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.
npx skills add KaydenClark/LLM_Workbench --skill to-tasksgit clone --depth 1 https://github.com/KaydenClark/LLM_WorkbenchWrote 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/to-tasks)<a href="https://agentmods.dev/skills/kaydenclark/llm_workbench/to-tasks"><img src="https://agentmods.dev/badge/skills/kaydenclark/llm_workbench/to-tasks/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/to-tasks"><img src="https://agentmods.dev/badge/skills/kaydenclark/llm_workbench/to-tasks.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 | $0.00018 | $0.01611 |
| Opus 5.5 | $0.00007 | $0.00644 |
| Sonnet 5.5 | $0.00004 | $0.00322 |
| Haiku 4.5 | $0.00002 | $0.00161 |
Grade A, and why
to-tasks 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 — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
To Tasks
The assigned SPEC.md names the destination. Each approved slice becomes its
own tasks/<TK-###>/TASK.md record inside the assigned Spec; a Spec that still
holds only the legacy slice table is converted first (step 4), and its table is
history, never a place for a new slice. The assigned Spec is the only store.
TASKBOARD.md is a generated projection, not a second tracker.
Confirm the assigned spec resolves through workbench/manifest.json before
changing it. Never recreate a root specs/ queue or a project-local
skill-discovery tree.
Cut Tasks when the Spec is activated (planned -> active), from live
Actuality at the real start of the work; a planned Spec gets no Tasks unless
the same request activates it. If the assigned Spec is still planned and the
request does not activate it, report that its decomposition waits for
activation and write no Task. When the same request activates it, activate it
with node workbench/tools/spec-workbench.mjs convert-tasks S-### --activate,
run once, and never edit its **Status:** by hand:
- A planned Spec that still holds unfinished slice-table rows: the command
turns those rows into
TASK.mdrecords and sets its**Status:**toactivein one step, writing nothing if any record fails to parse. - A planned record-backed Spec (it already has a
tasks/directory;to-specleaves a new Spec that way, with no Task row and an empty trackedtasks/): write its firstTASK.mdrecord(s) as in step 4, requesting each label withnext-id, then run the command once. It converts no row: it refuses, writing nothing, when the Spec has no record or a record fails to parse, and otherwise sets only its**Status:**toactive.
Either way the command changes no other header field, so update
Latest event and Next gate yourself. Tasks already cut into an existing
planned Spec stay as they are.
Process
- Read the assigned
SPEC.mdand the relevant source and tests. Do not select an unassigned spec. - Propose vertical slices that each deliver a checkable behavior in one context.
Apply the
/tracer-bulletdiscipline so each slice pierces every layer of the project's stack rather than one layer. Give each a short outcome-oriented title, real blockers, done criteria, and closing proof. Prefer an independent first tracer bullet; use expand-contract when a wide refactor cannot stay green by vertical slice. - If the owner has not already authorized this decomposition, present the
proposed set and ask for approval before changing durable work state, and
confirm that slice size and blocking edges are right. When the request or
the composing workflow already carries that planning authority, do not ask
again (
workbench/docs/adr/0045-skill-composition-within-inherited-scope.md); state the set you are writing and continue. - With that authority, use
spec-workbench.mjs next-id S-### --prefix TK --jsonfor each new label. It proposes a letter-bearing ID from the whole Workbench inventory without reserving it; save each planned Task before requesting the next. Preserve existing spec-qualified numeric Task references.- If the assigned Spec already has a
tasks/directory, it is record-backed: write onetasks/<TK-###>/TASK.mdper approved slice. It opens with the title line# TK-### - <slice>(the parser refuses a record without it) and carries one**Field:** valueline each forTask ID,Spec ID,Slice,Status(readyunless a real blocker applies),Stance(the normal stance for the executor:Builder,Auditor,ReviewerorReconciler),Blockers(none, or a comma-separated list ofS-###/TK-###ids), andDestination(spec-acceptance: <the Acceptance Criteria line this Task advances>, orwiki-claim: <the Wiki page>only when the Task's own destination is producing that page; a corrective Task never takes a Wiki claim as its destination, and a later gap against delivered work is a new Spec, not a Task). AddPlanned verificationnaming the check the Task expects to run. Add no table rows; a record-backed Spec's retainedVertical Implementation Slicestable is completed history only. When that Spec is stillplannedand the same request activates it, runconvert-tasks S-### --activateonce after writing the records. A record looks like this:
- If the assigned Spec already has a
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 · 122 lines · 18 tokens per session scan A dbf23deb835d
to-tasks is a skill published in the GitHub repository KaydenClark/LLM_Workbench (2 stars, last pushed 2d ago), licensed MIT. It adds 18 tokens to every session and 1,611 once invoked, about $0.0001 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 2026-10-07.
Other skills, from other repositories
story-readiness
Is a story implementation-ready? Checks clear acceptance criteria, open questions, ADR refs. READY/NEEDS WORK/BLOCKED/NOT ASSESSED.
recipe-create-meet-space
Create a Google Meet meeting space and share the join link.
workthreads
SpecStory Workthreads - a weekly work-thread rollup across a team's repos from SpecStory coding histories (any agent - Claude Code, Codex, Cursor, Gemini, and more). It groups the window's sessions into threads of work per project and labels each new / open / recently closed, so a lead sees what shipped, what is still…
atmos-config
Atmos root configuration: atmos.yaml discovery, precedence, deep merging, basepath, imports, minimal bootstrap, and routing to narrower Atmos skills.
projects
List all managed projects with status, branch, open PRs, and open issue counts — portfolio-level view.
handle-linkedin-connection-request-signal
Use this skill when someone sends a team member an inbound LinkedIn connection request. An inbound request is a deliberate, ACTIVE first-party intent signal — meaningfully stronger than a passive profile view — and it deserves different scoring rules. The skill resolves and enriches the requester, gates them through…