Restore support for custom prompt/command files in Positron Assistant

[randrescastaneda]

We (and I suspect other teams) built workflows around custom prompt files (.github/prompts/*.prompt.md, .vscode/positron/prompts/, etc.) that were previously usable as /name slash commands in Positron Assistant. That capability appears to no longer work — Posit Assistant does not read or expose these files as commands, and there is no setting (including chat.promptFilesLocations, which is a VS Code-core setting Posit Assistant doesn't consult) that restores it.

Why this matters

Several of us have built non-trivial systems on top of custom prompts/commands: multi-step workflows, team conventions, orchestration across agent files, shared contracts between prompts, etc. Losing native support means re-engineering these systems from scratch, which is a significant cost for teams that had already standardized around this pattern. This is especially frustrating because:

• The rest of the VS Code-adjacent ecosystem — VS Code itself (via Copilot Chat), Cursor, OpenCode, Claude Code, and Codex — all support user-defined prompt files or custom commands as a first-class, low-friction feature.
• Positron is a VS Code fork, so users reasonably expect this class of customization to carry over or have a clear equivalent.

Prompts/commands are not the same as skills

We understand Posit Assistant's answer to this is "convert your prompt files into skills" (SKILL.md, auto-loaded by description relevance). We want to push back on that being treated as a drop-in replacement:

• Skills are context-loading modules, not invocable workflows. They're designed to be discovered by relevance to a task description and loaded into context — not necessarily to be deterministically triggered on demand the way a slash command is.
• Prompts/commands are explicit and deterministic. When I type /cg-brainstorm (one of commands of this plugin ), I want that exact workflow to run, right now — not for the assistant to decide it's relevant based on matching my message against a description. For teams with prompt files that orchestrate multi-step processes (referencing other agent files, shared contracts, specific tool sequences), that determinism is the point.
• Not every prompt maps cleanly to "specialized knowledge." Some of our prompt files are procedural scripts (e.g., "run this exact sequence of steps against this file with these guardrails") rather than domain knowledge to load. Forcing these into the skill format loses the explicit-invocation guarantee and adds migration overhead for something that used to be a plain markdown file in a folder.

Ask

Could Posit Assistant add native support for user-defined prompt/command files (e.g., recognizing .github/prompts/*.prompt.md, or a Positron-specific equivalent), independent of the skills system? At minimum, it would help to have:

1. A documented, first-class way to register custom slash commands from markdown files (not routed through skill discovery).
2. Compatibility with (or an explicit statement about) chat.promptFilesLocations and related VS Code chat settings, since those are the conventions many of us already have in place.
3. Clarity in the docs on the intended distinction between "skills" and "commands/prompts," since right now the skills docs read as if they're meant to cover this use case, but in practice they don't behave the same way.

Happy to share examples of the workflows this would unblock if useful.

Thank you so much!!

[seeM]

Thank you for the thoughtful and detailed feedback @randrescastaneda! I've created a feature request for this over here https://github.com/posit-dev/positron/issues/15066. I'll forward it on to the Assistant team for their feedback.

[randrescastaneda]

That's wonderful @seeM, Thank you so much. I'm really looking forward to hearing from the team about this feature.