Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Gemini Skill Guide · jdrhyne | DecimalAI

jdrhyne/gemini

SkillScore68Maintained

Use when the user asks to run Gemini CLI for code review, plan review, or big context (>200k) processing. Ideal for comprehensive analysis requiring large context windows. Uses Gemini 3 Pro by default for state-of-the-art reasoning and coding.

agent-toolingby jdrhyne

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/jdrhyne-gemini/SKILL.md

decimalai skills pull jdrhyne-gemini --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/jdrhyne-gemini/SKILL.md https://app.decimal.ai/s/jdrhyne-gemini/SKILL.md

Raw file: /s/jdrhyne-gemini/SKILL.md ↗ · machine-readable summary at /s/jdrhyne-gemini.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 34d ago

—+172% tokens0% turns68% of whole cases pass with skill (15/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-03	✗→✓	▲ Improved	587%	0%
case-07	✗→✓	▲ Improved	136%	0%
case-08	✗→✓	▲ Improved	93%	0%
case-09	✗→✓	▲ Improved	96%	0%
case-10	✗→✓	▲ Improved	139%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

jdrhyne / agent-skills

codex/gemini/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1

Skill contents

SKILL.md

Gemini Skill Guide

When to Use Gemini

WHEN ASKED TO BE ACTIVATED
Code Review: Comprehensive code reviews across multiple files
Plan Review: Analyzing architectural plans, technical specifications, or project roadmaps
Big Context Processing: Tasks requiring >200k tokens of context (entire codebases, documentation sets)
Multi-file Analysis: Understanding relationships and patterns across many files

⚠️ Critical: Background/Non-Interactive Mode Warning

NEVER use --approval-mode default in background or non-interactive shells (like Claude Code tool calls). It will hang indefinitely waiting for approval prompts that cannot be provided.

For automated/background reviews:

✅ Use --approval-mode yolo for fully automated execution
✅ OR wrap with timeout: timeout 300 gemini ...
❌ NEVER use --approval-mode default without interactive terminal

Symptoms of hung Gemini:

Process running 20+ minutes with 0% CPU usage
No network activity
Process state shows 'S' (sleeping)

Fix hung process:

bash
# Check if hung
ps aux | grep gemini | grep -v grep

# Kill if necessary
pkill -9 -f "gemini.*gemini-3-pro-preview"

Running a Task

Ask the user (via AskUserQuestion) which model to use in a single prompt. Available models:

gemini-3-pro-preview ⭐ (flagship model, best for coding & complex reasoning, 35% better at software engineering than 2.5 Pro)
gemini-3-flash (sub-second latency, distilled from 3 Pro, best for speed-critical tasks)
gemini-2.5-pro (legacy option, strong all-around performance)
gemini-2.5-flash (legacy option, cost-efficient with thinking capabilities)
gemini-2.5-flash-lite (legacy option, fastest processing)

Select the approval mode based on the task:

default: Prompt for approval (⚠️ ONLY for interactive terminal sessions)
auto_edit: Auto-approve edit tools only (for code reviews with suggestions)
yolo: Auto-approve all tools (✅ REQUIRED for background/automated tasks)

Assemble the command with appropriate options:

-m, --model <MODEL> - Model selection
--approval-mode <default|auto_edit|yolo> - Control tool approval
-y, --yolo - Alternative to --approval-mode yolo
-i, --prompt-interactive "prompt" - Execute prompt and continue interactively
--include-directories <DIR> - Additional directories to include in workspace
-s, --sandbox - Run in sandbox mode for isolation

For background/automated tasks, ALWAYS use --approval-mode yolo or add timeout wrapper. NEVER use default in non-interactive shells.

Run the command and capture the output. For background/automated mode:

bash # Recommended: Use yolo for background tasks gemini -m gemini-3-pro-preview --approval-mode yolo "Review this codebase for security issues"

# Or with timeout (5 min limit) timeout 300 gemini -m gemini-3-pro-preview --approval-mode yolo "Review this codebase"

For interactive sessions with an initial prompt:

bash gemini -m gemini-3-pro-preview -i "Review the authentication system" --approval-mode auto_edit

After Gemini completes, inform the user: "The Gemini analysis is complete. You can start a new Gemini session for follow-up analysis or continue exploring the findings."

