Ready to supercharge your AI agents?

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

Get Started Free →

System Overview · vibeeval | DecimalAI

System Overview

SkillScore96Maintained

system-overview

Show how vibecosystem works - agents, skills, hooks, and self-learning pipeline.

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/system-overview/SKILL.md

decimalai skills pull system-overview --out .claude/skills/

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

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

Raw file: /s/system-overview/SKILL.md ↗ · machine-readable summary at /s/system-overview.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

25 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 — +76% lift-20% tokens0% turns96% of whole cases pass with skill (24/25)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	4%	0%
case-03	✗→✓	▲ Improved	-4%	0%
case-04	✗→✓	▲ Improved	-21%	0%
case-05	✗→✓	▲ Improved	-20%	0%
case-06	✗→✓	▲ Improved	-45%	0%

Showing 5 of 25 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/system-overview/SKILL.md

View on GitHub ↗

License MITImported 8/1/2026Version v2

Skill contents

SKILL.md

System Overview

Show users how Continuous Claude works - the opinionated setup with hooks, memory, and coordination.

When to Use

User asks "how does this work?"
User asks "what can you remember?"
User asks "what's different about this setup?"
User runs /system_overview

Response

CONTINUOUS CLAUDE SYSTEM OVERVIEW
=================================

MEMORY LAYER (PostgreSQL + pgvector)
------------------------------------
- 78,000+ temporal facts from past sessions
- Learnings extracted automatically at session end
- Semantic search with embeddings

RECALL: uv run python opc/scripts/recall_temporal_facts.py --query "your topic"

HOOKS (9 event types registered)
--------------------------------
SessionStart    → Load continuity ledger, rebuild symbol index
UserPromptSubmit → Skill activation check, context injection
PreToolUse      → Smart search routing (Grep → TLDR for code)
PostToolUse     → File claims, compiler feedback
PreCompact      → Save state before context compaction
Stop            → Extract learnings, create handoffs
SubagentStart   → Register spawned agents
SubagentStop    → Coordination, handoff creation
SessionEnd      → Cleanup

CONTINUITY SYSTEM
-----------------
Ledger:   thoughts/ledgers/CONTINUITY_CLAUDE-{session}.md
Handoffs: thoughts/shared/handoffs/{session}/*.yaml

Commands:
  /resume_handoff <path>  - Continue from handoff
  /create_handoff         - Create snapshot for transfer

TLDR CODE INTELLIGENCE
----------------------
5-layer analysis: AST → Call Graph → CFG → DFG → PDG
95% token savings vs reading raw files
Auto-intercepts Grep for .py/.ts/.go/.rs files

Pre-built index: /tmp/claude-symbol-index/symbols.json

SETUP
-----
Run: uv run python opc/scripts/setup/wizard.py

Options:
  [1] SQLite only (simple, offline)
  [2] PostgreSQL + pgvector (semantic search)

Key Files

| Component | Location | |-----------|----------| | Hook registration | .claude/settings.json | | Hook implementations | .claude/hooks/src/*.ts | | Rules (auto-injected) | .claude/rules/*.md | | Skills | .claude/skills/*/SKILL.md | | Setup wizard | opc/scripts/setup/wizard.py | | Recall script | opc/scripts/recall_temporal_facts.py | | Store learning | opc/scripts/core/store_learning.py | | Symbol index builder | opc/scripts/build_symbol_index.py |

Environment Variables

| Variable | Purpose | |----------|---------| | CONTINUOUS_CLAUDE_DB_URL | PostgreSQL connection | | VOYAGE_API_KEY | Embeddings (optional) | | BRAINTRUST_API_KEY | Tracing (optional) | | CLAUDE_PROJECT_DIR | Auto-set by Claude Code |

Illustrative Want to see what system-overview 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, 19:37 UTC · 25 cases

Lift

+76%

96% vs 20%

Agent turns

0%

1.0 vs 1.0

Tokens

-20%

1,450.2 vs 1,807.52

fail → pass · 19pass → pass · 5fail → fail · 1

Per-case results (25)

