T01 · Skill Instruction Hijacking
- Location
SKILL.md:583- Finding
Remote Content Is Elevated to System-Level Instructions
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill is a real cloud identity integration, but it gives a remote service ongoing influence over agent behavior and may upload conversation-derived data without clear per-use consent.
Install only if you are comfortable using Live Neon as a third-party identity store for agents. Treat fetched prompts as untrusted configuration unless a human reviews them, avoid approve-all for beliefs or responsibilities, do not enable heartbeat or scheduled syncing without explicit administrative approval, and redact secrets, personal data, source code, and confidential business context before using observation or source-sync features.
SKILL.md:583Remote Content Is Elevated to System-Level Instructions
SKILL.md:617Persistent Identity Updates and Recurring Heartbeat Can Preserve Poisoned State
SKILL.md:332Conversation-Derived Observations and User Quotes Are Proactively Uploaded
SKILL.md:703Security Documentation Misrepresents the Relevant Data Boundary
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Register and get your token in one call — no signup form, no email required:
curl -s -X POST https://persona.liveneon.ai/api/register \
-H "Content-Type: application/json" \
-d '{"org_name": "My Org"}' | jq .
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Optional: add email for account recovery later:
curl -s -X PATCH "$LIVE_NEON_BASE/account" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{"email": "you@example.com"}'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
API call:
curl -s -X POST "https://persona.liveneon.ai/api/register" \
-H "Content-Type: application/json" \
-d '{"org_name": "My Org"}'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
API call:
curl -s -X POST "$LIVE_NEON_BASE/pbd/process" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{"agentId": "AGENT_ID"}'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
API call:
curl -s -X POST "$LIVE_NEON_BASE/pbd/process" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{"agentId": "AGENT_ID"}'
The /ln review command documents an --approve-all mode that can automatically approve all pending beliefs and responsibilities. Since approved items feed the agent's future system prompt and behavior, mass approval removes human scrutiny and can let inaccurate, overbroad, or manipulated identity changes propagate persistently.
| Argument | Required | Description |
|---|---|---|
| `agentId` | yes | Agent UUID/slug |
| `--approve-all` | no | Auto-approve all pending items |
| `--bulk` | no | Use bulk API for batch operations |
**API calls:**
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
API calls:
# Fetch pending beliefs
curl -s "$LIVE_NEON_BASE/beliefs?agentId=$AGENT_ID&status=pending" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN"
# Fetch pending responsibilities
This endpoint posts free-form observations and source quotes that can include user corrections or interaction excerpts to a remote service. In context, the risk is not that external transmission exists, but that the content can include sensitive conversation data and can later shape persistent identity state if users are not adequately informed.
Single observation:
curl -s -X POST "$LIVE_NEON_BASE/observations" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
API calls:
# Group consensus
curl -s -X POST "$LIVE_NEON_BASE/groups/$GROUP_ID/consensus" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN"
# Org consensus
The manifest description says the skill manages identities, syncs content, runs discovery, reviews beliefs, and builds prompts, but the body also documents account registration, agent creation, and content-source creation. Those resource-creation capabilities materially change the risk profile because invoking the skill can create persistent remote accounts and assets, yet this is not disclosed up front.
This call creates a new remote agent resource, which is a persistent state-changing action not reflected in the top-level manifest description. The danger is primarily lack of transparent disclosure and user awareness that invoking the skill can provision durable remote entities.
curl -s -X POST "$LIVE_NEON_BASE/agents" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{"name": "Lead Engineer", "job_title": "Senior Backend Engineer"}' | jq .
This call creates a remote content source and configures ingestion from an external domain, which can expand what data the platform collects and processes. Because it is a state-changing integration setup step, it should be prominently disclosed and gated with consent.
curl -s -X POST "$LIVE_NEON_BASE/content-sources" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Approve them:
curl -s -X PATCH "$LIVE_NEON_BASE/beliefs/bulk" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{"updates": [
The proactive-trigger guidance encourages automatic use of /ln observe, /ln sync, and /ln discover after user corrections or significant work, but it does not clearly warn that conversation-derived content and linked-source content may be sent to a third-party service. That creates a privacy and data-governance risk because sensitive user inputs could be uploaded without explicit, situational consent.
The heartbeat section recommends recurring automated POSTs of accumulated observations plus content syncing and discovery, but it does not prominently warn that this continuously sends potentially sensitive data to a remote platform. Because it is scheduled and ongoing, the exposure can be broader and harder for users to notice or control.
This scheduled heartbeat POST sends accumulated observations to the remote service on a recurring basis, potentially without a fresh user prompt each time. In combination with the automation guidance, it increases the chance of continuous leakage of sensitive interaction summaries or organizational data.
# 1. Report observations accumulated since last heartbeat
curl -s -X POST "$LIVE_NEON_BASE/observations" \
-H "Authorization: Bearer $LIVE_NEON_TOKEN" \
-H "Content-Type: application/json" \
-d '{
The manifest does not mention that the skill can submit observations about conversations or agent behavior that may later influence the agent's stored identity. This omission reduces informed consent because users may not realize their interactions can be turned into durable remote state affecting future behavior.
The manifest omits consensus promotion features that can elevate derived beliefs or responsibilities from one agent to group- or organization-level state. That broader scope increases impact because changes can affect multiple agents beyond the originally targeted one.