Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
GitHub Copilot
Gemini CLI+20 moreClaude CodeCursorCodex CLIGitHub CopilotGemini CLI
OpenClaw
Manus
Grok+17 agentsswarm budget gate
Static audit for multi-agent swarm orchestration to prevent budget overruns caused by recursive spawns and racy ledgers.
$13
swarm budget gate
Example session with this skill installed
Audit this LangGraph repo for swarm-level budget safety. We have a $5 cap on each agent loop, so it should be fine. Check if we need a shared ledger for the parallel branches.
- Read your context and instructions
- Compiled the swarm budget gate
- Generated the document
I have completed the static audit of the LangGraph orchestration code. The swarm currently lacks a shared ledger, creating a significant financial unboundedness risk despite individual agent caps.
Swarm Budget Audit Report
| Metric | Status | Finding |
|---|---|---|
| Verdict | FAIL | Financial ceiling is unenforced across concurrent branches. |
| Spawn Inventory | INCOMPLETE | Found 3 explicit spawns; recursive delegation in tools/researcher.py is untraceable. |
| Shared Ledger | ABSENT | Each agent tracks local state["cost"] with no global aggregation. |
| Kill-Switch | PROCESS-BOUND | Signal handler only halts the main thread; sub-processes will persist. |
Financial Risk Arithmetic
- Per-Agent Cap: $5.00
- Max Concurrent Agents: 10 (per
config.yamlthread pool) - Theoretical Ceiling: $50.00 per execution cycle.
Gap: The orchestrator does not sum spend across branches. If 10 agents run in parallel, the swarm spend is 10x the assumed $5 limit.
Code Analysis: Race Condition Risk
The current implementation in nodes/execute.py uses a read-then-write pattern:
# FAIL: Read-then-write race condition
if current_spend < MAX_BUDGET:
result = await agent.run()
update_spend(result.cost)
Under concurrent dispatch, multiple agents will read the same current_spend value before any have written their updates, bypassing the $5 cap entirely.
Next steps
- Implement a reservation-based ledger using an atomic increment (e.g., Redis
INCRBYor SQL row lock). - Move the kill-switch flag to a shared durable store like a file or database.
- Explicitly pass a shared
BudgetContextobject to theresearcher.pytool to track recursive costs.
swarm-budget-gate.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
What you get
About this skill
Multi-agent swarms can route around individual agent caps by spawning recursive subagents or parallel branches. If your orchestrator lacks a shared ledger, the total bill is the sum of every cap across every agent spawned, which often lacks a hard ceiling. This skill provides a static audit of orchestration code to identify these financial blind spots before deployment.
What it does
- Spawn point inventory identifies every location where agents are instantiated, including recursive delegation and tool-triggered spawns.
- Ledger atomicity check verifies if shared budget tracking uses reservation-based atomic operations to prevent races during concurrent dispatches.
- Kill-switch validation audits whether a swarm-wide halt persists if the primary orchestrator process crashes or is killed.
- Theoretical ceiling calculation computes the actual financial risk by multiplying per-agent caps by maximum concurrent agent counts.
- Remediation mapping recommends specific runtime primitives or generates scoped ledger snippets to close identified gaps.
How it works
- Run the
Spawn Inventorycommand to grep the entire repository for framework-specific instantiation primitives. - Analyze the budget tracking logic to determine if spend is aggregated across all spawn points or isolated per agent.
- Test the kill-switch durability against externally stored states rather than local process flags.
- Review the final audit report for a PASS, FAIL, or INCOMPLETE verdict based on ironclad concurrency and enumeration rules.
Frameworks & tools
Works with LangGraph, CrewAI, AutoGen, and custom recursive-delegation harnesses. Compatible with Python and TypeScript orchestration layers.
Why this beats prompting it yourself
Generic prompts often overlook the race conditions inherent in concurrent agent dispatches. This skill enforces a strict "reserve-before-execute" logic and repo-wide spawn tracing that prevents agents from bypassing budget gates through retries or tool calls.
Use cases
- Auditing a LangGraph swarm for recursive subagent spend before production release.
- Investigating budget overruns where individual agents stayed under their local caps.
- Designing a race-free shared ledger for a high-concurrency CrewAI implementation.
Known limitations
Does not enforce budgets at runtime. Static analysis cannot fully trace dynamic agent construction from external webhooks or plugin registries.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean 1 day ago
- Passed all security checks, Safe to install