Back to skill

Security audit

Prd Implementation Ready Skills

Security checks for vulnerabilities and agentic risk

Overview

This is a Markdown-only PRD-writing skill with broad but disclosed guidance, and I found no code execution, credential access, persistence, or hidden behavior.

Install this if you want rigorous PRD/specification assistance. Review generated assumptions before treating them as final, especially for money movement, permissions, lifecycle states, and cross-system behavior, and be aware that the skill is heavily bilingual Chinese/English.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (25)

Memory Manipulation

High
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · SKILL.md (reported line 67)May include surrounding context.

md
- For every key business object with a lifecycle, draw a Mermaid `stateDiagram-v2` showing object status changes, allowed transitions, rollback/reopen paths, terminal states, and which flow step triggers each transition.
- Mark convergence points explicitly: which flow depends on which object status, which data is written back, which downstream flow is unlocked/frozen, and which exceptions interrupt another flow.

Diagrams clarify structure; they do not replace rules. After each diagram, still write the trigger, preconditions, actors, input/output, persisted object, status changes, exception branches, downstream impact, and testable acceptance criteria.

## Main File Navigation Map

Memory Manipulation

High
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.

Content

Scanner excerpt · references/necessity-gate.md (reported line 28)May include surrounding context.

md
Do not add a new:

- business object if an existing object plus a clear state/field covers the need;
- status if it is only a display label, sub-status, reason, flag, or Tab filter;
- page/Tab if a filter, section, drawer, or detail area carries the job clearly;
- field if no user decision, calculation, rule, display, audit, or integration uses it;

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 · README.md (reported line 13)May include surrounding context.

md
## What This Skill Helps With

Use this skill when a product requirement needs to become clear enough for an engineer or coding agent to implement without asking business questions.

It is strongest for:

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation criteria are extremely broad and can match many normal PRD, spec, workflow, and review tasks. Over-broad routing is risky because it can cause this skill to take control in situations where its heavyweight process, defaults, and document-shaping behavior are not appropriate, increasing the chance of unintended business-rule fabrication or scope takeover.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
82% confidence
Finding

This section encourages the agent to treat ambiguous user shorthand like 'you decide' or 'best practice' as authorization to close business decisions, which can let the model autonomously invent requirements. In a PRD-generation context, that is dangerous because fabricated defaults may become authoritative implementation inputs, especially for permissions, money flow, states, and idempotency rules.

Content

Scanner excerpt · SKILL.md (reported line 241)May include surrounding context.

md
- Before locking flows, run a business possibility scan for SaaS/multi-tenant, multiple-role, cross-organization, migration, compatibility, and external-system variants. Close each as supported, blocked, later, or a recommended default.
- Treat short answers ("加 / 不需要 / 你来 / 按最佳实践") as decisions to digest, not missing context.
- Use the minimum structure that makes the product buildable, testable, and understandable. Do not add entities, statuses, pages, components, tables, or prose just to look complete.
- **Formula Operand Semantics Gate:** Before writing any formula with guarantee/保底, max/min/取大取小, cap/floor, tier, threshold, rate, base amount, detail-calculated amount, or adjustment, define every operand's business meaning, source, entity scope, period, eligibility, comparison basis, rounding, and boundary behavior. A formula is not complete until a tester can compute it from source records without asking "保底是什么保底 / guarantee of what?"
- **Finance Fund-Flow Gate:** Before finalizing any balance, credit, support fund, prepayment, recharge, refund, freeze, occupation, release, write-off, profit sharing, or payment rule, prove the fund loop closes: source pool, usable amount, deduction/occupation order, mixed funding behavior, ledger effect, source trace, closure path, reversal, page visibility, and cross-module side effects. Do not invent a single-source, single-period, one-to-one, or audit-heavy path just because it is easy to write.
- **Settlement and Payment Closure Gate:** Before finalizing cost inclusion, reconciliation, billing, payment, or profit sharing, prove the state and money loop closes: audit status, payment status, locked cost period, duplicate prevention key, preserved detail dimensions, platform service fee, profit-sharing funds waterfall, payment order origin payload, payment completion side effects, centralized credit management, and void/rollback. A neat formula is not enough if the business event cannot be operated, paid, reversed, and audited
...[truncated 25 chars]

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 · SKILL.md (reported line 262)May include surrounding context.

md
- About to lock口径 / write defaults **without** first reading existing project docs (skipping research under time pressure).
- About to model a finance action as "choose one balance source" when partial insufficiency could require mixed funding, or as "automatic audit/payment" when the business only needs a deterministic ledger adjustment.
- About to tie a refund/release to the same receipt/source detail without proving that the occupied pool and the source trace have a one-to-one pairing.
- About to call a supplier pre-recharge or operating payment an internal ledger registration without checking whether real external payment happens.
- About to equate "finance-side" with the whole platform-side without defining organization scope, operator role, and whether platform can see all subjects.
- About to collapse detail dimensions into a single total before deciding which downstream calculation uses summary calculation and which report needs dimension analysis.
- About to mention invoice, payment trace, exception record, allocation detail, or any child object only inside an operation row without its own lifecycle/page/visibility.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The default prompt is broadly framed around clarifying any rough requirement into an implementation-ready PRD, without clear trigger boundaries, domain limits, or user-consent conditions. This can cause over-broad invocation and unintended use in contexts where the skill may inject strong assumptions or rewrite requirements beyond what the user explicitly requested.

