Kyvvu

Policy · critical

External content taint: high-impact actions require a fresh gate

If the agent has fetched external content earlier in this task, any subsequent high-impact action (code execution, outbound message, destructive or mutating resource call) must be IMMEDIATELY preceded by a step.gate. A single approval at task start does not whitewash the rest of the path - oversight must be re-confirmed AFTER external influence has entered the context.

The policy

A few lines of YAML in your repository.

Part of OWASP agentic default. Assign the pack, or copy this policy into a manifest of your own.

Verdict
Critical: the step is blocked before it runs, and an incident is opened.
Checked
Before every step, against everything the agent did earlier in the task.
Rule
not
OWASP agentic
ASI01, ASI06, ASI09
- name: "External content taint: high-impact actions require a fresh gate"  description: |    ASI01 (Agent Goal Hijack), ASI06 (Memory & Context Poisoning),    ASI09 (Human-Agent Trust Exploitation):    If the agent has fetched external content earlier in this task,    any subsequent high-impact action (code execution, outbound message,    destructive or mutating resource call) must be IMMEDIATELY preceded    by a step.gate. A single approval at task start does not whitewash    the rest of the path - oversight must be re-confirmed AFTER external    influence has entered the context.    This addresses the EchoLeak / indirect prompt injection pattern:    malicious instructions hidden in retrieved content silently redirect    the agent's plan. By requiring fresh approval after every external    ingestion, the policy forces the human (or a downstream policy gate)    to re-validate intent before any irreversible action.    Convention: requires step.resource GETs to set the property    target.trust = "external" when fetching content from outside an    internal allowlist.    This is a path-dependent compound policy and is the headline    demonstration of "policies on paths".    Applies to: ALL agents.  rule_type: not  params:    condition:      rule_type: all_of      params:        conditions:          - rule_type: any_of            params:              conditions:                - rule_type: current_is                  params:                    kind: step.exec                - rule_type: current_is                  params:                    kind: step.message                    verb: POST                - rule_type: current_is                  params:                    kind: step.resource                    verb: POST                - rule_type: current_is                  params:                    kind: step.resource                    verb: PATCH                - rule_type: current_is                  params:                    kind: step.resource                    verb: DELETE          - rule_type: history_contains            params:              kind: step.resource              count_attempts: true              verb: GET              property_filter:                target.trust: external          - rule_type: not            params:              condition:                rule_type: step_directly_preceded_by                params:                  required_kind: step.gate  severity: critical  enforcement_point: step_execution

Why it matters

This addresses the EchoLeak / indirect prompt injection pattern: malicious instructions hidden in retrieved content silently redirect the agent's plan. By requiring fresh approval after every external ingestion, the policy forces the human (or a downstream policy gate) to re-validate intent before any irreversible action.

Convention: requires step.resource GETs to set the property target.trust = "external" when fetching content from outside an internal allowlist.

This is a path-dependent compound policy and is the headline demonstration of "policies on paths".

Applies to: ALL agents.

How the rule works

not: Compound rule: passes if and only if the sub-condition fails.

  • condition: {"rule_type":"all_of","params":{"conditions":[{"rule_type":"any_of","params":{"conditions":[{"rule_type":"current_is","params":{"kind":"step.exec"}},{"rule_type":"current_is","params":{"kind":"step.message","verb":"POST"}},{"rule_type":"current_is","params":{"kind":"step.resource","verb":"POST"}},{"rule_type":"current_is","params":{"kind":"step.resource","verb":"PATCH"}},{"rule_type":"current_is","params":{"kind":"step.resource","verb":"DELETE"}}]}},{"rule_type":"history_contains","params":{"kind":"step.resource","count_attempts":true,"verb":"GET","property_filter":{"target.trust":"external"}}},{"rule_type":"not","params":{"condition":{"rule_type":"step_directly_preceded_by","params":{"required_kind":"step.gate"}}}}]}}. A single sub-condition (dict with rule_type and params).

OWASP agentic threats it addresses

In these packs

From Kyvvu's open policy library at github.com/Kyvvu/manifests, Apache 2.0.

Next

Watch it decide.

In your browser

Try it in the playground

The policy is loaded. Add the steps your agent takes and watch the engine decide, step by step.

On your machine

Five minutes, no account

kyvvu try runs a demo agent against the bundled baseline. Then assign this pack from the dashboard or the CLI.

For security teams

See it on your agents

A 30-minute demo on the agents you actually run, with the policies that fit them.