Safety Boundaries

Do not run Gemini unless the user explicitly asked for Gemini or for a second-opinion review that justifies it.
Do not use --approval-mode yolo unless the task is non-interactive and the user approved high-autonomy execution.
Do not save reports or write files unless the user asked for an artifact.
Do not pass secrets, private keys, or unrelated workspace data into prompts.

Quick Reference

| Use case | Approval mode | Key flags | | --- | --- | --- | | Background code review | yolo ✅ | -m gemini-3-pro-preview --approval-mode yolo | | Background analysis | yolo ✅ | -m gemini-3-pro-preview --approval-mode yolo | | Background with timeout | yolo ✅ | timeout 300 gemini -m gemini-3-pro-preview --approval-mode yolo | | Interactive code review | default | -m gemini-3-pro-preview --approval-mode default (interactive terminal only) | | Code review with auto-edits | auto_edit | -m gemini-3-pro-preview --approval-mode auto_edit | | Automated refactoring | yolo | -m gemini-3-pro-preview --approval-mode yolo | | Speed-critical background | yolo ✅ | -m gemini-3-flash --approval-mode yolo | | Cost-optimized background | yolo ✅ | -m gemini-2.5-flash --approval-mode yolo | | Multi-directory analysis | yolo (if background) | --include-directories <DIR1> --include-directories <DIR2> | | Interactive with prompt | auto_edit or default | -i "prompt" --approval-mode <mode> |

Model Selection Guide

| Model | Best for | Context window | Key features | | --- | --- | --- | --- | | gemini-3-pro-preview ⭐ | Flagship model: Complex reasoning, coding, agentic tasks | 1M input / 64k output | Vibe coding, 76.2% SWE-bench | | gemini-3-flash | Sub-second latency, speed-critical applications | 1M input / 64k output | Distilled from 3 Pro, TPU-optimized | | gemini-2.5-pro | Legacy: Strong all-around performance | 1M input / 65k output | Thinking mode, mature stability | | gemini-2.5-flash | Legacy: Cost-efficient, high-volume tasks | 1M input / 65k output | Thinking mode | | gemini-2.5-flash-lite | Legacy: Fastest processing, high throughput | 1M input / 65k output | Maximum speed, minimal latency |

Gemini 3 Advantages: 35% higher accuracy in software engineering, state-of-the-art on SWE-bench (76.2%), GPQA Diamond (91.9%), and WebDev Arena (1487 Elo). Knowledge cutoff: January 2025.

Coming Soon: gemini-3-deep-think for ultra-complex reasoning with enhanced thinking capabilities.

Common Use Cases

Code Review (Background/Automated)

bash
# For background execution (Claude Code, CI/CD, etc.)
gemini -m gemini-3-pro-preview --approval-mode yolo \
  "Perform a comprehensive code review focusing on:
   1. Security vulnerabilities
   2. Performance issues
   3. Code quality and maintainability
   4. Best practices violations"

# With timeout safety (5 minutes)
timeout 300 gemini -m gemini-3-pro-preview --approval-mode yolo \
  "Perform a comprehensive code review..."

Plan Review (Background/Automated)

bash
# For background execution
gemini -m gemini-3-pro-preview --approval-mode yolo \
  "Review this architectural plan for:
   1. Scalability concerns
   2. Missing components
   3. Integration challenges
   4. Alternative approaches"

Big Context Analysis (Background/Automated)

bash
# For background execution
gemini -m gemini-3-pro-preview --approval-mode yolo \
  "Analyze the entire codebase to understand:
   1. Overall architecture
   2. Key patterns and conventions
   3. Potential technical debt
   4. Refactoring opportunities"

Interactive Code Review (Terminal Only)

bash
# ONLY use default mode in interactive terminal
gemini -m gemini-3-pro-preview --approval-mode default \
  "Review the authentication flow for security issues"

Following Up

Gemini CLI sessions are typically one-shot or interactive. Unlike Codex, there's no built-in resume functionality.
For follow-up analysis, start a new Gemini session with context from previous findings.
When proposing follow-up actions, restate the chosen model and approval mode.
Use AskUserQuestion after each Gemini command to confirm next steps or gather clarifications.

