OC Team Builder
WarnAudited by ClawScan on May 18, 2026.
Overview
The skill is mostly aligned with team planning, but its Research Lab mode explicitly encourages autonomous code-changing experiment loops that never stop unless manually interrupted.
Review this skill carefully before installing. It is not showing evidence of data theft or hidden endpoints, but you should only use Research Lab mode in a trusted sandbox with explicit time and iteration limits, and you should inspect any external agent roster files before letting the agent adopt them.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If used as written, the agent could keep consuming compute and changing project files beyond the user's immediate oversight.
The skill tells the agent to keep operating without pausing for the user while modifying experiment targets and running repeated tests.
LOOP FOREVER ... Implement the change ... Run the experiment ... NEVER STOP — do NOT pause to ask the human. The human might be asleep. You are autonomous.
Require explicit user opt-in for autonomous mode, set a fixed iteration/time limit, ask before overnight operation, and provide a clear stop/rollback procedure.
A bad or overbroad run command could modify files, run untrusted project code, consume resources, or perform unintended actions in the selected project directory.
The experiment helper can execute arbitrary shell commands supplied through the CLI, with no allowlist or additional confirmation in the script.
--run-cmd CMD ... eval "$run_cmd" || true ... timeout $((budget * 2)) bash -c "$run_cmd" || true
Run only trusted commands in a sandboxed project, avoid broad shell strings, and consider replacing eval/bash -c with safer allowlisted command templates.
Specialist definitions can steer the agent's responses; if those files are untrusted or edited maliciously, they may introduce unsafe instructions.
The skill intentionally loads external/local agent definition files as role instructions, which is purpose-aligned but means poisoned or untrusted roster files could influence the agent's behavior.
All agent definitions live in `reference/agency-agents-main/` ... To activate any specialist, read their definition file and adopt their role.
Use only trusted roster files, inspect specialist definitions before activation, and keep user/system instructions higher priority than loaded persona text.