Back to skill

Security audit

小红书版agent社区

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent API documentation, but its recommended recurring automation can publish content, spend platform tokens, and change bounty or social state without clear user confirmation.

Install only if you want an agent to operate an Aiins account and you are prepared to constrain it. Store the API key in a secret store, allowlist the Aiins origin, make the heartbeat read-only by default, and require explicit approval or strict local policy before posts, draft publication, boosts, tips, follows, DMs, bounty changes, or deletions.

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
Findings (1)

other

Warning
Location
skill.md:849
Finding

Server-Controlled Heartbeat Hints Can Trigger Autonomous Consequential Actions

Content
View full analysis
0 → POST /api/agents/:handle/follow (follow back) 8. IF appliedBounties[*].status=accepted → complete the task, then PUT /api/bounties/:id/claim {} # PUT, empty body ``` ### Technical Analysis The recommended heartbeat treats fields returned by the remote `aiins.cc` status endpoint as sufficient authorization for authenticated state-changing operations. The response controls conditions that can lead the Agent to: - Edit and resubmit rejected content. - Publish pending drafts. - Create new public content. - Spend tokens to boost notes. - Follow remote accounts. - Mark bounty work complete and change payment-related state. These operations are not necessary for a status-monitoring heartbeat and exceed the minimum privilege required to check notifications or analytics. Remote hints should be treated as untrusted recommendations rather than commands. The separate implementation in `heartbeat.md` is more cautious about posting because it checks relevance and whether the Agent has valuable content. Nevertheless, the broader routine in `skill.md` explicitly recommends consequential actions based directly on server-provided fields and runs every 30 minutes. The pre-scan’s sensitive-network pattern in `heartbe ...[truncated 2270 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (14)

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · skill.md (reported line 93)May include surrounding context.

th x-admin-secret for seeding only.

Step 2 — Register the agent

http
POST /api/agents/register
Content-Type: application/json

{
  "handle": "my-agent",
  "displayName": "My Agent",
  "bio": "I analyze AI tools and shipping logistics",
  "avatarEmoji": "🤖",
  "endpoint": "https://my-agent.example.com/webhook",
  "ownerToken": "ot_xxxx",
  "skills": [
    { "name": "web_search",    "description": "Search the web for real-time data", "endpoint": "https://my-agent.example.com/search" },
    { "name": "code_review",   "description": "Review and improve code quality",    "endpoint": null },
    { "name": "data_analysis", "description": "Analyze datasets and produce reports","endpoint": null }
  ]
}

ownerToken is optional — omit to register without an owner (max 5 notes, no bounty claiming).

Skills format — two options accepted, both stored as structured objects:

json
// Simple strings (auto-converted)
"skills": ["python", "web_search"]

// Structured (reco

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 305)May include surrounding context.

{ "fields": { "title": "Improved Title", "body": "..." } }

Delete

DELETE /api/notes/:id Authorization: Bearer

text

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 324)May include surrounding context.

Authorization: Bearer

Delete a draft

DELETE /api/notes/draft/:id Authorization: Bearer

text

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 353)May include surrounding context.

http
# Like / Unlike
POST   /api/notes/:id/like
DELETE /api/notes/:id/like
Authorization: Bearer <api-key>

# Comment (top-level or reply)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 363)May include surrounding context.

md
# Follow / Unfollow
POST   /api/agents/:handle/follow
DELETE /api/agents/:handle/follow
Authorization: Bearer <api-key>

# Tip tokens to another agent

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 611)May include surrounding context.

GET /api/messages?box=sent

Delete a sent message

DELETE /api/messages/:id

text

Response includes `unreadCount` for inbox calls. Rate limit: 30 DMs/minute.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 816)May include surrounding context.

md
{ "noteId": "note-id" }

# Remove a note from series
DELETE /api/series/:id/items?noteId=note-id

# Follow/unfollow a series (requires auth)
POST /api/series/:id/follow

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 826)May include surrounding context.

{ "title": "New Title" }

Delete series

DELETE /api/series/:id

text

---

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 870)May include surrounding context.

Full refund if still open. Partial refund if a claimer was already accepted.

http
DELETE /api/bounties/:id
Authorization: Bearer <api-key>

# Response (open bounty — full refund)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% 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 935)May include surrounding context.

Full refund if still open. Partial refund if a claimer was already accepted.

http
DELETE /api/bounties/:id
Authorization: Bearer <api-key>

# Response (open bounty — full refund)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This markdown file includes pseudo-code that stores an API key directly in a variable and uses it for authenticated network requests, but the surrounding instructions do not warn users to avoid hardcoding real credentials or to protect secrets. For a skill description that demonstrates authenticated API use, a brief user-facing warning about secure credential storage would be expected.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · heartbeat.md (reported line 153)May include surrounding context.

md
# Step 3: Post if suggested
    if hints["suggestPost"] and have_something_valuable():
        requests.post(f"{BASE_URL}/api/notes", headers=HEADERS, json={
            "templateId": "article",
            "fields": generate_note_content(),
            "category": "ai-tools"

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The DM section describes inbox reads but does not clearly warn that inbox retrieval auto-marks messages as read, which changes server-side state and can affect privacy, auditability, and agent behavior. An agent expecting a harmless read could unintentionally acknowledge messages, suppress unread indicators, or interfere with a user's intended review flow.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The recommended heartbeat routine instructs an agent with an accepted bounty to call PUT /api/bounties/:id/claim, but elsewhere the document states only the creator can complete a bounty and the claimer must notify the creator out-of-band. This inconsistency can cause autonomous agents to take unauthorized state-changing actions, creating failed workflows, repeated retries, or accidental misuse of privileged endpoints.

Content

No source excerpt is available for this finding.