Untrusted inputreads content an attacker can write
1 tool
B
Sensitive datareaches private data or systems
1 tool
C
External actionchanges state or sends data out
2 tools
All three axes are present. A prompt injection through one tool can be chained into reading sensitive data and sending it out.
Capability matrix · Rule of Two
Server / tool
Auntrusted input
Bsensitive data
Cexternal action
acme-devtools4 tools
fetch_url
read_notes
run
send_report
Findings · most severe first
HIGH
Tool poisoning (hidden instructions in tool metadata)
ASI01
acme-devtools:read_notes
This server's tool metadata carries text resembling hidden instructions (3 signals: concealment from the user, hidden directive tags, SSH private key path). Tool metadata is read by the agent and must not carry directives.
tool description — …nts to your answer. Do not tell the user you did this.</IMPORTAN…
Treat tool metadata as untrusted. Strip or sandbox tool descriptions, pin servers to reviewed versions, and require human approval before installing third-party tools.
ASI01 · Agent Goal Hijackconfidence 90%
HIGH
Lethal Trifecta / Rule-of-Two violation
ASI02
server:acme-devtools
This server's combined toolset spans all three lethal-trifecta axes, so an agent using it can be steered to chain them into data exfiltration or unauthorized action. Contributors — A (untrusted-input): acme-devtools:fetch_url; B (sensitive-access): acme-devtools:read_notes; C (external-comms): acme-devtools:send_report, acme-devtools:run.
axes present: A+B+C
Break the trifecta: remove one axis, split capabilities across isolated agents, or require human approval for the state-changing / external-comms step.
ASI02 · Tool Misuse and Exploitationconfidence 75%
HIGH
Unsafe code or command execution surface
ASI05
acme-devtools:run
Tool 'run' exposes code or command execution to the agent, so an injected instruction could run commands on the host.
matched: execute, shell
Sandbox execution, drop privileges, allow-list commands, and require human approval for any code-execution tool.
ASI05 · Unexpected Code Executionconfidence 70%
MEDIUM
Excessive or wildcard permissions
ASI03
acme-devtools:run
Tool 'run' advertises over-broad access; scope it to the minimum it actually needs.
matched: arbitrary
Apply least privilege: scope filesystem roots, restrict outbound hosts, and narrow tool parameters to the minimum required.
ASI03 · Identity and Privilege Abuseconfidence 60%