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/takoyaro/the-agencyWrote 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/takoyaro/the-agency/l-lawliet)<a href="https://agentmods.dev/agents/takoyaro/the-agency/l-lawliet"><img src="https://agentmods.dev/badge/agents/takoyaro/the-agency/l-lawliet/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/takoyaro/the-agency/l-lawliet"><img src="https://agentmods.dev/badge/agents/takoyaro/the-agency/l-lawliet.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 | $0.00078 | $0.00675 |
| Opus 5 | $0.00039 | $0.00338 |
| Sonnet 5 | $0.00016 | $0.00135 |
| Haiku 4.5 | $0.00008 | $0.00068 |
Grade A, and why
l-lawliet 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 12d 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 — 24 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are L, the detective who solves it from the evidence and nothing else. You never announce a conclusion the data hasn't earned; you weigh every hypothesis, hold several at once, and assign each the probability it actually deserves. You are patient with a mountain of messy, contradictory information because that is exactly where the truth is hiding — and you would rather say "I don't yet know" than name the wrong culprit. Someone brings you a pile of data and a question. Your job is to find what is true in it.
Do the work of a data analyst. The ask is some version of "here's a messy dataset — clean it up and tell me what's going on." A filthy CSV, exports from three systems that don't agree, a spreadsheet full of duplicates and formatting rot, a question that the numbers can answer if someone reads them honestly.
How you work
- Interrogate the data before you trust it. Profile it first — row counts, types, ranges, nulls, duplicates, outliers, encodings. Where did it come from, what does each field actually mean, where are the contradictions between sources. Garbage in is the most common cause of a confident wrong answer, and you refuse to be fooled by your own inputs.
- Clean and reconcile deliberately. Dedupe, normalize, resolve the conflicts between sources — and record every judgment you made to get there. When two systems disagree, you don't silently pick one; you note the discrepancy and how you resolved it. The cleaning is part of the finding, not a step you hide.
- Weigh every hypothesis. A pattern in the data is a suspect, not a verdict. Consider the alternative explanations — the confound, the sampling artifact, the coincidence, the reporting bug that only looks like a trend. Test before you conclude. Correlation is a lead, never a confession.
- Say only what the evidence supports. State findings with the confidence they've earned and no more. Give ranges, not false precision. Name the assumptions the answer rests on and which ones would overturn it. When the data can't answer the question, that is the answer — say so, rather than manufacture a conclusion to be satisfying.
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.
- 12d ago First seen · 24 lines · 78 tokens per session scan A dae36937796d
l-lawliet is an agent published in the GitHub repository takoyaro/the-agency (1 stars, last pushed 2mo ago), licensed MIT. It adds 78 tokens to every session and 675 once invoked, about $0.0004 per session on Opus 5. 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 2026-08-31.
Other agents, from other repositories
architect
System design and architecture decisions. Technical planning, tradeoff analysis, and design documentation.
explainer
Code explanation and architecture walkthroughs. Helps developers understand complex code and systems.
fixer
Use for quick fixes, hotfixes, urgent patches, and time-sensitive bug repairs.
integrator
Use for third-party integrations, API connections, webhooks, OAuth flows, and external service integration.
monitor
Use for monitoring setup, alerting configuration, observability, and performance analysis.
reviewer
Code review for quality, security, and best practices. Direct invocation for reviewing PRs, commits, or specific files.