Back to skill

Security audit

summarize-pro

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with high confidence because scanners report undeclared file changes and delegation, alongside gaps between its advertised summarizing checks and their reported capabilities.

Consider avoiding installation until the documentation-editing behavior, reported risk of reference-file deletion, and automatic delegation are clarified. Scanner coverage is incomplete: some referenced files were missing and the binary icon was outside analyzer scope, so the scan cannot establish that the artifact 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (19)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The implementation substantially supports the declared offline utility-toolchain role, and no network access or unrelated execution behavior is evident. However, its primary operation is heuristic extraction and template generation rather than the advertised intent-driven summarization. More importantly, unsupported statements without recognized dates or numbers can pass citecheck, so it cannot establish the advertised sentence-level fidelity. The broken selfcheck command and source-independent deadline further contradict the claim that strict non-fabrication and completeness are runnable guarantees. The claimed 22-case golden set cannot be verified from this supplied file alone.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code's primary purpose is documentation restructuring, not faithful summarization or any of the declared verification and rendering functions. It performs persistent writes to two hard-coded local documentation files. Although it uses only the standard library and operates locally, those properties alone do not align its behavior with the declared purpose. No summarization, citation checking, commitment grading, or other advertised functionality is implemented in this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code's primary function is relocating documentation sections, not processing source material into faithful summaries. It reads and overwrites SKILL.md and creates or appends to references/boundaries.md and references/examples.md. Although it is offline and uses only the standard library, those shared implementation properties do not account for its materially different purpose and undeclared documentation-modification behavior. This assessment applies to the supplied chunk, not to any unseen toolkit.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The reported unconditional removal of render.md when its merge destination is absent is a concrete data-loss defect, not merely a purpose mismatch. If the maintenance utility is run against the configured installation, reference content can be deleted without being preserved, compromising the skill's documentation integrity.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied chunk's primary purpose is reorganizing skill documentation, not summarizing source material or implementing the declared verification toolchain. Its persistent edits to SKILL.md and reference files are not described by the declared purpose. Although it could be a development-time maintenance utility, the supplied code does not connect those edits to any declared runtime capability. This assessment concerns this chunk only; it does not establish whether the advertised summarization tools exist elsewhere.

Content

No source excerpt is available for this finding.

Citecheck can certify unsupported statements as PASS despite the manifest's sentence-level traceability guarantee.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

Citecheck checks only selected date and numeric tokens against substrings anywhere in the source, not whether the source supports each statement. Unsupported statements containing no recognized tokens receive PASS with zero checks, while fabricated claims reusing source numbers can also pass. This creates an integrity-check bypass that is especially consequential in a skill advertising sentence-level traceability for legal, medical, personnel, and incident summaries.

Content

No source excerpt is available for this finding.

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 script rewrites the skill's own instructions and reference documents, an undeclared maintenance capability unrelated to summarizing user content.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The manifest describes summarization and supporting offline validation tools, not modification of the skill's instruction files. Using the hard-coded paths at L11–L12, this module unconditionally moves sections of SKILL.md into reference files and overwrites SKILL.md; these operations also execute upon import because they are at module scope.

Content

No source excerpt is available for this finding.

The script rewrites the skill's own instructions and reference documentation rather than processing documents for summarization.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

Documentation maintenance alone is benign, but this implementation removes render.md whenever it exists, even if render-and-cli.md does not exist and no merge occurred. Running or importing the script against the configured installation can therefore destroy reference content without preserving it.

Content

No source excerpt is available for this finding.

The script rewrites the skill's own manifest and reference documentation rather than processing summarization inputs.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

L38–L41 append a section to references/ecosystem.md and replace its contents in SKILL.md; L54 overwrites SKILL.md at the hard-coded installation path defined in L08–L09. This is a skill-package maintenance capability, not an implementation requirement of the declared summarization, checking, redaction, validation, or rendering tools.

Content

No source excerpt is available for this finding.

