Back to skill

Security audit

COE Consensus

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed local quorum-evaluation API that verifies signed votes and does not execute actions or install persistent background behavior.

Install only if you need a local signed-vote quorum evaluator. Run it on loopback or behind a properly hardened reverse proxy, keep bearer tokens and environment configuration restricted to the service account, and remember that its output is a checksum-backed evaluation result, not an automatic action executor or durable decision ledger.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The reported implementation includes privileged security-sensitive behaviors such as signing with private keys, bearer-token authentication, environment-driven security config, and HTTP hardening, none of which are reflected in the declared purpose. Undisclosed key handling and auth logic are especially dangerous because reviewers may not apply the stricter controls required for secret management and exposed services.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The reported implementation includes privileged security-sensitive behaviors such as signing with private keys, bearer-token authentication, environment-driven security config, and HTTP hardening, none of which are reflected in the declared purpose. Undisclosed key handling and auth logic are especially dangerous because reviewers may not apply the stricter controls required for secret management and exposed services.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The reported implementation includes privileged security-sensitive behaviors such as signing with private keys, bearer-token authentication, environment-driven security config, and HTTP hardening, none of which are reflected in the declared purpose. Undisclosed key handling and auth logic are especially dangerous because reviewers may not apply the stricter controls required for secret management and exposed services.

Content

No source excerpt is available for this finding.

Possible Typosquatting: 'uvicorn' resembles popular package 'gunicorn'

High
Category
Supply Chain
Confidence
70% confidence
Finding

Package name closely resembles a popular package, suggesting possible typosquatting. Attackers publish malicious packages with similar names to trick developers into installing them.

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 · README.md (reported line 7)May include surrounding context.

md
`JEP_CONSENSUS_POLICY` maps tenant names to configuration, e.g. `{"team-a":{"policy":"simple_majority","validators":["v1","v2","v3","v4"]}}`. Optional operator fields: `trust_weights`, `threshold`, `f`. Request-supplied weights/thresholds are rejected. Trusted keys authorize proposers; only configured, independently signing validators count.

At most 128 signed events are accepted per evaluation. A V or T event must refer to exactly one earlier same-target judgment and contain `reference_digest=skill.core.event_digest(judgment)`. If an event supplies its own `hash`, it must equal `event_digest(event)`. The signature covers this reference digest and session context. One issuer contributes at most one vote per assertion; conflicting statements disqualify that issuer. Zero confidence contributes no vote. No verification means unresolved. Termination requires the original issuer.

Majority needs at least three distinct confirmations and more than half of configured membership. Weighted quorum needs three distinct confirmations and strictly more than the operator threshold (at least half of total configured weight). BFT quorum needs n >= 3f+1 and n-f confirmations. This is a stateless quorum evaluator, not a distributed BFT protocol or action executor. Re-evaluating the same round cannot multiply voters; downstream consumers must use the round ID for idempotent execution and preserve evidence if they need a durable decision ledger.

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 · SECURITY.md (reported line 34)May include surrounding context.

md
`JEP_CONSENSUS_POLICY` maps tenant names to configuration, e.g. `{"team-a":{"policy":"simple_majority","validators":["v1","v2","v3","v4"]}}`. Optional operator fields: `trust_weights`, `threshold`, `f`. Request-supplied weights/thresholds are rejected. Trusted keys authorize proposers; only configured, independently signing validators count.

At most 128 signed events are accepted per evaluation. A V or T event must refer to exactly one earlier same-target judgment and contain `reference_digest=skill.core.event_digest(judgment)`. If an event supplies its own `hash`, it must equal `event_digest(event)`. The signature covers this reference digest and session context. One issuer contributes at most one vote per assertion; conflicting statements disqualify that issuer. Zero confidence contributes no vote. No verification means unresolved. Termination requires the original issuer.

Majority needs at least three distinct confirmations and more than half of configured membership. Weighted quorum needs three distinct confirmations and strictly more than the operator threshold (at least half of total configured weight). BFT quorum needs n >= 3f+1 and n-f confirmations. This is a stateless quorum evaluator, not a distributed BFT protocol or action executor. Re-evaluating the same round cannot multiply voters; downstream consumers must use the round ID for idempotent execution and preserve evidence if they need a durable decision ledger.

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 15)May include surrounding context.

md
`JEP_CONSENSUS_POLICY` maps tenant names to configuration, e.g. `{"team-a":{"policy":"simple_majority","validators":["v1","v2","v3","v4"]}}`. Optional operator fields: `trust_weights`, `threshold`, `f`. Request-supplied weights/thresholds are rejected. Trusted keys authorize proposers; only configured, independently signing validators count.

At most 128 signed events are accepted per evaluation. A V or T event must refer to exactly one earlier same-target judgment and contain `reference_digest=skill.core.event_digest(judgment)`. If an event supplies its own `hash`, it must equal `event_digest(event)`. The signature covers this reference digest and session context. One issuer contributes at most one vote per assertion; conflicting statements disqualify that issuer. Zero confidence contributes no vote. No verification means unresolved. Termination requires the original issuer.

Majority needs at least three distinct confirmations and more than half of configured membership. Weighted quorum needs three distinct confirmations and strictly more than the operator threshold (at least half of total configured weight). BFT quorum needs n >= 3f+1 and n-f confirmations. This is a stateless quorum evaluator, not a distributed BFT protocol or action executor. Re-evaluating the same round cannot multiply voters; downstream consumers must use the round ID for idempotent execution and preserve evidence if they need a durable decision ledger.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill references environment-based configuration (JEP_CONSENSUS_POLICY) but declares no explicit tool scope or permissions boundary in the skill manifest. That creates an authorization ambiguity: a caller or host may expose broader environment access than intended, increasing the chance of secret/config leakage or misuse of privileged runtime settings.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
98% confidence
Finding

The module docstring states "COE Consensus Skill v1.0.0" while the code sets version to "2.0.0". This is an active contradiction between inline documentation and the actual version metadata, which can mislead consumers about the skill they are running.

Content

No source excerpt is available for this finding.