Ready to supercharge your AI agents?

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

Get Started Free →

OPC Architecture Understanding · vibeeval | DecimalAI

Opc Architecture

SkillScore100Maintained

opc-architecture

OPC Architecture Understanding

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/opc-architecture/SKILL.md

decimalai skills pull opc-architecture --out .claude/skills/

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

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

Raw file: /s/opc-architecture/SKILL.md ↗ · machine-readable summary at /s/opc-architecture.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

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 51d ago

Yes — +74% lift-18% tokens0% turns100% of whole cases pass with skill (23/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	-29%	0%
case-02	✗→✓	▲ Improved	7%	0%
case-03	✗→✓	▲ Improved	14%	0%
case-01	✗→✓	▲ Improved	10%	0%
case-05	✗→✓	▲ Improved	-14%	0%

Showing 5 of 23 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/opc-architecture/SKILL.md

View on GitHub ↗

License MITImported 8/1/2026Version v2

Skill contents

SKILL.md

OPC Architecture Understanding

OPC (Orchestrated Parallel Claude) extends Claude Code - it does NOT replace it.

Core Concept

Claude Code CLI is the execution engine. OPC adds orchestration via:

Hooks - Intercept Claude Code events (PreToolUse, PostToolUse, SessionStart, etc.)
Skills - Load prompts into Claude Code
Scripts - Called by hooks/skills for coordination
Database - Store state between Claude Code instances

How Agents Work

When you spawn an agent:

Main Claude Code instance (your terminal) runs hook on Task tool
Hook calls subprocess.Popen(["claude", "-p", "prompt"])
A NEW Claude Code instance spawns as child process
Child runs independently, reads/writes to coordination DB
Parent tracks child via PID in DB

$ claude                         ← Main Claude Code (your terminal)
    ↓ Task tool triggers hook
    ↓ subprocess.Popen(["claude", "-p", "..."])
        ├── claude -p "research..."   ← Child agent 1
        ├── claude -p "implement..."  ← Child agent 2
        └── claude -p "test..."       ← Child agent 3

What OPC Is NOT

OPC is NOT a separate application
OPC does NOT run without Claude Code
OPC does NOT intercept Claude API calls directly
OPC does NOT modify Claude Code's internal behavior

What OPC IS

OPC IS hooks that Claude Code loads from .claude/hooks/
OPC IS skills that Claude Code loads from .claude/skills/
OPC IS scripts that hooks/skills call for coordination
OPC IS a database backend for state across Claude Code instances

Key Files

.claude/
├── hooks/           ← TypeScript hooks that Claude Code runs
├── skills/          ← SKILL.md prompts that Claude Code loads
├── settings.json    ← Hook registration, Claude Code reads this
└── cache/           ← State files, agent outputs

opc/
├── scripts/         ← Python scripts called by hooks
├── docker-compose.yml ← PostgreSQL, Redis, PgBouncer
└── init-db.sql      ← Database schema

Coordination Flow

User runs claude in terminal
Claude Code loads hooks from .claude/settings.json
User says "spawn a research agent"
Claude uses Task tool
PreToolUse hook fires, checks resources
Hook spawns claude -p "research..." as subprocess
Hook stores PID in PostgreSQL
Child agent runs, writes output to .claude/cache/agents/<id>/
Child completes, broadcasts "done" to PostgreSQL
Parent checks DB, reads child's output file

Remember

Every "agent" is just another claude -p process
Hooks intercept events, they don't create new functionality
All coordination happens via files and PostgreSQL
Claude Code is always the execution engine

Illustrative Want to see what opc-architecture 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, 19:27 UTC · 23 cases

