feat(skills): add tsh-creating-router-skills - #71
Conversation
IgorWnek
changed the title
|
|
||
| - **Conciseness** — The context window is a shared resource. Only add context the LLM doesn't already have. | ||
| - **Progressive disclosure** — Discovery tier (~100 tokens): name + description. Activation tier (<5000 tokens): full SKILL.md body. Resource tier (on demand): templates, examples. | ||
| - **Separation of concerns** — A skill defines HOW to perform a task. It must NOT define WHO the agent is (agent territory) or WHAT triggers the workflow (prompt territory). |
There was a problem hiding this comment.
Do you mean only the prompt-level workflow trigger here? I’m asking because the current wording can be read as “a skill should not say when to use it”, while the canonical guidance in line 164 of .github/skills/tsh-creating-skills/SKILL.md file explicitly says the description should cover both what the skill does and when to use it, and in line 299 also mentions trigger keywords.
| @@ -0,0 +1,50 @@ | |||
| --- | |||
| name: <stack-leaf-skill-name> | |||
There was a problem hiding this comment.
what do you think about encoding the tsh- prefix directly in the placeholder here, so the template reinforces the canonical naming pattern instead of a generic one?
| @@ -0,0 +1,92 @@ | |||
| --- | |||
| name: <skill-name> | |||
There was a problem hiding this comment.
what do you think about encoding the tsh- prefix directly in the placeholder here, so the template reinforces the canonical naming pattern instead of a generic one?
| @@ -0,0 +1,50 @@ | |||
| --- | |||
| name: <pattern-leaf-skill-name> | |||
There was a problem hiding this comment.
what do you think about encoding the tsh- prefix directly in the placeholder here, so the template reinforces the canonical naming pattern instead of a generic one?
| - Directory name uses gerund form with `tsh-` prefix | ||
| - `name` field in frontmatter matches directory name | ||
| - Description is concise (1–2 sentences) | ||
| - Body stays under 5000 tokens |
There was a problem hiding this comment.
in .github/skills/tsh-creating-skills/SKILL.md in lines 55, 220 and 300 we used "lines" but here we use "tokens" - what do you think about making it more consistent?
| ## Tool Usage Guidelines | ||
|
|
||
| - **`read` / `search`** — Use before creating to check existing patterns. When creating a new agent, read 1–2 existing agents in `.github/agents/` to match conventions. When creating a new skill, read existing skills in `.github/skills/`. Use `search` to find references to the artifact being modified and check for consistency impacts. | ||
| - **`edit`** — Use to create new files or modify existing ones. For new files, use the `create_file` capability. For modifications, read the file first, then apply targeted edits. |
There was a problem hiding this comment.
could we align this wording with the declared toolset? Mentioning create_file here introduces a capability name that is not present in the agent’s tool list, which may make the execution path slightly less explicit
2 participants
Adds the
tsh-creating-router-skillsskill for building router-style skills in one domain.Includes reusable templates, a worked example, docs updates, and changelog entries.