Analyzes task complexity, identifies required agents, groups them into optimal batches, and executes batches sequentially (agents within each batch run in parallel). Each batch gets a dedicated context window. Results are written to files and synthesized by an orchestrator. Achieves 60-77% token sav
SkillScore67Maintained
Automatic
◆AI Engineer pack
Last verified 8/27/2026
What it changes
Passed 13 of 20 test cases with the skill, and 6 of 20 without it.
Without skill29%
With skill67%
Checks it passed that the model alone missed
Allocates 8 total agents for task execution.
Limits total agent allocation to at most 2 agents.
Rates task complexity prior to agent batch planning.
and 13 more in the full report
10 fixed6 passed either way7 still failing1 broke
Pass-rate lift
+38 pp
Tokens
+3%
Agent turns
0%
Response time
-11%
Model gemini-3.6-flashMeasured 8/27/2026v116 of 24 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 10
Prompt
/batch "Fix typo in user notification banner text and add debug log line to logger"
What was checked, without and with the skill
Checked
Without skill
With skill
Assigns a task complexity rating in the 1 to 3 range.
Missed
Passed
Limits total agent allocation to at most 2 agents.
Missed
Passed
case-04 · gemini-3.6-flash · 8/27/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md37 linessingle file
/batch
Execute multi-agent tasks using intelligent batching for token efficiency.
Analyzes task complexity, identifies required agents, groups them into optimal batches, and executes batches sequentially (agents within each batch run in parallel). Each batch gets a dedicated context window. Results are written to files and synthesized by an orchestrator. Achieves 60-77% token savings vs naive parallel execution.
Plan batches -- group agents into batches of 3-5 (cohesive grouping).
Execute Batch 1 -- run first group, write results to work/batch1-results.md.
Execute Batch 2 -- read Batch 1 results, run second group, write to work/batch2-results.md.
Synthesize -- orchestrator reads all batch results, produces final output.
Complexity
Batch Size
Example
1-3
1-2 agents
Fix typo, add logging
4-5
3 agents
Add API endpoint with tests
6-7
5 agents
Build feature with frontend/backend
8-10
8 agents
Full system redesign
Examples
# Full-stack feature
/batch "Build user authentication with React frontend, Go backend, and tests"
# Performance work
/batch "Optimize database queries and add caching layer"
# Security audit
/batch "Security assessment of payment processing system"
Takes a problem or decision, generates multiple distinct approaches, evaluates each on pros/cons/effort, and provides a reasoned recommendation. Forces structured thinking before jumping to implementation.