Back to skill

Security audit

ci-long-red-triage

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious because scanners report unsafe cleanup of a fixed temporary directory and instructions that could change or publish repository workflows; no clear malicious intent is reported.

Consider whether you are comfortable with the skill's reported directory cleanup and repository workflow or publishing instructions before installing. The scan is partial: although both listed components were inspected, a local path-like reference could not be resolved, so evidence is incomplete. The other scanner completed with no findings, but that does not establish that the skill is safe.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (5)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
97% confidence
Finding

The command recursively deletes /tmp/fh before using it as a clean test HOME, without checking whether its existing contents belong to this task. This is the same bounded data-loss issue identified by the SQP-2 finding, not evidence of malicious intent or filesystem-wide deletion.

Content

Scanner excerpt · SKILL.md (reported line 58)May include surrounding context.

md
复现 CI 条件:

    rm -rf /tmp/fh && mkdir -p /tmp/fh && HOME=/tmp/fh go test -count=1 ./...

干净 HOME 下全过,才算真的修好。

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 · SKILL.md (reported line 58)May include surrounding context.

md
复现 CI 条件:

    rm -rf /tmp/fh && mkdir -p /tmp/fh && HOME=/tmp/fh go test -count=1 ./...

干净 HOME 下全过,才算真的修好。

The CI reproduction command recursively deletes a fixed directory without warning that existing contents will be lost.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The reproduction command unconditionally removes the predictable directory /tmp/fh without establishing that the current task created or owns its contents. Creating a clean HOME is appropriate for this CI-triage task, but a preexisting directory could contain unrelated files that would be irreversibly deleted.

Content

No source excerpt is available for this finding.

L64 directs workflow changes for infrastructure failures despite the manifest explicitly excluding code changes for them.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The manifest at L03 states that infrastructure failures are not code changes, but L64 instructs the agent to add continue-on-error: true to steps affected by storage quotas or notification failures. This changes repository workflow behavior to tolerate an infrastructure problem rather than merely diagnosing and reporting it, even though the instruction preserves hard failures for core gates.

Content

No source excerpt is available for this finding.

The workflow directs pushes to remote main without warning that this publishes commits and modifies the shared branch.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow makes pushing local main directly to remote main part of completing CI triage, without checking publication authorization or whether the repository requires a pull request. Although publishing fixes is related to the task and the push is not forced, following this default could publish unintended commits or trigger deployments when permissions and branch protections allow it.

Content

No source excerpt is available for this finding.