejected (license wording + security-doc wording tripping scope alerts). Overall verdict `suspicious` only because of the temp-path line.
- **Fix (matching the scanner's own advice):** replace the fixed path with `STALE_DIR=$(mktemp -d "${TMPDIR:-/tmp}/clawhub-diff.XXXXXX")` + `trap 'rm -rf "$STALE_DIR"' EXIT`. `mktemp -d` guarantees a brand-new path each run, so the delete can never hit pre-existing files.
- Lesson: **documented shell snippets are scanned too** — even a harmless "pull-

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/audit-patterns.md (reported line 47)May include surrounding context.

  1. Pin reviewed commits, use git checkout --detach <full-sha>
  2. Verify repos still exist before publishing — git ls-remote fails on auth issues; the GitHub API is more reliable:
    bash
    curl -s "https://api.github.com/repos/<owner>/<repo>/commits?per_page=1" \
      | python3 -c "import json,sys; d=json.load(sys.stdin); print(d[0]['sha'] if isinstance(d,list) and d else d.get('message'))"
    
    Not Found → the repo was renamed/deleted. Search users/<owner>/repos for the new address and note "renamed from X" in the docs.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/audit-patterns.md (reported line 81)May include surrounding context.

md
## 7. Excessive agency / consent

**Pattern**: the skill makes the agent auto-reuse the user's prior data, auto-run batch file operations, or maintain persistent staging directories — with no consent checks.
**Fix**:
- Auto-reusing user data → **scope limit**: same user, same session, time window only; never across users or sessions
- Batch file operations → confirm scope with the user before executing

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · LICENSE (reported line 20)May include surrounding context.

text
this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING BUT NOT LIMITED TO THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · LICENSE (reported line 24)May include surrounding context.

text
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)

External Script Fetching

Low
Category
Supply Chain
Confidence
15% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · references/audit-patterns.md (reported line 47)May include surrounding context.

  1. Pin reviewed commits, use git checkout --detach <full-sha>
  2. Verify repos still exist before publishing — git ls-remote fails on auth issues; the GitHub API is more reliable:
    bash
    curl -s "https://api.github.com/repos/<owner>/<repo>/commits?per_page=1" \
      | python3 -c "import json,sys; d=json.load(sys.stdin); print(d[0]['sha'] if isinstance(d,list) and d else d.get('message'))"
    
    Not Found → the repo was renamed/deleted. Search users/<owner>/repos for the new address and note "renamed from X" in the docs.