What's new in tl v0.9.3: build sourcing workflows from your AI agent, and install extra skills your org distributes

A new release of the tl CLI is out today: v0.9.3. The headline is workflows: your AI agent can now design a multi-stage sourcing funnel and hand it to the ThoughtLeaders app as a real, openable workflow — no more assembling stages by hand after the research is done. Alongside that, there’s a new way for your organization to publish extra skills that your agent will install and update automatically, and tl balance now shows you where your credits are coming from. Claude Code, Gemini, Codex, and OpenCode all pick everything up on the next tl update.

Workflows, built by your agent

Until now, the research half and the assembly half of sourcing lived in two different places. Your agent would find you a strong shortlist of channels via tl-keyword-research or the recommender, and then you’d flip over to the app and rebuild that shortlist as a workflow — a query stage, a filter stage for the guide brands, another for the renewal winners, another for the look-alikes. It worked, but it was a lot of clicking to reproduce something the agent had already figured out.

v0.9.3 closes that gap. Two new things ship together:

• tl-create-workflow — a new AI skill that turns a sourcing goal (“find me creators for a home-improvement brand who look like our best renewals”) into a proper populated workflow blueprint: an entry query stage plus the guide-brands → renewal-winners → look-alikes → value-vs-price funnel the TL methodology teaches. The skill delegates the entry stage sourcing to the tools that are best at it (keyword research, channels search, the recommender), keeps the funnel shape shallow and readable, and hands you a blueprint you can create directly.
• tl workflow create — a new command that takes a blueprint (from the skill, or hand-written) and creates the workflow in the app in one call. You get back a workflow ID and an “Open in app” link. The workflow it creates is identical to one you’d build by hand — same stages, same subscriptions, same everything — and shows up in the web list and detail views right away.

The prompt shapes that unlock this look like:

• “I want to source YouTube channels for a mattress brand. Their best renewals are Kelly Stamps and Casey Neistat. Build me a workflow that finds look-alikes to those two and filters down to channels that could clear a $30k eCPM guarantee.”
• “Design a workflow for a coffee brand launching to gen-Z: a keyword-research entry stage on ‘coffee’, filtered by our top three guide brands in the category.”
• “Turn the shortlist we just researched into an actual workflow in the app — I want to send it to the account manager.”

The agent walks the funnel design with you, shows you the blueprint before it creates anything, and only then makes the call. If your account isn’t ready for the direct-create path yet, the skill falls back to hand-off instructions for the app — the blueprint is the same either way. Nothing changes about how workflows behave once they exist; this just removes the “now translate what my agent found into stages” step from the middle.

A skill delivery system for your organization

Bundled skills (the ones that ship with tl — tl, tl-keyword-research, tl-save-report, and friends) cover the shared surface everyone uses. But the ThoughtLeaders team, and organizations working with us, sometimes want to distribute extra skills on top — a workflow tailored to how one team runs outreach, a helper for a specific vertical, an internal playbook. That used to mean copy-pasting files. Now there’s a proper channel for it.

v0.9.3 introduces four new commands:

• tl skill list — see the extra skills your organization has made available (with the ones you’ve already installed marked).
• tl skill download <name> — install one into the same skill folder your bundled skills live in. Your agent picks it up the moment you next invoke it.
• tl skill update — refresh your installed extras (checked quietly in the background on every tl call; when something’s out of date you get an unobtrusive nudge).
• tl skill remove <name> — uninstall an extra without touching anything else.

Installation is atomic and safety-checked, so a half-downloaded skill can never leave your setup in a broken state. tl setup and tl update leave your installed extras alone — nothing you install with tl skill gets clobbered by a client refresh.

For most users, the practical effect is: when your account manager mentions “we’ve published a new skill for that”, you type tl skill list and tl skill download <name>, and your AI agent knows how to do the new thing.

Where your credits come from

tl balance has always shown you the total credit balance on your account. What it hasn’t shown is how that total is composed — how many of those credits are your plan (top-up) credits, which refill on the plan’s cycle, versus purchased credits, which you bought outright and don’t reset.

v0.9.3 splits the two out. The output now names both pools, in the order they’re spent: plan credits go first, purchased credits stay in reserve. Practically, that means:

• You can see at a glance whether you’re chewing through your plan quota or into your reserve.
• Purchased credits survive past the plan’s session/topup resets, so you always know what’s still there if you hit an unusually heavy research week.
• Older servers that don’t yet return the split still render the classic combined balance — nothing breaks.

Prompts that get more useful with this:

• “What’s my credit balance? How much of that is my plan vs bought?”
• “Am I going to run out of plan credits this cycle at my current usage?”

A couple of small safety fixes

Two quieter changes worth naming:

• Skill installs now reject drive-lettered paths on Windows — the kind of paths that would otherwise let a maliciously-packaged skill escape its install directory. Belt-and-braces defense; the server already rejected them, and now the client does too.
• The skill staleness check has its own short timeout — the background “are my installed skills fresh?” check that runs alongside your tl calls now bounds itself tightly, so it never sits waiting on a slow response and holds up your actual command.

Updating

Run tl update to grab v0.9.3, or just keep working — the auto-updater catches you up on your next call. tl changelog shows the running log of what’s landed since your last release.

If your team has a workflow-shape sourcing question that’s been living in a shared doc, this is the release to ask your agent to turn it into a real one.