Ready to supercharge your AI agents?

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

/hub:init — Create New Session · alirezarezvani | DecimalAI

alirezarezvani/init

SkillScore77Maintained

Create a new AgentHub collaboration session with task, agent count, and evaluation criteria. Use when the user runs /hub:init or asks to start a multi-agent competition on a task.

agent-toolingby alirezarezvani

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/alirezarezvani-init/SKILL.md

decimalai skills pull alirezarezvani-init --out .claude/skills/

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

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

Raw file: /s/alirezarezvani-init/SKILL.md ↗ · machine-readable summary at /s/alirezarezvani-init.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 49d ago

—+19% tokens0% turns77% of whole cases pass with skill (17/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-07	✗→✓	▲ Improved	4%	0%
case-08	✗→✓	▲ Improved	-16%	0%
case-09	✗→✓	▲ Improved	-16%	0%
case-10	✗→✓	▲ Improved	30%	0%
case-11	✗→✓	▲ Improved	13%	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

alirezarezvani / claude-skills

.gemini/skills/init/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v3

Skill contents

SKILL.md

/hub:init — Create New Session

Initialize an AgentHub collaboration session. Creates the .agenthub/ directory structure, generates a session ID, and configures evaluation criteria.

Usage

/hub:init                                                    # Interactive mode
/hub:init --task "Optimize API" --agents 3 --eval "pytest bench.py" --metric p50_ms --direction lower
/hub:init --task "Refactor auth" --agents 2                  # No eval (LLM judge mode)

What It Does

If arguments provided

Pass them to the init script:

bash
python {skill_path}/scripts/hub_init.py \
  --task "{task}" --agents {N} \
  [--eval "{eval_cmd}"] [--metric {metric}] [--direction {direction}] \
  [--base-branch {branch}]

If no arguments (interactive mode)

Collect each parameter:

Task — What should the agents do? (required)
Agent count — How many parallel agents? (default: 3)
Eval command — Command to measure results (optional — skip for LLM judge mode)
Metric name — What metric to extract from eval output (required if eval command given)
Direction — Is lower or higher better? (required if metric given)
Base branch — Branch to fork from (default: current branch)

Output

AgentHub session initialized
  Session ID: 20260317-143022
  Task: Optimize API response time below 100ms
  Agents: 3
  Eval: pytest bench.py --json
  Metric: p50_ms (lower is better)
  Base branch: dev
  State: init

Next step: Run /hub:spawn to launch 3 agents

For content or research tasks (no eval command → LLM judge mode):

AgentHub session initialized
  Session ID: 20260317-151200
  Task: Draft 3 competing taglines for product launch
  Agents: 3
  Eval: LLM judge (no eval command)
  Base branch: dev
  State: init

Next step: Run /hub:spawn to launch 3 agents

Baseline Capture

If --eval was provided, capture a baseline measurement after session creation:

Run the eval command in the current working directory
Extract the metric value from stdout
Append baseline: {value} to .agenthub/sessions/{session-id}/config.yaml
Display: Baseline captured: {metric} = {value}

This baseline is used by result_ranker.py --baseline during evaluation to show deltas. If the eval command fails at this stage, warn the user but continue — baseline is optional.

After Init

Tell the user:

Session created with ID {session-id}
Baseline metric (if captured)
Next step: /hub:spawn to launch agents
Or /hub:spawn {session-id} if multiple sessions exist

Illustrative Want to see what alirezarezvani/init 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