Error Handling

Stop and report failures whenever gemini --version or a Gemini command exits non-zero.
Request direction before retrying failed commands.
Before using high-impact flags (--approval-mode yolo, -y, --sandbox), ask the user for permission using AskUserQuestion unless already granted.
When output includes warnings or partial results, summarize them and ask how to adjust using AskUserQuestion.

Troubleshooting Hung Gemini Processes

Detection

bash
# Check for hung processes
ps aux | grep -E "gemini.*gemini-3" | grep -v grep

# Look for these symptoms:
# - Process running 20+ minutes
# - CPU usage at 0%
# - Process state 'S' (sleeping)
# - No network connections

Diagnosis

bash
# Get detailed process info
ps -o pid,etime,pcpu,stat,command -p <PID>

# Check network activity
lsof -p <PID> 2>/dev/null | grep -E "(TCP|ESTABLISHED)" | wc -l
# If result is 0, process is hung

Resolution

bash
# Kill hung Gemini processes
pkill -9 -f "gemini.*gemini-3-pro-preview"

# Or kill specific PID
kill -9 <PID>

# Verify cleanup
ps aux | grep gemini | grep -v grep

Prevention

ALWAYS use --approval-mode yolo for background/automated tasks
Add timeout wrapper for safety: timeout 300 gemini ...
Never use --approval-mode default in non-interactive shells
Monitor first run with ps to ensure process completes

Tips for Large Context Processing

Be specific: Provide clear, structured prompts for what to analyze
Use include-directories: Explicitly specify all relevant directories
Choose the right model:

Use gemini-3-pro-preview for complex reasoning, coding tasks, and maximum analysis quality (recommended default)
Use gemini-3-flash for speed-critical tasks requiring sub-second response times
Use gemini-2.5-flash for cost-optimized high-volume processing

Leverage Gemini 3's strengths: 35% better at software engineering tasks, exceptional at agentic workflows and vibe coding
Break down complex tasks: Even with large context, structured analysis is more effective
Return structured findings: Ask Gemini for markdown or JSON that can be quoted directly in the final answer

CLI Version

Requires Gemini CLI v0.16.0 or later for Gemini 3 model support. Check version: gemini --version

Illustrative Want to see what jdrhyne/gemini would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/24/2026, 18:55 UTC · 22 cases