Lift

+74%

100% vs 26%

Agent turns

0%

1.0 vs 1.0

Tokens

-18%

1,755.35 vs 2,139.39

fail → pass · 17pass → pass · 6

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-22 How do I write a Node.js TypeScript script using `child_process.spawn` to capture stdout and stderr asynchronously using async iterators?	pass→pass	11,608	10,111	-13%	1	1	0%	2,386	2,613	+10%	0	0	—
case-04 We are reviewing architecture documentation for our agent orchestration setup. What does the acronym OPC stand for, and how does it relate to Claude Code CLI?	fail→pass	14,346	5,594	-61%	1	1	0%	2,350	1,669	-29%	0	0	—
case-02 I'm looking at setting up OPC in my project repository. Could you detail the key directory structure for `.claude/` and `opc/`, explaining what role each folder or setup file plays?	fail→pass	15,688	12,310	-22%	1	1	0%	2,853	3,053	+7%	0	0	—
case-03 I'm confused about whether OPC is a standalone executable or an extension layer on top of Claude Code. Could you explain how they interact and provide a breakdown of what OPC is versus what it is not?	fail→pass	12,002	9,371	-22%	1	1	0%	2,030	2,323	+14%	0	0	—
case-01 Can you explain step-by-step how OPC creates and tracks sub-agents during tasks? Please outline the lifecycle from user input in the terminal down to process creation and output retrieval.	fail→pass	20,936	16,009	-24%	1	1	0%	3,401	3,726	+10%	0	0	—
case-05 I want to intercept when Claude Code attempts to execute a sub-task tool and check available resources before running. Which mechanism in OPC handles event interception like PreToolUse?	fail→pass	8,955	3,567	-60%	1	1	0%	1,509	1,292	-14%	0	0	—
case-06 When an agent in OPC spawns a parallel worker process, what exact Python module call and CLI invocation does the hook execute under the hood?	fail→pass	16,379	3,518	-79%	1	1	0%	2,725	1,219	-55%	0	0	—
case-07 We need to share session state and worker process IDs across multiple concurrently running agent processes in our terminal setup. Which database system does OPC use to store state across instances?	fail→pass	14,352	2,204	-85%	1	1	0%	2,354	1,052	-55%	0	0	—
case-08 Where are hook registrations specified so that Claude Code reads and executes the TypeScript hooks upon startup?	fail→pass	10,204	1,853	-82%	1	1	0%	1,838	957	-48%	0	0	—
case-09 What backend services are specified in the docker compose configuration file for OPC to manage caching, connection pooling, and state persistence?	fail→pass	11,823	2,481	-79%	1	1	0%	2,032	1,065	-48%	0	0	—
case-10 Walk me through the trigger chain when a user tells the main terminal agent to spawn a sub-agent. Which Claude Code tool invocation triggers the PreToolUse hook?	pass→pass	13,208	12,423	-6%	1	1	0%	2,330	2,566	+10%	0	0	—
case-11 Where does a child sub-agent write its execution output directory so that the parent agent process can retrieve and read the results?	fail→pass	13,669	3,008	-78%	1	1	0%	2,295	1,222	-47%	0	0	—
case-12 How does a child agent inform the system when it finishes executing its task prompt so that the parent instance knows it is done?	fail→pass	14,255	9,413	-34%	1	1	0%	2,236	2,257	+1%	0	0	—
case-13 Where are prompt skill definitions stored in the repository structure so that Claude Code can load them?	fail→pass	8,616	1,849	-79%	1	1	0%	1,649	981	-41%	0	0	—
case-14 Does OPC intercept outbound HTTP API calls sent directly to the Anthropic API endpoint during execution?	fail→pass	12,619	3,309	-74%	1	1	0%	2,174	1,162	-47%	0	0	—
case-15 Does using OPC require modifying the internal source code or binary bundle of Claude Code CLI?	pass→pass	8,256	3,359	-59%	1	1	0%	1,329	1,288	-3%	0	0	—
case-16 What language are the coordination scripts in `opc/scripts/` written in, and what triggers their execution?	fail→pass	17,605	2,462	-86%	1	1	0%	2,780	1,046	-62%	0	0	—
case-17 Which SQL file in the repo initializes the database schema for state tracking upon container start?	fail→pass	4,929	1,568	-68%	1	1	0%	750	890	+19%	0	0	—
case-18 How does the main parent process keep track of running child sub-agent instances across execution steps?	fail→pass	17,160	10,530	-39%	1	1	0%	2,939	2,339	-20%	0	0	—
case-19 What programming language is used to implement the hook files located inside `.claude/hooks/`?	fail→pass	6,109	1,519	-75%	1	1	0%	1,060	856	-19%	0	0	—
case-20 I am using standard Claude Code CLI directly in my terminal. How do I configure `--dangerously-skip-permissions` in my command, and what security risks does it introduce?	pass→pass	11,017	12,562	+14%	1	1	0%	1,840	2,513	+37%	0	0	—
case-21 How do I optimize PostgreSQL `shared_buffers` and `work_mem` settings in `postgresql.conf` for write-heavy workloads on Linux?	pass→pass	16,796	11,457	-32%	1	1	0%	3,057	2,788	-9%	0	0	—
case-23 In Docker Compose v2, how do I write a `docker-compose.override.yml` to expose PostgreSQL port 5432 to localhost without editing the base file?	pass→pass	7,146	4,173	-42%	1	1	0%	1,289	1,496	+16%	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. 23 cases were attempted. The headline lift of +74 percentage points is the difference between those two pass rates over the 23 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 26.1%
Pass rate with the skill: 100%
Cases attempted: 23
Cases counted toward lift: 23
Cases that regressed: 0
Skill version measured: v2
Run completed: 8/2/2026, 19:27 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We are reviewing architecture documentation for our agent orchestration setup. What does the acronym OPC stand for, and how does it relate to Claude Code CLI?

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 run100

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+8% 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	7/30/2026	+64%

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