8/10/2026, 04:15 UTC · 22 cases

Measured on 18 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

+45%

77% vs 32%

Agent turns

1.0 vs 1.0

Tokens

+19%

1,387.22 vs 1,170.56

fail → pass · 11pass → pass · 6fail → fail · 4pass → 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 to spin up an AgentHub session to work on optimizing our API response latency. Set up 3 parallel workers using `pytest bench/api_perf.py` as the benchmark script to monitor `response_time_ms` where smaller numbers mean better performance.	fail→fail	4,691	4,096	-13%	1	1	0%	387	974	+152%	0	0	—
case-02 Can you initialize a new AgentHub collaboration for rewriting our user onboarding guide? We want 3 agents working on this, but we won't be using an automated benchmark suite—just rely on LLM judging.	fail→fail	8,596	3,905	-55%	1	1	0%	1,831	926	-49%	0	0	—
case-03 Please set up an AgentHub workspace to refactor our authentication module with 2 agents. Fork the workspace from our `develop` branch and run `pytest tests/test_auth.py` to evaluate progress using `error_count` (lower target).	fail→fail	7,324	4,413	-40%	1	1	0%	1,094	950	-13%	0	0	—
case-04 We already initialized session 20260317-143022 for API optimization. Now start the 3 parallel background agents to begin work on the task.	fail→fail	6,263	6,342	+1%	1	1	0%	1,014	1,979	+95%	0	0	—
case-05 Session 20260317-143022 has finished running all 3 agents. Rank the generated candidate branches against our baseline metric to pick the winning code.	pass→pass	5,447	8,944	+64%	1	1	0%	848	2,834	+234%	0	0	—
case-06 Check if the background workers for session 20260317-143022 are still running or if any worker process crashed.	pass→fail	11,409	4,427	-61%	1	1	0%	1,707	991	-42%	0	0	—
case-07 Initialize a session for task 'Speed up database queries' with 2 agents, using `pytest tests/db_bench.py` and metric `query_time_sec` (lower is better). What happens right after the session directory structure and config are created?	fail→pass	7,066	3,217	-54%	1	1	0%	1,319	1,368	+4%	0	0	—
case-08 When running the initialization step with `--eval "npm test"`, the initial baseline test run fails with exit code 1 because the project environment is missing a seed file. Should session initialization abort or fail completely?	fail→pass	8,451	2,825	-67%	1	1	0%	1,525	1,281	-16%	0	0	—
case-09 I want to interactively create an AgentHub session without passing command line flags. What is the very first parameter I must provide when prompted?	fail→pass	7,315	1,599	-78%	1	1	0%	1,228	1,036	-16%	0	0	—
case-10 During interactive session initialization, if I press Enter to skip specifying the agent count, how many parallel agents will be configured by default?	fail→pass	4,459	1,395	-69%	1	1	0%	700	910	+30%	0	0	—
case-11 When setting up an AgentHub session interactively, if no base branch is explicitly entered, which branch is selected as the default?	fail→pass	4,495	1,627	-64%	1	1	0%	798	898	+13%	0	0	—
case-12 Initialize an AgentHub session for 'Maximize throughput' with agent count 4, evaluation command `python bench.py`, and metric `req_per_sec`. Which direction parameter must be passed for this throughput metric?	fail→pass	3,506	3,010	-14%	1	1	0%	621	1,309	+111%	0	0	—
case-13 Set up a session with task 'Fix memory leak' and eval command `valgrind --log-file=mem.log ./bin`. If an evaluation command is provided, is `--metric` required or optional?	fail→pass	3,459	6,675	+93%	1	1	0%	615	2,103	+242%	0	0	—
case-14 When initializing an AgentHub session where a metric name is specified, is the metric direction parameter required or optional?	pass→pass	7,042	3,557	-49%	1	1	0%	1,184	1,218	+3%	0	0	—
case-15 Where is the session configuration file saved on disk after initializing session ID `20260317-100000`?	fail→pass	10,610	1,761	-83%	1	1	0%	2,139	1,049	-51%	0	0	—
case-16 When initializing a content draft session with no `--eval` command provided, what evaluation mode label should be displayed in the summary output?	fail→pass	5,387	1,522	-72%	1	1	0%	960	1,015	+6%	0	0	—
case-17 What local directory structure is created at the start of initializing an AgentHub session?	pass→pass	7,322	4,100	-44%	1	1	0%	1,375	1,528	+11%	0	0	—
case-18 If a workspace already contains multiple existing AgentHub sessions, what specific advice should be given to the user after initializing a new session regarding how to spawn agents?	fail→pass	8,512	2,556	-70%	1	1	0%	1,528	1,187	-22%	0	0	—
case-19 I am writing an automated shell script to invoke the AgentHub initialization script `hub_init.py` for task 'Refactor CSS styles'. How should the task string parameter be passed?	pass→pass	6,421	3,313	-48%	1	1	0%	1,322	1,359	+3%	0	0	—
case-20 What is the initial value of the State field reported in the AgentHub session summary output immediately after setup?	fail→pass	9,190	1,493	-84%	1	1	0%	1,680	1,006	-40%	0	0	—
case-21 When an AgentHub session is initialized without an `--eval` command (LLM judge mode), does the process attempt to run a baseline measurement script?	pass→pass	5,842	2,541	-57%	1	1	0%	1,047	1,149	+10%	0	0	—
case-22 Initialize a session for task 'Add logging'. What identifier format is generated for the session ID?	pass→pass	7,169	9,227	+29%	1	1	0%	1,167	1,741	+49%	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 18 counted toward the lift figure. The other 4 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 +45 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 cases got worse with the skill loaded, and they are 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: 31.8%
Pass rate with the skill: 77.3%
Cases attempted: 22
Cases counted toward lift: 18
Cases that regressed: 2
Skill version measured: v3
Run completed: 8/10/2026, 04:15 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Prompt

Initialize a session for task 'Speed up database queries' with 2 agents, using `pytest tests/db_bench.py` and metric `query_time_sec` (lower is better). What happens right after the session directory structure and config are created?

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 run77

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+42% 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 48 days ago

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