Measured on 21 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+55%

68% vs 14%

Agent turns

1.0 vs 1.0

Tokens

+172%

3,257.52 vs 1,196.95

fail → pass · 13pass → pass · 2fail → fail · 6pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I need you to run a full code review on our repository using Gemini. Provide a summary covering potential security vulnerabilities, performance bottlenecks, and maintainability concerns. Please prompt me to choose the model first before kicking off the command.	fail→fail	10,072	34,113	+239%	1	1	0%	1,211	3,960	+227%	0	0	—
case-02 Please run Gemini to analyze our new microservices architecture specification located in docs/architecture.md. I want a detailed breakdown of scalability risks, missing components, and integration bottlenecks. Make sure to query me regarding model selection first.	fail→fail	3,401	9,026	+165%	1	1	0%	515	3,275	+536%	0	0	—
case-03 Can you execute a multi-directory analysis across our codebase with Gemini to identify technical debt and structural anti-patterns? Deliver the results as a structured markdown report. Remember to ask me which model variant I prefer before starting the process.	fail→pass	3,073	3,133	+2%	1	1	0%	461	3,166	+587%	0	0	—
case-04 Could you review the changes in src/auth.ts to ensure error handling is properly implemented?	fail→fail	2,602	2,346	-10%	1	1	0%	322	3,019	+838%	0	0	—
case-05 I need to run an automated refactoring session using the OpenAI Codex CLI. How should I invoke codex to update src/utils.ts?	pass→fail	10,860	7,882	-27%	1	1	0%	1,643	3,911	+138%	0	0	—
case-06 Please run a non-interactive code review using the Claude CLI tool on the current workspace.	fail→fail	3,811	20,256	+432%	1	1	0%	295	4,852	+1545%	0	0	—
case-07 We need to schedule an automated background review of our API backend in our non-interactive agent runner using Gemini with model gemini-3-pro-preview. Which approval mode flag should be passed to prevent the process from hanging?	fail→pass	8,892	2,897	-67%	1	1	0%	1,319	3,110	+136%	0	0	—
case-08 I am sitting at an interactive terminal session and want to run Gemini with model gemini-3-pro-preview to review my code with suggested auto-edits. Which approval mode flag should be used?	fail→pass	9,981	2,491	-75%	1	1	0%	1,579	3,049	+93%	0	0	—
case-09 Write a bash command for an automated background Gemini review of security issues using model gemini-3-pro-preview that includes a 5-minute safety timeout to prevent stuck execution.	fail→pass	9,429	3,794	-60%	1	1	0%	1,571	3,084	+96%	0	0	—
case-10 Which specific Gemini model variant should be selected when running background automated checks where sub-second latency is required?	fail→pass	9,492	2,473	-74%	1	1	0%	1,265	3,019	+139%	0	0	—
case-11 Our team wants to run high-volume background processing using Gemini on a budget. Which legacy flash model variant supports thinking capabilities while staying cost-efficient?	fail→pass	11,470	2,673	-77%	1	1	0%	1,825	3,083	+69%	0	0	—
case-12 Construct a background command using Gemini with model gemini-3-pro-preview that analyzes both packages/core and packages/ui simultaneously without changing directories.	fail→pass	6,164	5,180	-16%	1	1	0%	1,020	3,503	+243%	0	0	—
case-13 What CLI option allows starting an interactive Gemini session while passing an initial prompt string like 'Review auth logic' directly on startup?	pass→pass	12,308	1,756	-86%	1	1	0%	1,913	2,935	+53%	0	0	—
case-14 Our automated background pipeline seems stuck on a preview model review. What exact shell command filters running processes to detect hung Gemini processes?	fail→pass	17,068	3,871	-77%	1	1	0%	1,212	3,269	+170%	0	0	—
case-15 An automated background process running gemini-3-pro-preview has been sleeping for 25 minutes with 0% CPU. What precise command forcefully terminates all hung instances of this process?	fail→pass	6,222	1,808	-71%	1	1	0%	852	2,933	+244%	0	0	—
case-16 How do you check network socket activity for a specific PID (e.g. 12345) to verify if a sleeping process has active connections?	pass→pass	11,948	5,886	-51%	1	1	0%	2,048	3,754	+83%	0	0	—
case-17 During a background session, `gemini --version` returns exit code 1. What action should be taken according to standard tool error procedures?	fail→pass	7,898	2,300	-71%	1	1	0%	1,255	2,974	+137%	0	0	—
case-18 Before executing a Gemini command with high-autonomy flags like `-y` or `--sandbox` that were not explicitly pre-approved by the user, what tool must be invoked to obtain consent?	fail→fail	9,323	1,852	-80%	1	1	0%	1,391	3,001	+116%	0	0	—
case-19 What is the minimum version of the Gemini CLI tool required to support Gemini 3 model variants?	fail→pass	6,459	1,777	-72%	1	1	0%	1,010	2,875	+185%	0	0	—
case-20 Once a Gemini analysis command finishes running in a background workflow, what statement should be delivered to the user regarding session continuity?	fail→pass	7,254	1,900	-74%	1	1	0%	978	2,965	+203%	0	0	—
case-21 When proposing follow-up analysis actions after completing a Gemini command, what specific details must be restated to the user?	fail→fail	8,899	2,303	-74%	1	1	0%	1,118	2,988	+167%	0	0	—
case-22 Which legacy Gemini model variant provides maximum speed and minimal latency for high-throughput processing?	fail→pass	6,009	1,872	-69%	1	1	0%	848	2,958	+249%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 13.6%
Pass rate with the skill: 68.2%
Cases attempted: 22
Cases counted toward lift: 21
Cases that regressed: 1
Skill version measured: v1
Run completed: 8/24/2026, 18:55 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-03

Prompt

Can you execute a multi-directory analysis across our codebase with Gemini to identify technical debt and structural anti-patterns? Deliver the results as a structured markdown report. Remember to ask me which model variant I prefer before starting the process.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run68

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+69% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 34 days ago

Other measured skills in the registry, with their headline benchmark lift.