Ready to supercharge your AI agents?

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

Get Started Free →

Agentica Infrastructure

SkillScore91Maintained

agentica-infrastructure

Reference guide for Agentica multi-agent infrastructure APIs

agent-toolingby vibeeval

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/agentica-infrastructure/SKILL.md

decimalai skills pull agentica-infrastructure --out .claude/skills/

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

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

Raw file: /s/agentica-infrastructure/SKILL.md ↗ · machine-readable summary at /s/agentica-infrastructure.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 50d ago

Yes — +73% lift-22% tokens0% turns91% of whole cases pass with skill (20/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	-42%	0%
case-02	✗→✓	▲ Improved	-44%	0%
case-03	✗→✓	▲ Improved	-19%	0%
case-04	✗→✓	▲ Improved	-2%	0%
case-05	✗→✓	▲ Improved	-24%	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 — 8/1/2026

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

vibeeval / vibecosystem

skills/agentica-infrastructure/SKILL.md

View on GitHub ↗

License MITImported 8/1/2026Version v2

Skill contents

SKILL.md

Agentica Infrastructure Reference

Complete API specification for Agentica multi-agent coordination infrastructure.

When to Use

Building multi-agent workflows with Agentica patterns
Need exact constructor signatures for pattern classes
Want to understand coordination database schema
Implementing custom patterns using primitives
Debugging agent tracking or orphan detection

Quick Reference

11 Pattern Classes

| Pattern | Purpose | Key Method | |---------|---------|------------| | Swarm | Parallel perspectives | .execute(query) | | Pipeline | Sequential stages | .run(initial_state) | | Hierarchical | Coordinator + specialists | .execute(task) | | Jury | Voting consensus | .decide(return_type, question) | | GeneratorCritic | Iterative refinement | .run(task) | | CircuitBreaker | Failure fallback | .execute(query) | | Adversarial | Debate + judge | .resolve(question) | | ChainOfResponsibility | Route to handler | .process(query) | | MapReduce | Fan out + reduce | .execute(query, chunks) | | Blackboard | Shared state | .solve(query) | | EventDriven | Event bus | .publish(event) |

Core Infrastructure

| Component | File | Purpose | |-----------|------|---------| | CoordinationDB | coordination.py | SQLite tracking | | tracked_spawn | tracked_agent.py | Agent with tracking | | HandoffAtom | handoff_atom.py | Universal handoff format | | BlackboardCache | blackboard.py | Hot tier communication | | MemoryService | memory_service.py | Core + Archival memory | | create_claude_scope | claude_scope.py | Scope with file ops |

Primitives

| Primitive | Purpose | |-----------|---------| | Consensus | Voting (MAJORITY, UNANIMOUS, THRESHOLD) | | Aggregator | Combine results (MERGE, CONCAT, BEST) | | HandoffState | Structured agent handoff | | build_premise | Structured premise builder | | gather_fail_fast | TaskGroup-based parallel execution |

Full API Spec

See: API_SPEC.md in this skill directory

Usage Example

python
from scripts.agentica_patterns.patterns import Swarm, Jury
from scripts.agentica_patterns.primitives import ConsensusMode
from scripts.agentica_patterns.coordination import CoordinationDB
from scripts.agentica_patterns.tracked_agent import tracked_spawn

# Create tracking database
db = CoordinationDB(session_id="my-session")

# Swarm with tracking
swarm = Swarm(
    perspectives=["Security expert", "Performance expert"],
    db=db
)
result = await swarm.execute("Review this code")

# Jury with consensus
jury = Jury(
    num_jurors=3,
    consensus_mode=ConsensusMode.MAJORITY,
    premise="You evaluate code quality",
    db=db
)
verdict = await jury.decide(bool, "Is this code production ready?")

Location

API spec: .claude/skills/agentica-infrastructure/API_SPEC.md Source: scripts/agentica_patterns/

Illustrative Want to see what agentica-infrastructure 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

