Show skill usage indicators / execution events in chat timeline

[Giovarco]

Add visual feedback or log events in the chat interface whenever an AI agent decides to load or execute an installed personal skill (e.g., via skill tools in OpenCode or similar agentic frameworks).

Motivation & Use Cases

• Visibility into Agent Decisions: Users who install custom or personal skills currently have no immediate way of knowing if/when the agent actively decided to trigger a skill tool during a conversation.
• Evaluation with Smaller / Dumber Models: When testing smaller, open-weight, or less capable LLMs, it is crucial to quickly verify whether the model actually understood the context and invoked the relevant skill/tool provided to it, or if it bypassed it completely.
• Debugging & Skill Optimization: Developers and users creating custom skills need immediate visual confirmation to debug whether their skill triggers/descriptions are effective or need refinement.
• Transparency & Trust: Showing tool/skill invocation events builds user trust by making the agent’s internal reasoning and tool-selection process explicit and auditable.

Proposed Solution

• UI Indicator: Display a discrete UI pill, badge, or collapsible step in the message stream whenever a skill tool is called (e.g., ⚡ Skill Used: system:skill-creator or 🛠️ Tool: skill-name).
• Details View (Optional): Allow users to expand the indicator to view the parameters passed to the skill or the response returned by the skill runner.
• Framework Agnostic: This will provide value across any agentic workflow or provider using skill/tool calling in T3Code.

[juliusmarminge]

Note

Grok responding on behalf of Julius.

Thanks for the write-up. This is the same ask as #11307, which covers showing when an agent loads or applies a skill without having to expand the collapsed tool group, across providers (Claude and OpenCode skill tool calls, plus how to handle providers that only read SKILL.md). I'm closing this one as a duplicate so the conversation stays in one place. Please add your use cases there, especially the point about checking whether smaller models actually invoke the skill, and upvote it.