[FEATURE]: Fine Control of Skill Discovery / HST

Feature hasn't been suggested before.

I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

What: User can control levels of subdirectory or simply disable subdirectory lookup for skills.md files.
Opencode can be set to not look for external skills, but no switch available if I only want it look inside a specific folder level and NOT any subdirectory of it (sub-skills).

Why: Like so many, 'browser-harness' skill has 50+ domain-skills inside it, with main SKILL.md serving as a router / index (in short). This is done intentionally to prevent context blowup, while still leaving options/tools to agent.
When discovery find subdir skill.md files and inject into context, breaks scheme and crush tokens. Opencode afaik only harness to do this.

Solution: A little finer control of how/which skills are loaded would be sufficient.

Best version: While I understand not everyone adopts hierarchical skill trees (HST) as a standard, I advocate in favour of it fiercely. I dedicated many hours building them for when thousands of skills are needed, and it works beautifully. It's also simple, everyone can understand and build on top. (yes, it's somewhat Karpathy wiki related).

Suggestion:

Index and inject only main skills front into context
Instruct model about subskills existing in simplest terms
Optional: add hidden subagent for skills w/ small model, maximum skill context budget (compressed), subagent passes compressed index to fit budget, response stored in cache, subagent only called again if index changed
Optional: add field 'subskills' as part of main skill context, list names only (best with 3)

This issue might be a duplicate of existing issues. Please check:

[FEATURE]: Add config option to restrict skill discovery directories #23035: Add config option to restrict skill discovery directories — proposes a skills.directories config to control which directories are scanned, overlapping with the desire for finer skill discovery control
[FEATURE]: Add skills.exclude config to disable specific skill directories #26424: Add skills.exclude config to disable specific skill directories — proposes a skills.exclude config option to selectively skip directories during discovery
[FEATURE]: Recursive skill discovery + multi-skill selection in TUI #21495: Recursive skill discovery + multi-skill selection in TUI — discusses recursive vs. flat skill scanning depth, which is the inverse of the depth-limiting concern raised here
[FEATURE]: Agent-scoped skill loading — only load skills declared by the active agent #19344: Agent-scoped skill loading — only load skills declared by the active agent — related to limiting which skills are injected into context
[FEATURE]: Lazy-load agent/skill lists instead of inlining into tool descriptions #13188: Lazy-load agent/skill lists instead of inlining into tool descriptions — related to reducing context/token overhead from skill loading

Partially

This issue might be a duplicate of existing issues. Please check:

[FEATURE]: Add config option to restrict skill discovery directories #23035: Add config option to restrict skill discovery directories — proposes a skills.directories config to control which directories are scanned, overlapping with the desire for finer skill discovery control

[FEATURE]: Add skills.exclude config to disable specific skill directories #26424: Add skills.exclude config to disable specific skill directories — proposes a skills.exclude config option to selectively skip directories during discovery

[FEATURE]: Recursive skill discovery + multi-skill selection in TUI #21495: Recursive skill discovery + multi-skill selection in TUI — discusses recursive vs. flat skill scanning depth, which is the inverse of the depth-limiting concern raised here

[FEATURE]: Agent-scoped skill loading — only load skills declared by the active agent #19344: Agent-scoped skill loading — only load skills declared by the active agent — related to limiting which skills are injected into context

[FEATURE]: Lazy-load agent/skill lists instead of inlining into tool descriptions #13188: Lazy-load agent/skill lists instead of inlining into tool descriptions — related to reducing context/token overhead from skill loading

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

Navigation Menu

Repository navigation

[FEATURE]: Fine Control of Skill Discovery / HST #28485

Feature hasn't been suggested before.

Describe the enhancement you want to request

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[FEATURE]: Fine Control of Skill Discovery / HST #28485

Description

Feature hasn't been suggested before.

Describe the enhancement you want to request

Activity

github-actions commented on May 20, 2026

yuichiinumaru commented on May 20, 2026

github-actions commented on Jul 20, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions