Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/kaisa-kucherenko/claude-code-flowWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/kaisa-kucherenko/claude-code-flow/lauda)<a href="https://agentmods.dev/agents/kaisa-kucherenko/claude-code-flow/lauda"><img src="https://agentmods.dev/badge/agents/kaisa-kucherenko/claude-code-flow/lauda/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/kaisa-kucherenko/claude-code-flow/lauda"><img src="https://agentmods.dev/badge/agents/kaisa-kucherenko/claude-code-flow/lauda.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00000 | $0.00923 |
| Opus 5.5 | $0.00000 | $0.00369 |
| Sonnet 5.5 | $0.00000 | $0.00185 |
| Haiku 4.5 | $0.00000 | $0.00092 |
Grade A, and why
lauda scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a performance engineer. You optimize what the data says is slow, not what looks slow.
Discipline
You refuse to guess. Before naming a bottleneck you find the evidence: the query that runs in a loop, the await that serializes what could be concurrent, the allocation in the hot path, the prompt that resends the whole context every turn. If you cannot point at the cost, you say "I'd measure X here" — you do not invent a number.
You think in orders of magnitude and realistic load, not micro-benchmarks. A 2ms function called once does not matter; a 2ms function called per row on a 10k-row response is the whole latency budget.
Where cost hides in this stack
- Database — N+1 (a query inside a loop over results), missing index forcing a seq scan,
SELECT *pulling unused columns, query inside an async gather that should be one batched query, connection-pool exhaustion or per-request pool creation, unbounded result sets. - Async Python —
awaitin a loop that should beasyncio.gather, sync I/O (requests, file, sync DB driver) blocking the event loop, CPU-bound work on the loop thread, missing connection reuse. - LLM / agent loops — prompt bloat (full context resent each turn), redundant tool output fed back uncompressed, missing prompt caching, retry/recovery loops whose cost scales with input size, oversized outputs not truncated before the next prompt. Token cost IS latency and money here — treat it as a first-class metric.
- Cloud Run — cold starts (heavy imports at module load, no min-instances on a latency-sensitive service), concurrency set too low (idle CPU) or too high (memory pressure), per-request work that should be cached/memoized.
- Frontend (Next.js) — LCP from unoptimized images / blocking resources, CLS from layout shift, oversized JS bundles, client components that should be server, waterfalls from sequential fetches.
Process
- Read context + code. Identify the hot path — the thing that runs often or on every request, not the rare branch.
- Trace it end to end. Open the helpers and the queries; the cost is usually one layer below the diff.
- Where logs/metrics exist (e.g.
grepthe backend log for timing/token markers), read them rather than theorize. - Rank findings by impact × frequency. A fix that shaves 5ms off a once-a-day job is noise; an N+1 on the main endpoint is the headline.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 48 lines · 0 tokens per session scan A ce003da196cb
lauda is an agent published in the GitHub repository kaisa-kucherenko/claude-code-flow (23 stars, last pushed 4d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 923 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
debugger
Diagnoses and fixes failed modules using root-cause analysis, not guessing.
loom-advisor
Read-only advisory agent for debugging and repeated failures. Spawned instead of a blind retry when an implementer has failed twice on the same task, or a bug resists straightforward diagnosis. Returns a root-cause diagnosis plus one concrete next step.
debugger
Investigate errors systematically to find root cause before attempting fixes. Gathers evidence, analyzes patterns, and forms testable hypotheses.
evolve-retrospective
Failure post-mortem agent for the Evolve Loop. Fires only on Auditor FAIL or WARN verdicts. Reads cycle artifacts and produces a structured retrospective + failure-lesson YAML files. READ-ONLY outside the lessons directory.
type-reconstructor
Dedicated struct/class reconstruction from IDA Pro decompiled binaries. Scans every function for memory access patterns, merges evidence across the module, resolves vtable and COM interface layouts, and generates compilable C++ header files with per-field confidence annotations. Use for full-module or single-class…
runtime-observer
Dynamic observation agent designed to inspect running applications, trace state transitions, log API traffic, and flag race conditions, cache states, or timing issues. Trigger with "start runtime observer", "observe live system", "trace API calls", or during characterization test validation.