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 new to this environment. Could you explain how this custom Claude setup works under the hood? I'd like a complete breakdown outlining the memory architecture, lifecycle event hooks, context persistence, and key configuration paths.	fail→pass	18,275	14,983	-18%	1	1	0%	3,084	3,205	+4%	0	0	—
case-02 What can you remember across our sessions and what makes this setup different from standard Claude? Please provide a detailed overview of your memory layers, command utilities, and setup options.	fail→fail	13,495	10,782	-20%	1	1	0%	2,265	2,665	+18%	0	0	—
case-03 Run /system_overview to show me how this setup functions. Please format your response as a structured breakdown covering the core components, hook lifecycle, and environment configurations.	fail→pass	11,110	6,204	-44%	1	1	0%	1,913	1,831	-4%	0	0	—
case-04 We are reviewing our custom memory layer configuration. While standard Claude environments rely on transient context, what specific database engine and vector extension store our temporal facts in this setup, and what CLI script handles querying them?	fail→pass	8,471	2,170	-74%	1	1	0%	1,405	1,106	-21%	0	0	—
case-05 Instead of sending whole source files directly to the prompt context, how does the code intelligence component analyze source files before feeding context, and what token savings percentage does it achieve?	fail→pass	13,743	6,037	-56%	1	1	0%	2,342	1,877	-20%	0	0	—
case-06 In typical projects, configuration settings live in package.json or tsconfig.json. Where are event hooks registered in this workspace configuration, and where are their TypeScript source implementations located?	fail→pass	11,092	2,370	-79%	1	1	0%	1,982	1,094	-45%	0	0	—
case-07 Standard agent setups often save session logs in /tmp or .log files. Where are the session continuity ledger markdown files stored on disk, and where are handoff YAML files saved in this project?	fail→pass	10,368	2,560	-75%	1	1	0%	1,665	1,171	-30%	0	0	—
case-08 Rather than running standard npm scripts or pip install commands, which script runs the interactive setup wizard for this environment, and what command tool prefix is used to execute it?	fail→pass	13,338	2,148	-84%	1	1	0%	2,571	1,103	-57%	0	0	—
case-09 When configuring the system setup wizard, what are the two database storage options offered to the user?	fail→pass	9,569	1,588	-83%	1	1	0%	1,564	976	-38%	0	0	—
case-10 While OPENAI_API_KEY is common across LLM tooling, which optional environment variable configures the vector embedding provider key in this setup, and which variable handles tracing?	fail→pass	14,574	1,914	-87%	1	1	0%	2,764	1,060	-62%	0	0	—
case-11 Rather than rebuilding code symbols dynamically in memory on every request, where is the pre-built code symbol index stored in the file system?	fail→pass	11,192	2,114	-81%	1	1	0%	2,142	1,070	-50%	0	0	—
case-12 What exact slash commands exist in this environment for resuming a session from a saved snapshot and for creating a new snapshot transfer?	fail→pass	5,941	2,515	-58%	1	1	0%	952	1,059	+11%	0	0	—
case-13 When searching code with Grep, what smart routing action does the PreToolUse hook perform for source files?	fail→pass	8,929	2,364	-74%	1	1	0%	1,391	1,075	-23%	0	0	—
case-14 What specific operational tasks are executed by the PostToolUse event hook during file operations and build steps?	fail→pass	11,296	7,249	-36%	1	1	0%	1,796	1,948	+8%	0	0	—
case-15 When a session halts, what dual responsibilities are handled by the Stop event hook?	fail→pass	6,352	2,588	-59%	1	1	0%	1,017	1,128	+11%	0	0	—
case-16 Before generating a response to a prompt, what tasks does the UserPromptSubmit hook execute?	fail→pass	13,539	2,813	-79%	1	1	0%	2,145	1,156	-46%	0	0	—
case-17 Which two lifecycle hooks manage subagent execution, and what is their primary coordination role?	fail→pass	10,100	2,696	-73%	1	1	0%	1,704	1,219	-28%	0	0	—
case-18 When context limits are reached and context compaction occurs, what safety step does the PreCompact hook perform?	pass→pass	8,839	2,192	-75%	1	1	0%	1,299	1,031	-21%	0	0	—
case-19 In the workspace configuration table, which script path is designated for storing extracted learnings?	fail→pass	5,346	2,027	-62%	1	1	0%	774	1,075	+39%	0	0	—
case-20 In which project directory are system rules located for automatic context injection?	fail→pass	4,983	1,487	-70%	1	1	0%	908	919	+1%	0	0	—
case-21 Which script path in the project structure is responsible for rebuilding the symbol index?	fail→pass	5,195	1,764	-66%	1	1	0%	724	1,015	+40%	0	0	—
case-22 When searching code with Grep across all files, which specific programming language file extensions trigger automatic code intelligence interception?	pass→pass	10,270	2,058	-80%	1	1	0%	1,652	1,004	-39%	0	0	—
case-23 I need to write a custom TypeScript hook function in `.claude/hooks/src/my-hook.ts` that listens for prompt submissions and validates input length using node's `fs` module. How should I write this TypeScript function?	pass→pass	14,465	8,578	-41%	1	1	0%	2,929	2,510	-14%	0	0	—
case-24 How do I install PostgreSQL and enable the pgvector extension on an Ubuntu 22.04 LTS server using apt and psql commands?	pass→pass	9,792	6,382	-35%	1	1	0%	1,977	1,941	-2%	0	0	—
case-25 What is the syntax for creating an AST (Abstract Syntax Tree) parser using Python's native `ast` module to analyze function definitions in a file?	pass→pass	10,582	7,504	-29%	1	1	0%	2,223	2,017	-9%	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. 25 cases were attempted. The headline lift of +76 percentage points is the difference between those two pass rates over the 25 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 20%
Pass rate with the skill: 96%
Cases attempted: 25
Cases counted toward lift: 25
Cases that regressed: 0
Skill version measured: v2
Run completed: 8/2/2026, 19:37 UTC

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

Prompt

I'm new to this environment. Could you explain how this custom Claude setup works under the hood? I'd like a complete breakdown outlining the memory architecture, lifecycle event hooks, context persistence, and key configuration paths.

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 run96

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-16% 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 51 days ago

Model	Method	Date	Lift
gemini-3.6-flash	verified	7/29/2026	+73%

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