Back to skill

Security audit

沙箱环境下的 git 安全操作与仓库恢复

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marked this skill suspicious because scanners report concerns about credential exposure, publishing data, authorization workarounds, and repair steps that could conceal repository differences; the scan was partial, so evidence is incomplete.

Consider avoiding installation until the credential-handling, public-repository, authorization-workaround, and tracking-reference guidance has been reviewed. Skillspector's analysis was partial: the manifest could not be parsed, several references were unresolved, and a bundled image was outside analyzer scope, so the scan does not provide complete coverage.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (8)

Credential-discovery commands print access tokens without warning about exposure in tool output or session logs.

High
Category
Output Handling
Confidence
99% confidence
Finding

The credential-discovery commands print credential-helper secrets and token environment variables into agent-visible output, where session logs or transcripts may retain them. The surrounding workflow also recommends embedding tokens in command-line URLs, increasing exposure through command logging and process inspection even if the remote URL is subsequently cleaned.

Content

No source excerpt is available for this finding.

The repository-creation workflow selects public visibility without warning about subsequent data exposure.

High
Category
Output Handling
Confidence
97% confidence
Finding

The repository-creation example explicitly selects public visibility without requiring the user to approve publication. In a recovery workflow, an agent could follow this example and subsequently push private source code, configuration, or historical secrets to a publicly readable repository.

Content

No source excerpt is available for this finding.

Frames switching credentials as a way to bypass a workflow authorization restriction without user involvement.

High
Category
Prompt Injection
Confidence
95% confidence
Finding

The skill treats an OAuth workflow-scope rejection as a reason to use an existing SSH identity without user involvement. SSH still enforces its own repository permissions, but automatically switching to a potentially broader identity can exceed the authorization intended for the task and enable workflow changes that execute in CI.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

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

Content

Scanner excerpt · SKILL.md (reported line 471)May include surrounding context.

text

2026-09-20 本机实测:SSH **直连即可**(github.com:22 没被墙),且比 https+socks5 还快。
`~/.ssh/config` 里已有 `IdentityFile /c/Users/<user>/.ssh/id_ed25519_gh`,无需额外配置。

**推完要手动修跟踪引用** —— 用 SSH URL 推不会更新 `origin` 的 https 跟踪引用,
于是 `git status -sb` 会显示假的 `ahead N`(按 §3 直写即可):

Reconstructs filtered syntax indirectly to evade a PowerShell security check.

Medium
Category
Prompt Injection
Confidence
93% confidence
Finding

The skill explicitly recommends reconstructing percent characters to get past a PowerShell security filter rather than using an approved mechanism. Although the immediate purpose is benign placeholder detection, teaching an agent to disguise blocked syntax undermines enforcement and can generalize to more dangerous operations.

Content

No source excerpt is available for this finding.

The documented repair for a missing remote-tracking reference substitutes local HEAD for the remote commit, potentially concealing divergence.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The repair writes local HEAD into origin/main without establishing that the remote actually points to that commit. This fabricates synchronization, potentially concealing unpushed changes or remote divergence; subsequent independent remote checks mitigate this risk but do not make the repair correct.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · SKILL.md (reported line 454)May include surrounding context.

bash
TOKEN=$(printf "protocol=https\nhost=github.com\n\n" | git credential fill | sed -n 's/^password=//p')
curl -sS -o /dev/null -D - -H "Authorization: token $TOKEN" https://api.github.com/user \
  | grep -i x-oauth-scopes
# 期望里要有 workflow;只有 gist, read:org, repo 就是缺

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 · SKILL.md (reported line 454)May include surrounding context.

bash
TOKEN=$(printf "protocol=https\nhost=github.com\n\n" | git credential fill | sed -n 's/^password=//p')
curl -sS -o /dev/null -D - -H "Authorization: token $TOKEN" https://api.github.com/user \
  | grep -i x-oauth-scopes
# 期望里要有 workflow;只有 gist, read:org, repo 就是缺