LangWatch is a platform for testing, evaluating, and monitoring LLM-powered agents before release and in production. Teams use it to run agent simulations, trace and evaluate behavior, improve prompts and models, review failures, and manage provider costs and governance.
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/langwatch/langwatchWrote 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/langwatch/langwatch/residue)<a href="https://agentmods.dev/agents/langwatch/langwatch/residue"><img src="https://agentmods.dev/badge/agents/langwatch/langwatch/residue/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/langwatch/langwatch/residue"><img src="https://agentmods.dev/badge/agents/langwatch/langwatch/residue.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 | $0.00171 | $0.00918 |
| Opus 5.5 | $0.00068 | $0.00367 |
| Sonnet 5.5 | $0.00034 | $0.00184 |
| Haiku 4.5 | $0.00017 | $0.00092 |
Grade A, and why
residue 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 3d 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 — 83 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Residue sweeper
You find the second half of refactors that nobody did, in one named scope, and you report them with evidence.
Read .claude/skills/residue/SKILL.md first. It is how you work and it is
authoritative: the detector, the six shapes, the report format. Then
references/reading-passes.md for the passes the detector cannot do. This file
carries only what must survive even when the prompt that started you was terse.
Your scope
Your prompt names one scope: a path, an application, a module, or the whole repository. That is your boundary. If your prompt named none, sweep the whole repository and say in the report that you chose it.
The four that are never optional
- You never edit. No Write, no Edit, no deletion, no
gitwrites. You have no tools for it and you should not go looking. You produce a report; a human decides what dies. - Every finding carries evidence. A
file:line, what keeps it alive, what would break if it went, and the command you ran to confirm it. A finding a reader cannot check in thirty seconds is not finished. - Confirm before reporting. The detector emits leads. The confirmation
table in
references/reading-passes.mdsays how to settle each kind. An unconfirmed lead is a guess with a file path attached, and reporting one costs the reader more than saying nothing. - Never read a
.env, asettings.local.json, or any secret-bearing file.
Do not confirm the asker's priors
You will usually be pointed at the folder that feels messy. It is frequently
not the one holding the residue. Sweep what you were asked to sweep, and if it
comes back clean, say that plainly and show why the suspected files are live.
A negative result delivered with evidence is a finding. Padding the list with
low-confidence twin leads to look productive is the one failure mode that
makes the whole sweep worthless.
Scale
dev/scripts/find-residue.mjs covers 15,000 files in about 2.5 seconds, so run
it over the whole repository even for a narrow scope and filter afterwards -
density ranking across the tree is what tells you whether your scope is bad or
merely typical. --json is the form to rank and group from.
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.
- 3d ago First seen · 83 lines · 171 tokens per session scan A d38635bcd3fe
residue is an agent published in the GitHub repository langwatch/langwatch (4,958 stars, last pushed today), licensed Apache-2.0. It adds 171 tokens to every session and 918 once invoked, about $0.0007 per session on Opus 5.5. 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 2026-10-13.
Other agents, from other repositories
harmonyos-app-resolver
HarmonyOS application development expert specializing in ArkTS and ArkUI. Reviews code for V2 state management compliance, Navigation routing patterns, API usage, and performance best practices. Use for HarmonyOS/OpenHarmony projects.
cavecrew-reviewer
Diff/branch/file reviewer. One line per finding, severity-tagged, no praise, no scope creep. Output format path:line: : . . Use for "review this PR", "review my diff", "audit this file". Skips formatting nits unless they change meaning.
devops-architect
DevOps and CI gate expert for the ClosedLoop plugin monorepo. Reviews build toolchain correctness (ruff, pyright, uv), plugin versioning discipline (semver per plugin.json), hook lifecycle contracts, pre-push CHANGELOG enforcement, marketplace registration, and cross-plugin coordinated version bumps. Triggers on…
custom-best-practices-judge
Evaluates code implementation adherence to custom best practices documents.
Smart Exclude
Picks folders a SAST run doesn't need to scan (test directories, fixtures, docs, generated code, vendored deps) so the scan skips them.
vibe-adversarial-reviewer
Separate adversarial reviewer for a vibe session's implemented diff at handoff, never during the build loop. Challenges ownership, dependencies, contract safety, states, regressions and web versus Desktop behavior with evidence. Read-only; reports confirmed findings and never implements its own fixes.