llm06-agentic-abuse-hardening-v2
10 rules · Keryn Software Licence · rules 0.8.0 · all packs
How to read a rule
- Severity
- How bad the finding is if it is real:
errorfails a default gate,warningandinfodo not. - Confidence
- P0 exploitable as written; P1 dangerous in context; P2 a heuristic or a posture check: confirm by reading the code.
- False-positive risk
- low: an exact API or flag, so hits are almost always real; medium: the pattern is legitimate in many programs, so check each hit; high: it finds a lead to confirm, so expect benign hits.
- Edge
- The relationship the finding sits on. reaches: content or traffic gets to a thing; calls: one thing invokes another; reads: one thing reads from another; writes: one thing writes to another; sends: data leaves for another party; grants: authority is given or left unchecked; depends_on: something relies on a component it does not verify; trusts: one party believes another without evidence. The boundary it crosses, when there is one: user->prompt (untrusted input reaching a prompt); model->code (model output reaching code, a shell, a query or a page); secret->outside (a secret leaving the trusted side); supply-chain (code, models or prompts pulled in from outside).
- Taxonomy chips
- References for readers who work in OWASP LLM 2025, OWASP Agentic, CWE or NIST AI RMF. They say where a rule belongs, not that a framework is satisfied.
seceng.llm06.tool-call-without-approval-gate
Section titled “seceng.llm06.tool-call-without-approval-gate”info python llm06-agentic-abuse-hardening
Agent or tool execution call occurs without a visible human approval gate in the surrounding scope.
Confidence P1: dangerous in context. False-positive risk medium: the pattern is legitimate in many programs, so check each hit.
Edge inspected: grants authority is given or left unchecked.
seceng.llm06.tool-call-without-approval-gate-js
Section titled “seceng.llm06.tool-call-without-approval-gate-js”info javascript, typescript llm06-agentic-abuse-hardening
Agent or tool execution call occurs without a visible human approval gate in the surrounding scope.
Confidence P1: dangerous in context. False-positive risk medium: the pattern is legitimate in many programs, so check each hit.
Edge inspected: grants authority is given or left unchecked.
seceng.llm06.tool-schema-additional-properties
Section titled “seceng.llm06.tool-schema-additional-properties”warning python llm06-agentic-abuse-hardening
Tool or function schema allows arbitrary additional properties or disables strict validation. The model may inject unexpected parameters.
Confidence P2: a heuristic or a posture check: confirm by reading the code. False-positive risk low: an exact API or flag, so hits are almost always real.
Edge inspected: reads one thing reads from another.
seceng.llm06.tool-schema-additional-properties-js
Section titled “seceng.llm06.tool-schema-additional-properties-js”warning javascript, typescript llm06-agentic-abuse-hardening
Tool or function schema allows arbitrary additional properties or disables strict validation. The model may inject unexpected parameters.
Confidence P2: a heuristic or a posture check: confirm by reading the code. False-positive risk low: an exact API or flag, so hits are almost always real.
Edge inspected: reads one thing reads from another.
seceng.llm06.langchain-tool-missing-args-schema
Section titled “seceng.llm06.langchain-tool-missing-args-schema”warning python langchain
LangChain tool is registered without an args_schema or structured input model. Tool parameters may be hallucinated or injected by the LLM.
Confidence P1: dangerous in context. False-positive risk high: it finds a lead to confirm, so expect benign hits.
Edge inspected: reads one thing reads from another.
CWE-20 LLM06 ASI01 ASI02 ASI09
seceng.llm06.model-controlled-user-id-query
Section titled “seceng.llm06.model-controlled-user-id-query”warning javascript, typescript llm06-agentic-abuse-hardening
Model-controlled tool arguments appear to supply user_id, account_id, or tenant_id to a database query without a visible session authorization check.
Confidence P2: a heuristic or a posture check: confirm by reading the code. False-positive risk medium: the pattern is legitimate in many programs, so check each hit.
Edge inspected: grants authority is given or left unchecked.
CWE-639 LLM06 LLM08 ASI03 ASI04 ASI02
seceng.llm06.model-controlled-file-path
Section titled “seceng.llm06.model-controlled-file-path”warning javascript, typescript llm06-agentic-abuse-hardening
Model-controlled tool arguments are used as a file path without visible sandboxing or allowlist validation.
Confidence P2: a heuristic or a posture check: confirm by reading the code. False-positive risk medium: the pattern is legitimate in many programs, so check each hit.
Edge inspected: reads one thing reads from another.
CWE-22 CWE-73 LLM06 ASI02 ASI10
seceng.llm06.model-controlled-url-fetch
Section titled “seceng.llm06.model-controlled-url-fetch”warning javascript, typescript llm06-agentic-abuse-hardening
Model-controlled tool arguments are used as a URL fetch target without visible allowlist validation. This can enable SSRF or data exfiltration.
Confidence P2: a heuristic or a posture check: confirm by reading the code. False-positive risk medium: the pattern is legitimate in many programs, so check each hit.
Edge inspected: reaches content or traffic gets to a thing; it crosses the user->prompt boundary (untrusted input reaching a prompt).
seceng.llm06.agent-broad-tools-without-approval
Section titled “seceng.llm06.agent-broad-tools-without-approval”info python llm06-agentic-abuse-hardening
Agent is initialized with a broad tool list and no visible human approval or max-iteration control. This increases blast radius for tool abuse.
Confidence P1: dangerous in context. False-positive risk high: it finds a lead to confirm, so expect benign hits.
Edge inspected: grants authority is given or left unchecked.
CWE-250 CWE-770 LLM06 LLM10 ASI02 ASI08
seceng.llm06.tool-output-direct-db-write
Section titled “seceng.llm06.tool-output-direct-db-write”warning javascript, typescript llm06-agentic-abuse-hardening
Tool or model output is written directly to persistent storage without visible validation. This can enable state poisoning or unauthorized updates.
Confidence P2: a heuristic or a posture check: confirm by reading the code. False-positive risk medium: the pattern is legitimate in many programs, so check each hit.
Edge inspected: writes one thing writes to another.