v1.0.0

AgentDeal

SuspiciousClawScan verdict for this skill. Analyzed Apr 30, 2026, 3:33 PM.

Analysis

AgentDeal looks purpose-built for negotiation, but it gives the agent authority to use credentials, make deal commitments, and keep checking/responding in the background without enough declared boundaries.

GuidanceUse AgentDeal only after verifying the service and setting strict limits: keep the agent in `readonly` or `needs_approval` unless you are comfortable with autonomous commitments, store API keys securely, avoid sharing unnecessary sensitive information, and do not enable heartbeat monitoring without clear stop conditions.

Findings (7)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

Abnormal behavior control

Checks for instructions or behavior that redirect the agent, misuse tools, execute unexpected code, cascade across systems, exploit user trust, or continue outside the intended task.

Tool Misuse and Exploitation

SeverityHighConfidenceHighStatusConcern

SKILL.md

`POST /agents/negotiations/{id}/messages` — types: `proposal`, `counter_proposal`, `acceptance`, `rejection`, `clarification`, `agreement`, `handoff`; ... `full` | Decide and commit within constraints

The skill exposes external API actions that can send acceptances and agreements, and it defines a full authority mode that can commit within constraints without requiring per-action human approval.

User impactThe agent could send binding-seeming deal confirmations, accept offers, or make commitments on your behalf if configured too broadly.

RecommendationDefault to `readonly` or `needs_approval`; require explicit human approval before any acceptance, agreement, purchase, legal, medical, real-estate, or financial commitment.

Agentic Supply Chain Vulnerabilities

SeverityLowConfidenceHighStatusNote

metadata

Source: unknown; Homepage: none

The skill depends on the external AgentDeal service, but the registry metadata provides limited provenance for the publisher or project homepage.

User impactYou have less independent information for verifying the service before giving it deal data or credentials.

RecommendationVerify the AgentDeal domain, publisher, and official documentation before using real credentials or sensitive negotiations.

Cascading Failures

SeverityMediumConfidenceHighStatusConcern

references/heartbeat.md

If 15-30 minutes since last AgentDeal check: 1. Check active negotiations for new messages ... 4. Respond to messages or escalate to owner as needed

The heartbeat loop can repeatedly act across active negotiations, so a bad decision or bad input could propagate through ongoing deal communications rather than staying limited to one request.

User impactA mistaken response could be repeated or amplified across multiple negotiation rounds or active negotiations.

RecommendationLimit heartbeat use to specific negotiations, require approval for outbound commitments, set a maximum number of automated responses, and define clear stop conditions.

Rogue Agents

SeverityMediumConfidenceHighStatusConcern

references/heartbeat.md

Add to your heartbeat file: ## AgentDeal (every 15-30 minutes) ... Every 30 minutes when no active negotiations (check for new invites)

The skill instructs users to add persistent periodic activity, including checks even when there are no active negotiations, without a matching metadata declaration or removal instructions.

User impactThe agent could continue monitoring and responding after the initial task, consuming API limits and acting on new invites unless explicitly stopped.

RecommendationOnly enable the heartbeat with explicit consent, document how to disable it, set quiet hours and end dates, and require owner approval before responding to new invites.

Permission boundary

Checks whether tool use, credentials, dependencies, identity, account access, or inter-agent boundaries are broader than the stated purpose.

Identity and Privilege Abuse

SeverityMediumConfidenceHighStatusConcern

SKILL.md

Save the `api_key` and send your human the `claim_url`. Use `Authorization: Bearer ***` for all subsequent requests.

The runtime instructions require creating, storing, and using a bearer API key, even though the registry metadata declares no primary credential or required environment variables.

User impactIf the API key is exposed or mishandled, someone could act as your negotiation agent and access or change negotiation data.

RecommendationDo not paste real tokens into chat; store credentials in a managed secret/config mechanism, rotate keys when exposed, and update the skill metadata to declare its credential requirements.

Sensitive data protection

Checks for exposed credentials, poisoned memory or context, unclear communication boundaries, or sensitive data that could leave the user's control.

Memory and Context Poisoning

SeverityMediumConfidenceHighStatusNote

SKILL.md

**Document everything** — every message, concession, and agreement

The skill intentionally records negotiation history and deal details, which may include sensitive commercial or personal context that can influence later negotiation behavior.

User impactSensitive terms, concessions, constraints, and summaries may be stored and reused during later negotiation steps.

RecommendationAvoid sharing unnecessary secrets, review stored messages and alignment reports, and require owner approval before disclosing personal or confidential information.

Insecure Inter-Agent Communication

SeverityMediumConfidenceHighStatusNote

references/negotiation-guide.md

AgentDeal supports group negotiations: Multiple agents can join a negotiation; Each represents their owner's interests; Alignment reports consider all parties

The skill is explicitly designed for multi-agent negotiation, so deal information is exchanged with other agents and participants through the platform.

User impactInformation you share in a negotiation may be seen by other agents or owners, and their messages may influence the negotiation flow.

RecommendationVerify participants and invite tokens, share only negotiation-appropriate information, and use webhook signature verification when integrating external systems.