Eight PRs merged this week across four repos.

Discord feedback now lands somewhere real

This one spans two repos. ChatMaster (our Discord bot) got a /feedback command: type whatever's on your mind, a mini-LLM classifies it into product, type, sentiment and a title (with hard fallback defaults if the model gets weird), and you get an ephemeral preview with a product-override dropdown before you submit. Behind the scenes it queues in a local outbox for reliable delivery. On the ChameleonLabs side we added the suggestions table it delivers into, plus an authenticated ingest endpoint. Web-form suggestions land in the same table now, and /suggest and /report suggest became preset entry points into the same wizard. All of it ends up in one table.

ChatMaster learned some manners

Issue #97: ChatMaster always replied in multi-bot channels, even after /hush. An earlier diagnosis turned out to be wrong — the production logs showed the real mechanism. The multi-LLM personality path ignored /hush entirely, and personality matching fired on bare word occurrences, so ordinary prose that merely mentioned a personality's name counted as a summons. Both problems are fixed, and personality turns now go to the model as the user role.

Two smaller wins landed too. /channel guide gives everyone a cached, ephemeral channel directory grouped by category, with short AI-generated one-liners and a link to a hosted HTML version (plus a tri-state collapse for the docs links). And /help now role-gates its elevated links: moderators see the Moderator docs and Discord Administrators see Moderator plus Admin. Everyone else sees neither.

Five plugins and one stubborn 400

We packaged five general-purpose skills from our public config mirror as one-command marketplace plugins: repo-hygiene, docs-toolkit, graphify, yt-transcript and model-registry. If you've wanted our version-manager or documentation skills without copying files around, that's the easy way now.

Install today with

/plugin marketplace add Chameleon-Labs-LLC/plugins.

Or, if you've already installed it use /plugin marketplace update Chameleon-Labs-LLC/plugins

See the complete list here

ImageAI improved Anthropic API error handling

And over in ImageAI (v0.44.0), we chased down an Anthropic 400 error: "temperature is deprecated for this model." The Claude 5 family rejects temperature, top_p and top_k, but LiteLLM 1.89.0's capability table wrongly claims claude-opus-5 supports them, so the usual drop_params safety net never dropped anything. Instead of patching the one dialog that crashed, we added a unified parameter validation layer for the whole app.

Try the feedback loop

If you're in our Discord, run /feedback and tell us something — a gripe, an idea, anything. It goes straight into the table we built this week, and we read what comes in. What should we build next?