Deep Research via Gemini CLI Extension
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill mostly does the advertised Gemini research workflow, but review is warranted because it runs an auto-updating third-party extension and launches a background polling script that may make extra calls or writes.
Install only if you trust the external Gemini Deep Research extension and are comfortable with Google account/API-key use and possible billing. Consider pinning or manually reviewing the extension instead of auto-updating it, watch for the background polling process, and review saved reports before treating them as trusted notes.
Findings (5)
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.
A future or compromised upstream extension update could run locally with access to the user's Gemini configuration and report output path.
The setup directs users to install an external GitHub MCP extension with auto-update instead of a pinned, reviewed version. That extension is later executed by the skill's scripts.
gemini extensions install https://github.com/allenhutchison/gemini-cli-deep-research --auto-update
Use a pinned/reviewed extension version where possible, consider disabling auto-update, and declare the external dependency and credential requirements in metadata.
The background task may run longer than expected, make extra paid API/status calls, and write the report more than once.
The background poll script runs the poller twice and, on completion, runs the saver twice, despite the user-facing instructions describing a bounded 40-minute polling workflow and a single save.
node "$SKILL_DIR/poll-research.js" "$TASK_DIR" >> poll-out.log 2>&1 ... RESULT=$(cat <<< "$(node "$SKILL_DIR/poll-research.js" "$TASK_DIR")") ... node "$SKILL_DIR/save-report.js" "$TASK_DIR" >> save-out.log 2>&1 ... "$(node "$SKILL_DIR/save-report.js" "$TASK_DIR")"
Change the script to run each action once, capture output with tee or a temp file, and show the user the PID and a clear stop command.
Running the skill executes the locally installed Gemini Deep Research extension under the user's account.
The script starts the local MCP server with fixed arguments rather than arbitrary shell input. This is expected for the integration, but it executes code from the installed extension.
const server = spawn('node', ['dist/index.js'], { cwd: EXT_PATH, stdio: ['pipe', 'pipe', 'pipe'] });Install the extension only from a source you trust and avoid modifying the extension directory with untrusted code.
Research requests can consume paid Google quota and may be associated with the configured Google account.
The skill requires Google account authentication and a paid API key. This is expected for Gemini Deep Research, but it uses account and billing authority.
Prerequisite: a paid Google AI API key ... gemini auth ... gemini extensions config gemini-deep-research
Use a dedicated or least-privileged API key where possible, monitor billing/quota, and ensure the registry metadata declares the credential requirement.
Sensitive research topics or untrusted generated content may remain in local notes/temp files and could influence later workflows if reused as context.
The skill persists externally generated research reports and task metadata locally, including the topic, output path, and research ID.
保存位置：~/ObsidianVault/Default/DeepResearch/ ... task.json ← task parameters + research ID
Store reports in a dedicated folder, review generated content before relying on it, and clean temporary task directories for sensitive research.