verify-cli
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_CONTEXT_INJECTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill's primary function is to execute arbitrary shell commands provided via the
$ARGUMENTSparameter within a terminal harness session. This creates a potential surface for command injection if input is not carefully handled. - Evidence:
SKILL.mdinstructions specify parsing$ARGUMENTSand launching the command in an isolated tmux session. - Evidence:
references/harness-loop.mdprovides shell templates that substitute<command-under-test>directly intotmux new-sessionandscriptcommands. - [DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the
!command syntax inSKILL.mdto execute shell commands at skill load time to determine the local environment and project scripts. - Evidence:
tmux: !command -v tmux >/dev/null 2>&1 && tmux -V || echo "(none)"`` - Evidence:
Repo-native harness hints: !out=$(jq -r '.scripts | keys[]' package.json 2>/dev/null | grep -iE 'e2e|smoke|demo|tui|cli' | head -5); echo "${out:-(none)}"`` - [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted output from terminal sessions, which could contain malicious instructions designed to influence the agent's behavior during assertion and reporting phases.
- Ingestion points: Captured terminal transcripts in
references/harness-loop.md(e.g.,tmux capture-pane -pt "$S" > "$SCRATCH/00-initial.txt"). - Boundary markers: The skill does not define specific delimiters or instructions for the agent to ignore potentially malicious content within the transcripts.
- Capability inventory: The skill has extensive capabilities to execute shell commands, manage tmux sessions, and run Python scripts.
- Sanitization: There is no evidence of sanitization or filtering applied to terminal outputs before they are processed by the agent.
- [DYNAMIC_EXECUTION]: The skill includes a template for a Python driver to facilitate interactive flows without tmux, which executes commands via
os.execvpat runtime. - Evidence:
references/harness-loop.mdprovides a Python script usingpty.fork()andos.execvp(sys.argv[1], sys.argv[1:])to wrap the command under test.
Audit Metadata