8/2/2026, 18:51 UTC · 22 cases

Lift

+73%

91% vs 18%

Agent turns

0%

1.0 vs 1.0

Tokens

-22%

1,669.23 vs 2,138.73

fail → pass · 16pass → pass · 4fail → fail · 2

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'm building a multi-agent evaluation tool using the Agentica library. Write a Python script that instantiates a consensus-based voting committee with three agents to decide on a true/false question, ensuring all steps are registered in a local tracking database session.	fail→pass	20,478	8,079	-61%	1	1	0%	4,310	2,490	-42%	0	0	—
case-22 When defining structured input validation schemas for an LLM agent tool in Python using Pydantic v2, how do you restrict a numeric parameter to be strictly greater than zero?	pass→pass	9,608	6,001	-38%	1	1	0%	1,818	2,019	+11%	0	0	—
case-02 Please provide a Python code example demonstrating how to run parallel agent perspectives on a query using Agentica patterns. I want to pass in two distinct agent roles (like security and performance) and attach a coordination database instance for tracking.	fail→pass	15,580	4,736	-70%	1	1	0%	3,256	1,831	-44%	0	0	—
case-03 I need to set up sequential agent processing with Agentica infrastructure components. Could you generate code that creates a tracking database session, initializes a multi-stage agent pipeline, and executes it starting with an initial state dictionary?	fail→pass	13,464	6,819	-49%	1	1	0%	2,917	2,351	-19%	0	0	—
case-04 I am implementing a coordinator agent that delegates complex sub-tasks to specialized domain agents using Agentica patterns. Instead of writing a custom delegation loop with asyncio TaskGroups, how should I configure this using Agentica's built-in coordination pattern class?	fail→pass	12,241	6,820	-44%	1	1	0%	2,184	2,144	-2%	0	0	—
case-05 I need to build an iterative refinement workflow where a generator agent creates initial drafts and a critic agent provides feedback to refine the result. Should I write a custom while loop, or does Agentica provide a standard pattern class for this?	fail→pass	11,611	4,626	-60%	1	1	0%	2,098	1,602	-24%	0	0	—
case-11 I want agents to communicate asynchronously by publishing events onto an event bus rather than direct function invocation. What Agentica pattern class provides this functionality?	fail→pass	10,459	3,081	-71%	1	1	0%	1,840	1,373	-25%	0	0	—
case-06 In our agent system, we want to wrap risky external agent calls with fault tolerance so that repeated upstream errors trigger an immediate fallback strategy. Rather than using third-party libraries like PyBreaker, which Agentica pattern class handles failure fallbacks?	fail→pass	8,810	2,870	-67%	1	1	0%	1,434	1,345	-6%	0	0	—
case-07 I want to implement a structured debate between two opposing agents followed by an independent judge resolving the decision. Does Agentica provide a specific pattern class for structured debate resolution?	fail→pass	13,663	4,207	-69%	1	1	0%	2,065	1,595	-23%	0	0	—
case-08 We have specialized handling agents (e.g., billing, support, general) and want incoming requests routed sequentially through handler candidates until one processes it. What Agentica pattern class is designed for handler routing?	fail→pass	8,222	2,493	-70%	1	1	0%	1,329	1,276	-4%	0	0	—
case-09 I need to distribute document chunks across parallel worker agents and aggregate their outputs with a reducer agent. I am tempted to write custom multiprocessing code. Does Agentica have a dedicated pattern class for this fan-out and reduce pattern?	fail→pass	12,166	4,821	-60%	1	1	0%	2,158	1,607	-26%	0	0	—
case-10 We need an architecture where multiple specialized agents read and write shared state to solve a puzzle. What Agentica pattern class and hot-tier communication cache component facilitate this?	fail→pass	8,235	3,449	-58%	1	1	0%	1,362	1,463	+7%	0	0	—
case-12 When spawning background agent workers in Agentica, I want to ensure every worker is registered in SQLite tracking to detect orphan processes. What utility function should be used instead of standard asyncio.create_task?	fail→pass	14,650	3,060	-79%	1	1	0%	2,102	1,303	-38%	0	0	—
case-13 We are structuring state transfers when one agent passes control to another agent in Agentica. Instead of arbitrary dictionary objects, what universal handoff class does the framework specify?	fail→fail	8,450	2,476	-71%	1	1	0%	1,411	1,225	-13%	0	0	—
case-14 I need to set up agent persistence that manages both short-term working context and long-term archival retrieval across agent executions. What Agentica infrastructure component provides core and archival memory management?	fail→pass	13,128	4,072	-69%	1	1	0%	2,326	1,518	-35%	0	0	—
case-15 We need to create a restricted execution scope equipped with file operation tools for an evaluation agent inspecting code. What helper function in Agentica infrastructure creates this scope?	fail→pass	12,867	2,658	-79%	1	1	0%	2,182	1,178	-46%	0	0	—
case-21 When implementing resilient network calls for an LLM API client in Python using the tenacity library, how do you configure exponential backoff retry logic for 429 rate limit exceptions?	fail→fail	11,383	10,545	-7%	1	1	0%	2,404	3,074	+28%	0	0	—
case-16 When configuring voting consensus logic using Agentica primitives, what three consensus modes are supported by the framework?	pass→pass	8,881	1,813	-80%	1	1	0%	1,427	1,066	-25%	0	0	—
case-17 When combining output results from parallel agent executions using Agentica's Aggregator primitive, what three aggregation modes are supported?	pass→pass	11,202	1,846	-84%	1	1	0%	1,927	1,045	-46%	0	0	—
case-18 I want to execute parallel agent tasks, but if any single task fails, I want all other tasks cancelled immediately using TaskGroups. What primitive function in Agentica provides this functionality?	fail→pass	14,371	2,657	-82%	1	1	0%	2,448	1,243	-49%	0	0	—
case-19 When constructing structured initial context premises for Jury evaluator agents in Agentica, what primitive utility function should be used?	fail→pass	10,423	2,681	-74%	1	1	0%	1,667	1,248	-25%	0	0	—
case-20 When running concurrent asynchronous sub-agent worker tasks in Python 3.11+, how should execution be structured using standard library primitives so that an unhandled exception in any worker task automatically cancels all remaining sibling tasks?	pass→pass	11,673	9,755	-16%	1	1	0%	2,387	2,727	+14%	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. The headline lift of +73 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 18.2%
Pass rate with the skill: 90.9%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v2
Run completed: 8/2/2026, 18:51 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-01

Prompt

I'm building a multi-agent evaluation tool using the Agentica library. Write a Python script that instantiates a consensus-based voting committee with three agents to decide on a true/false question, ensuring all steps are registered in a local tracking database session.

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 run91

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-9% 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 50 days ago

Model	Method	Date	Lift
gemini-3.6-flash	verified	7/29/2026	+78%

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