What's new in tl v0.9.4: your agent finds the extra skills your org publishes, and raw-SQL detours end sooner
The extra skills your organization publishes are now discoverable to your AI agent the first time it looks for them. And two of the most common raw-SQL wrong turns now come with a helpful hint on the spot, so a curious deep-dive doesn't dead-end on a naming mismatch.
A new release of the tl CLI is out today: v0.9.4. The payoff shows up in the small moments — when your agent notices a capability it didn’t see before, and when its exploratory raw-SQL question comes back with a helpful pointer instead of a puzzling error. Nothing you have to change in how you work; run tl update (or let the auto-updater catch you) and your next session picks it up.
Your agent now finds the extras your organization publishes
Last release introduced a way for your organization to publish extra skills on top of the ones bundled with tl — a workflow tailored to how your team runs outreach, a helper for a specific vertical, an internal playbook — and the tl skill list / download / update / remove commands to install them. All of that shipped and works.
The catch: the main tl skill your agent reads first didn’t actually mention those commands, so agents weren’t reaching for them on their own. If nobody told the agent, the extras stayed invisible.
v0.9.4 documents the whole surface right where your agent will notice it. From this release on, when your account manager says “we’ve published a new skill for that”, you can ask your agent directly:
- “What extra ThoughtLeaders skills does my organization have available? Install any that look useful.”
- “There’s a new skill called
tl-brand-book. Install it and use it to draft the outreach brief for the Acme campaign.” - “Check if any of my installed skills are out of date and update them.”
The agent knows where to go and what to run, without you having to paste the command yourself.
Fewer dead ends when your agent goes deep on the data
For most account managers and marketing users, the agent handles data questions through the standard commands and never touches raw SQL. But for the tougher one-off analyses — things like “pull every sponsorship on channels over 500K subs that renewed in the last six months, and tell me which content categories over-index” — the agent sometimes drops into the raw database lane to answer precisely.
When it does that, the biggest source of wasted turns has been column names that sound right but aren’t. Two of the most common ones, straight from usage logs:
- The agent tries
channel_urlon the channels table — the actual column is calledurl. - The agent tries
urlon the sponsorships table — the actual column is calledmedia_url.
Neither is a wild guess; they’re the names anyone would try first. Before v0.9.4, the agent would get a bare “unknown column” error and burn a round-trip working out what to try next. Now it gets an inline hint on the spot (“did you mean url?”, “did you mean media_url?”) and gets to the right answer on the next attempt — often invisibly, because it happens in the middle of the same investigation.
You’ll notice this indirectly: your deep-dive research prompts feel a little snappier, and questions that used to occasionally stumble in the middle just… don’t. Try one of the shapes that tends to trigger the raw-SQL lane:
- “Show me sponsorship URLs and view counts for every ad my top-100 channels ran in Q2, ranked by views per subscriber.”
- “For channels that started sponsoring us this year, list their channel URLs and the sponsorship titles that ran alongside ours.”
Either question would have been the kind that occasionally tripped on the naming mismatch. In v0.9.4, it lands cleanly.
A little housekeeping
One last fix worth naming: the plugin metadata file that Claude Code and the other AI agents read to identify your tl installation had drifted a version behind after v0.9.3. All three version files are now in sync, so when your agent asks “what version of tl am I on?”, every part of the answer agrees.
Updating
Run tl update to grab v0.9.4, 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, and tl skill list is a nice starting point if you’ve been meaning to check what extras your organization has published.
Enjoy the smoother sessions.