Back to skill

Security audit

story-review:多视角对抗式审查

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report substantial capabilities beyond its stated story-review purpose, and some files were only partly inspected.

The verdict is suspicious, with medium overall confidence; it is not a claim that the author is malicious. Consider avoiding installation until the skill’s extra editing and persistent tracking capabilities are clearly disclosed and its permissions are understood. Coverage is partial: 19 of 23 components were fully inspected, and four were only partly inspected.

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
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (21)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code implements a punctuation checker/editor, not the declared reviewer-agent workflow. It contains no agent spawning, full/lean review modes, solo-review fallback, or rubric fallback. Its primary behavior is reading local files and writing normalized text back unless --check is specified. Although punctuation checking could support story review, automatic in-place editing is a materially different, undeclared capability. This assessment is limited to the supplied code chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk's primary function is transactional project-state maintenance, not the declared multi-agent review workflow. Its init and commit paths make substantial persistent changes under the project's tracking directory, including archiving legacy files and removing obsolete character views; draft also writes a JSON transaction file. These are substantive capabilities not disclosed by the description, rather than merely internal review implementation details. Although such a tool could support a broader review skill, the supplied code does not establish that connection or implement the described review behavior. No network access or malicious intent is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied module is dedicated to word-count accounting and record integrity. It contains no full/lean review modes, reviewer-agent spawning, solo-review fallback, rubric loading, or review-trigger handling. Although word-count checks could support a broader story-review workflow, this chunk implements a substantial, distinct chapter-length management capability not conveyed by the declared description. Its file access is local; no network activity, subprocess execution, or file modification is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Although prose linting could support a broader story-review workflow, this chunk implements a substantial standalone style gate, not merely orchestration details for the declared review process. Its checks focus on punctuation, stock expressions, sentence structure, and pattern density. No agent-based review or rubric fallback is present. The visible code does not rewrite files, execute subprocesses, or access the network; the mismatch concerns functional scope, not evidence of malicious behavior. Other unsupplied skill components could implement the declared workflow.

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 function is a standalone regex- and heuristic-based prose checker, not the agent-based review workflow described. Such a checker could support a larger review skill, but no integration with that workflow is shown here. Its local file reads and report generation are consistent with text review and are not independently suspicious. No file rewriting, network access, or external command execution is present.

Content

No source excerpt is available for this finding.

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 146)May include surrounding context.

md
node scripts/normalize-punctuation.js --check <正文文件...>

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 150)May include surrounding context.

md
`format` findings;旁白引号外的 `em-dash` 只采用 `check-ai-patterns.js` 的语义改写建议(去重丢弃 `normalize-punctuation.js` 同位置的 `em-dash`;对话引号内被打断或拖音的 `——` 不报错);默认 `--quote-mode ke

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 147)May include surrounding context.

md
node scripts/check-ai-patterns.js --check --fail-on=blocking <正文文件...>

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 150)May include surrounding context.

md
double-hyphen`、`markdown-divider` 归入 `format` findings;旁白引号外的 `em-dash` 只采用 `check-ai-patterns.js` 的语义改写建议(去重丢弃 `normalize-punctuation.js` 同位置的 `em-dash`;对话引号内被

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 151)May include surrounding context.

md
- `check-ai-patterns.js` 的 blocking / advisory 作为 `prose` 候选,按 `style_resolution`、原文功能及 `expression.md` / `narrative-focus.md` / `story-logic.md` 复核,确认造成具体阅读损失后

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 148)May include surrounding context.

md
node scripts/check-degeneration.js --check <正文文件...>

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 152)May include surrounding context.

md
- `check-degeneration.js` 的 blocking(复读/截断/tier1 工程词)作为 S1/S2 `prose` findings(建议“重新生成该段,不是改写”),advisory(tier2 章节/歧义词)作为 S3。

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 · references/tracking-transaction.md (reported line 25)May include surrounding context.

