Multi-viewpoint Debates
Analysis
The skill is mostly coherent, but its helper script prints copy-paste shell commands with the user’s debate topic inserted unescaped, which could change what runs if the topic contains shell syntax.
Findings (4)
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.
Checks for instructions or behavior that redirect the agent, misuse tools, execute unexpected code, cascade across systems, exploit user trust, or continue outside the intended task.
echo "Copy and run these three commands in sequence:" ... clawdbot sessions_spawn --task "... **Decision**: EOF echo "$TOPIC"
The script prints shell commands for the user to copy and run, while inserting the raw topic inside a double-quoted command argument. If the topic contains quotes, command substitutions, or shell separators, the generated command can be altered when pasted into a shell.
Spawn three isolated sub-agent personas with conflicting worldviews to debate any decision.
Creating additional agent sessions is the core feature and is clearly disclosed, but it means using the skill can trigger multiple autonomous model sessions for one user request.
Automatically captures debate outputs to an archive for future reference and pattern analysis.
The description suggests automatic capture, while the workflow later says to manually collect responses, create a markdown file, and update an index. This is a capability/retention ambiguity rather than evidence of hidden capture.
Checks for exposed credentials, poisoned memory or context, unclear communication boundaries, or sensitive data that could leave the user's control.
Your debates live in a searchable archive ... Over time, your archive becomes a personal decision-making manual.
The skill encourages persistent storage and future reuse of debate outputs. This is disclosed and user-controlled, but those files may contain sensitive decision context.