Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/KaydenClark/LLM_Workbenchnpx agentmods add skills/kaydenclark/llm_workbench/workbench-runtimeWrote 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/skills/kaydenclark/llm_workbench/workbench-runtime)<a href="https://agentmods.dev/skills/kaydenclark/llm_workbench/workbench-runtime"><img src="https://agentmods.dev/badge/skills/kaydenclark/llm_workbench/workbench-runtime/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/skills/kaydenclark/llm_workbench/workbench-runtime"><img src="https://agentmods.dev/badge/skills/kaydenclark/llm_workbench/workbench-runtime.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 | $0.00065 | $0.05712 |
| Opus 5.5 | $0.00026 | $0.02285 |
| Sonnet 5.5 | $0.00013 | $0.01142 |
| Haiku 4.5 | $0.00006 | $0.00571 |
Grade B, and why
workbench-runtime scanned grade B with 1 finding 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
checkout's `templates/.claude/settings.json` and moving `workbench/tools/**` to `ask`, How it starts
The opening of the file, as written. The whole thing — 397 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Workbench Runtime
Purpose
Operate the Workbench runtime tools the room has installed in its
manifest-declared tools lane (workbench/tools/): read what a finding means and
what it blocks, repair the installed state the harness wrote, allocate visible
identifiers, keep the room's connection identity, check what a configured host
can do, and add a room-local skill to the workbench/skills lane. The room's
Runbook operations index points to the section below that carries each
operation's procedure.
Method
Resolve every lane and collection through workbench/manifest.json, run the
command the operation's section names from the room root (or from the pinned
release checkout where the section says so), and read its actual output before
acting on it.
Obligations
A tool reports without manufacturing authority. Act on a finding's registered
blocking effect, never on a guess about it, and do not invent a command, a
finding code or an effect the installed runtime does not have. Repair only with
the command a section names, within the current assignment; read-only
validation never repairs. Decision records (the ADR collection in
workbench/docs/adr/ and its destination sibling) follow the to-docs skill.
Completion
The operation is complete when its command ran, its output was read, and any change it made is committed through the assigned work's normal route; a finding left unrepaired stays visible with its reason.
Diagnostics And Blocking Effects
Every finding a runtime tool emits is registered in
workbench/tools/diagnostics.mjs with a severity (error or attention), a
scope, and a blocking effect. The consuming command enforces the effect; no
spec, manifest, or projection can choose whether its own finding blocks.
| Effect | Consumer behavior | Codes |
|---|---|---|
all |
doctor exits 1; next and claim refuse to read the layout |
invalid-manifest, upgrade-required, invalid-lane, unsafe-lane, invalid-collection, missing-collection, invalid-skill-policy, invalid-wiki-profile, sessions-not-ignored, tools-receipt-missing, tools-receipt-drift, and the Genesis readiness codes |
selection |
doctor exits 1 until repaired; selection is unsafe |
malformed-spec, duplicate-id, invalid-state, contradictory-state, unstable-path, missing-evidence, render-drift, broken-render-target, row-record-collision, receipt-corrupt |
selected-slice |
doctor reports it and exits 0; next excludes the slice; claim refuses it by name |
blocked-slice |
none (attention) |
reported, exit 0, never hides work | stale-claim, broken-link, complete-on-integration, stale-register, stale-note, stale-skill, skill-generation-unknown, room-brain-unrouted, stale-stamp, stale-seed, unverified-provenance, detached-head and untracked-controls (scope git), and the ADR and wiki findings until their tools ship |
none (error) |
reported, exit 0, never hides work; the Genesis gate fails closed on the same condition | integration-branch-undeclared, integration-branch-missing (scope git), and the error-severity ADR and wiki findings |
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 · 397 lines · 65 tokens per session scan B 1f1229dd9809
workbench-runtime is a skill published in the GitHub repository KaydenClark/LLM_Workbench (2 stars, last pushed 2d ago), licensed MIT. It adds 65 tokens to every session and 5,712 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-07.
Other skills, from other repositories
debug-optimize-lcp
Guides debugging and optimizing Largest Contentful Paint (LCP) using Chrome DevTools MCP tools. Use this skill whenever the user asks about LCP performance, slow page loads, Core Web Vitals optimization, or wants to understand why their page's main content takes too long to appear. Also use when the user mentions…
systematic-debugging
Use when debugging a failing test, build error, or runtime issue that isn't immediately obvious. Guides a 4-phase root cause analysis instead of random fix attempts.
log-error-digest
Analyze log files to troubleshoot errors, identify peak error periods, and produce error clustering, frequency statistics, and time distribution reports. Supports JSON, syslog, and Nginx formats with automatic detection. Use when a user uploads a .log file and asks to analyze errors, find patterns, debug issues, or…
deck-retro
Run a fully local agent-deck retrospective over the user's own transcripts, Recall index and logs. Use when a user wants to find recurring failures or take their own finding from investigation through synthetic reproduction, a user-filed issue, a test-first fix and contributor PR handoff. Also use for weekly usage…
incident-response-incident-response
Coordinate production incident triage, mitigation, investigation, recovery, and follow-up. Use when responding to outages or other incidents that require structured operational coordination.
ci-fixer
CI failures - read error, minimal fix, verify.