Researching In Parallel

Research any topic thoroughly by running three sub-agents with distinct analytical lenses (breadth, critique, evidence), then giving their outputs to a final...

This skill appears to do exactly what it says: orchestrate three research sub-agents, fetch full-text sources, and save structured outputs to a workspace. Key things to consider before installing or running it: - Workspace placement: the SKILL.md requires installing the skill directory inside the agent's workspace so sub-agents can read its templates. That is necessary for the skill to work, but it also means the skill (and sub-agents it spawns) will have read/write access to files under that workspace. Do not install it into a workspace that contains sensitive files you do not want the skill to read. - File reads during setup: the skill states it may perform preparatory reads (reading skill-config.json, checking the workspace, looking for provided-sources files) without waiting for an explicit spawn confirmation. If you have sensitive files in the workspace, consider moving them or using a new empty directory for runs. - Network and scraping: sub-agents will use web_fetch, a browser tool, and a PDF extractor to retrieve full-text sources from the web. This is expected for deep research, but it means the agent will make outbound requests to external sites. If your environment restricts external network access, note that research depth will be reduced. - Saved extracts: the default skill-config.json sets save_source_extracts = true. Source extracts are intended to contain only the content actually used, but they will be written to the workspace. If you prefer not to retain extracts, set save_source_extracts to false before running. - Model selection: the config includes provider/model strings as defaults. These are guidance for sessions_spawn; actual model usage is governed by your OpenClaw allowlist and runtime. If you want to constrain which models/providers the skill uses, add an allowlist in your openclaw.json (agents.defaults.models). - Inspect files if concerned: because the skill is instruction-driven, you can review SKILL.md, the prompt templates in references/prompts, and scripts/assemble_prompts.py before installing. There are no hidden download URLs or credential requests in the package. Overall, the package is internally consistent with its research orchestration purpose. If you will run research on sensitive topics or in a sensitive workspace, create a dedicated empty workspace directory for runs, and consider disabling source-extract saving or restricting network access as appropriate.

Type: OpenClaw Skill Name: researching-in-parallel Version: 1.4.2 The OpenClaw AgentSkills bundle 'researching-in-parallel' is classified as benign. Its purpose is to conduct deep research using multiple sub-agents, which inherently requires access to powerful tools like `sessions_spawn`, `web_fetch`, `browser`, `PDF tool`, and file read/write within a designated workspace. All instructions across `SKILL.md` and the various prompt templates (`references/prompts/*.md`) are transparently aligned with this stated research objective. There is no evidence of intentional harmful behavior such as data exfiltration, unauthorized persistence, remote control, or obfuscation. The skill emphasizes user confirmation, AI provenance, and responsible tool usage, and all file operations are confined to the specified workspace. While the tools used are powerful and could be misused if the agent's environment or user input were compromised, the skill itself does not direct or intend such malicious actions.