Content

No source excerpt is available for this finding.

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/coverage-matrix.md (reported line 117)May include surrounding context.

md
Rules:

- Do not invent the authoritative anchor. A created time, sync time, approval time, or receipt time is only valid when the business object truly uses it as the anchor.
- If partial insufficiency can happen, decide whether users/system can combine multiple resources or sources; do not force a single-source rule without confirmation.
- Separate pool from trace. The current available pool controls use; source trace explains history. A one-to-one source pairing is required only when the business says so.
- Define reserve/release even when there is no money: stock reservation, coupon lock, appointment slot hold, content permission grant, quota occupation, task capacity allocation.
- Show the role perspective: the acting user, owner, approver, counterparty, platform/admin, or tenant may need different visibility and disabled reasons.

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · references/data-flow.md (reported line 41)May include surrounding context.

md
Use this for any money or non-finance resource: inventory, coupon, quota, points, capacity, task slots, asset usage, content permission, membership rights, API usage, visibility, approval authority. A resource/value flow is data flow plus business conservation: the system must show where the resource comes from, when it is reserved, how it is allocated, when it is consumed, and how it is released.

| Resource / Value | Source | Authoritative Anchor | Owner / Pool | Trace Source | Eligibility | Reserve / Occupy Trigger | Allocate / Combine Rule | Partial Insufficiency Rule | Consume / Settle Trigger | Release / Reverse Rule | Role Perspective / Counterparty Visibility | Confirm |
|---|---|---|---|---|---|---|---|---|---|---|---|---|

Rules:

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file includes untranslated Chinese text within otherwise English guidance, which effectively imposes a mixed-language reading requirement on users. Because the file does not state that it is intended for bilingual users or a Chinese-specific context, this can violate a language/locale policy requiring user choice or justified locale constraints.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Lines in this section contain Chinese terms such as '正文' and '同步更新 / 自动回滚 / 联动处理' embedded in operational instructions. Without an explicit statement that the skill targets a Chinese-language environment, the file forces a specific locale on readers rather than offering a language choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This section references labels like '主文件', '端', and '跨端' in natural-language instructions. Those terms impose a specific language/locale expectation on users, and the file does not provide opt-in, translation, or a documented reason for the locale restriction.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The red-flag section includes several Chinese-language example phrases, which makes the policy guidance dependent on understanding a specific language. Since no locale scope or user language choice is stated, this is a natural-language policy concern under the language/locale rule.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file contains extensive Chinese-language examples and prescribed wording, such as lines L011-L016 and L063-L104, but does not indicate that language selection is optional or limited to a specific region or audience. Under the language/locale policy, forcing a specific language without user opt-in can be a policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instruction requires final confirmation using specific Chinese phrases: "继续 / 按这个写 / 没问题 / 可以开始". This creates a locale/language constraint in the skill's natural-language behavior without indicating that other languages are acceptable or that the user can choose their preferred language.

Content

No source excerpt is available for this finding.

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 12)May include surrounding context.

text
permit persons to whom the Software is furnished to do so.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
74% confidence
Finding

The manifest presents the skill name and description in Chinese, but the default prompt is fixed in English. This can create an implicit language-policy mismatch if the skill forces responses or operation in a language the user did not explicitly choose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The markdown contains a required literal status marker, "数据流待确认," as the prescribed wording to use. It also includes several Chinese-only example phrases later in the document, which can impose a specific language/locale in a generally applicable skill reference without stating that the skill is region-specific or offering an alternative language option.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
74% confidence
Finding

The file instructs users to apply the gate when a PRD says phrases such as "按总额", "不按车/不按明细", and "汇总后传导." In a generic reference document, this language-specific phrasing can violate locale policy if the skill is expected to work across languages and no opt-in or region-specific justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The file says that if the PRD uses phrases like "系统生成 / 自动生成 / 自动统计 / 自动同步," certain fields must be named. This is another instance of enforcing examples in a single language in a general-purpose skill reference, with no stated locale limitation or user language choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
72% confidence
Finding

The instruction to avoid saying "同步更新 / 自动回滚 / 联动处理" embeds policy-relevant guidance in Chinese-only phrasing. Without an explicit statement that the document is intended for Chinese-language use, this can amount to a locale constraint imposed without user opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document includes direct Chinese phrasing such as "按规则处理" in otherwise English guidance, which imposes a language assumption without stating that the skill is intended only for Chinese-language PRDs. Similar mixed-language requirements recur later, but this line is a clear example of the policy issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file mandates handling phrases like "本月", "当月", and "账期内" as part of core workflow, but does not explain that the skill is region- or language-specific. That creates a natural-language locale constraint that is not optional for users.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file contains natural-language instructions and examples that require Chinese output elements, such as 00_PRD总览_全局规则_主业务流程. Similar mixed-language requirements continue throughout the document, but there is no explicit opt-in, language-choice mechanism, or justification that the skill is intended only for a Chinese-language or region-specific workflow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

Lines L190-L192 introduce a default UI recommendation specifically for 'B端后台' and present the guidance entirely in Chinese. This imposes a language/locale preference in natural-language content without user opt-in or an explicit justification that the skill is only for a Chinese-language context.

Content

No source excerpt is available for this finding.