{PYTHON} {当前 skill 根}/scripts/tracking_commit.py init --project {书项目根} --input {书项目根}/.story/work/init.json {PYTHON} {当前 skill 根}/scripts/tracking_commit.py check --project {书项目根} {PYTHON} {当前 skill 根}/scripts/tracking_commit.py draft --project {书项目根} --chapter {N} {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter check --project {书项目根} --chapter {N} [--min-chars {下限} --max-chars {上限}] {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter commit --project {书项目根} --chapter {N} --input {书项目根}/.story/work/第{NNN}章/tracking.json {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter accept-current-length --project {书项目根} --chapter {N} --input {书项目根}/.story/work/第{NNN}章/tracking.json

text

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 · references/tracking-transaction.md (reported line 26)May include surrounding context.

{PYTHON} {当前 skill 根}/scripts/tracking_commit.py check --project {书项目根} {PYTHON} {当前 skill 根}/scripts/tracking_commit.py draft --project {书项目根} --chapter {N} {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter check --project {书项目根} --chapter {N} [--min-chars {下限} --max-chars {上限}] {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter commit --project {书项目根} --chapter {N} --input {书项目根}/.story/work/第{NNN}章/tracking.json {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter accept-current-length --project {书项目根} --chapter {N} --input {书项目根}/.story/work/第{NNN}章/tracking.json

text

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 · references/tracking-transaction.md (reported line 27)May include surrounding context.

{PYTHON} {当前 skill 根}/scripts/tracking_commit.py draft --project {书项目根} --chapter {N} {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter check --project {书项目根} --chapter {N} [--min-chars {下限} --max-chars {上限}] {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter commit --project {书项目根} --chapter {N} --input {书项目根}/.story/work/第{NNN}章/tracking.json {PYTHON} {story-long-write skill 根}/scripts/storyctl.py chapter accept-current-length --project {书项目根} --chapter {N} --input {书项目根}/.story/work/第{NNN}章/tracking.json

text

- `init`:仅在 `_tracking-state.json` 不存在时执行(旧追踪文件自动移入 `追踪/_旧追踪存档/`);执行前读 [tracking-initialization.md](tracking-initialization.md)。

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.

Generic review phrases can activate the fiction-review skill for unrelated review requests.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The activation phrases “审查一下” and “帮我审一下” mean roughly “review this” and “help me review this,” and can readily occur in requests about code, contracts, or other non-fiction material. Although the body identifies novel review as the skill’s purpose, the trigger declaration does not require that context or exclude unrelated reviews.

Content

No source excerpt is available for this finding.

The review skill includes a workflow for initializing persistent novel-authoring state.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

L07 instructs the agent to write initialization JSON, execute tracking_commit.py init, and delete the temporary file; L33 describes generating persistent tracking context from character snapshots, foreshadowing, and timeline events. This initializes an authoring/import workflow rather than performing the multi-perspective review described by story-review, and the manifest does not declare responsibility for creating project tracking state.

Content

No source excerpt is available for this finding.

The review skill includes instructions for mutating authoritative story state and accepting chapters, capabilities not justified by its stated review purpose.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest describes adversarial review with parallel reviewers and solo/rubric fallbacks. This reference instead directs initialization that archives existing tracking files, atomic chapter commits that update authoritative continuity and word-count records and delete working directories, and acceptance of out-of-range chapter lengths with a force override. These are project-management mutations rather than necessary review operations.

Content

No source excerpt is available for this finding.

The review skill includes a punctuation normalizer that overwrites supplied documents by default.

Medium
Category
Excessive Agency
Confidence
92% confidence
Finding

The CLI defaults to modification and directly overwrites supplied files whenever normalization changes their contents, without explicit write consent or a backup. Omitting --check can therefore cause unintended manuscript changes or data loss; the review workflow's explicit --check invocation mitigates normal use but does not make this standalone interface safe by default.

Content

No source excerpt is available for this finding.

The review skill includes project-state initialization and chapter-append capabilities unrelated to conducting a review.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The manifest describes adversarial review with reviewer-agent and rubric fallbacks, but this script can initialize authoritative story state, relocate existing tracking files, and regenerate tracking views (L1255–L1267). It also accepts append transactions that advance the committed chapter (L1048–L1049, L1120–L1121), with both initialization and commits exposed through the CLI (L1695–L1698). These project-lifecycle operations exceed what is needed to review a story, even if updating tracking after review-driven revisions is intended.

Content

No source excerpt is available for this finding.