The script rewrites the skill's own instructions and reference files, an undeclared documentation-maintenance capability unrelated to summarizing user content.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The top-level loop reads the hard-coded SKILL.md path, moves its boundary/exception and input/output sections into reference files, and overwrites SKILL.md with pointers. These are modifications to the skill's governing documentation, not summary generation or any of the declared extract, citecheck, selfcheck, redact, validate, or render operations.

Content

No source excerpt is available for this finding.

The handoff configuration expands faithful summarization into automatic debugging and content-production delegation.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The configuration directs the agent to delegate work and pass source excerpts to other skills merely because a meeting contains a matching task. Meeting content is not user authorization to start debugging or content production, so crafted source material could trigger unintended downstream processing and disclosure to additional skill contexts.

Content

No source excerpt is available for this finding.

The routing instructions expand a summarization skill into automatic delegation to debugging and content-ideation skills.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The instructions mandate handing tasks and excerpts to another skill when matching subjects occur in the summarized material. This crosses from summarizing a document into acting on its contents without an explicit user request, allowing untrusted meeting text to induce unwanted delegation.

Content

No source excerpt is available for this finding.

The personal view invents a same-day response deadline absent from the source.

Medium
Category
Excessive Agency
Confidence
100% confidence
Finding

The personal view unconditionally inserts a request that unconfirmed assignees respond today, regardless of whether the source establishes that deadline. Because the message is presented as ready to send to a group, users may distribute an invented operational requirement as part of a faithful meeting summary. The effect is bounded to generated text, but directly undermines the skill's no-fabrication guarantee.

Content

No source excerpt is available for this finding.

Dynamic import via __import__()

Medium
Category
Dangerous Code Execution
Confidence
75% confidence
Finding

Dynamic import() can load arbitrary modules at runtime, bypassing static analysis and potentially importing malicious code.

Content

Scanner excerpt · tests/run_golden.py (reported line 70)May include surrounding context.

python
if src != exp.get("source"):
            fails.append("来源识别 %s,期望 %s" % (src, exp.get("source")))
        norm = sp.normalize_source(text, src)
        n_spk = len(__import__("re").findall(r"(?m)^\S{1,12}\s*[::]", norm))
        if n_spk < exp.get("speaker_lines_min", 0):
            fails.append("折行后说话人行 %d,期望 ≥%d" % (n_spk, exp.get("speaker_lines_min", 0)))
        return (not fails), fails

Dynamic import via __import__()

Medium
Category
Dangerous Code Execution
Confidence
75% confidence
Finding

Dynamic import() can load arbitrary modules at runtime, bypassing static analysis and potentially importing malicious code.

Content

Scanner excerpt · tests/run_golden.py (reported line 194)May include surrounding context.

python
body = sp.build_skeleton(fmt, facts)

    if exp.get("skeleton_has_numbered_lines") and not __import__("re").search(r"^\d+\.\s", body, __import__("re").M):
        fails.append("骨架缺少编号行")
    if "bullets_count_between" in exp:
        import re

Dynamic import via __import__()

Medium
Category
Dangerous Code Execution
Confidence
75% confidence
Finding

Dynamic import() can load arbitrary modules at runtime, bypassing static analysis and potentially importing malicious code.

Content

Scanner excerpt · tests/run_golden.py (reported line 194)May include surrounding context.

python
body = sp.build_skeleton(fmt, facts)

    if exp.get("skeleton_has_numbered_lines") and not __import__("re").search(r"^\d+\.\s", body, __import__("re").M):
        fails.append("骨架缺少编号行")
    if "bullets_count_between" in exp:
        import re

Standalone triggers such as “timeline,” “key points,” and “action items” do not distinguish summarization from other tasks.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

The trigger list includes generic terms without requiring a request to summarize existing source material. These terms can also occur in requests to create a project timeline or propose action items, potentially activating this source-faithful summarizer for an unrelated task.

Content

No source excerpt is available for this finding.

The routing cues for `wechat-viral-topic` are broad enough to match unrelated writing requests.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

The “Reach for it when” entry includes “what should I write this week” and “headline” without restricting them to WeChat content planning. These cues could route general writing or headline requests to this specialized skill; the skill name alone does not establish an explicit activation constraint.

Content

No source excerpt is available for this finding.