chore(tools): remove Claude Code-specific sections from skill cheatsheet
Keep cheatsheet focused on the Agent Skills open standard only. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
parent
07f1a44c5c
commit
6dc9ce0090
|
|
@ -325,7 +325,7 @@ If zero findings: report "All {N} rules passed. No findings." and list all passe
|
|||
|
||||
## Skill Spec Cheatsheet
|
||||
|
||||
Quick-reference for the Agent Skills open standard and Claude Code extensions.
|
||||
Quick-reference for the Agent Skills open standard.
|
||||
For the full standard, see: [Agent Skills specification](https://agentskills.io/specification)
|
||||
|
||||
### Structure
|
||||
|
|
@ -351,48 +351,6 @@ For the full standard, see: [Agent Skills specification](https://agentskills.io/
|
|||
- Skills are filesystem-based directories, not API payloads — Claude reads them via bash/file tools
|
||||
- Keep SKILL.md focused; offload detailed reference to separate files
|
||||
|
||||
### Cross-platform portability
|
||||
- Same SKILL.md format works on Claude Code, Claude API, Claude.ai, Agent SDK
|
||||
- Runtime differs per surface: Code has full network access, API has none, claude.ai varies
|
||||
- Skills don't sync across surfaces — deploy separately to each
|
||||
|
||||
### Extra frontmatter fields (Claude Code only)
|
||||
- `disable-model-invocation`: `true` = only user can invoke via `/name`
|
||||
- `user-invocable`: `false` = hidden from `/` menu, only Claude auto-loads it
|
||||
- `allowed-tools`: restrict tools available during skill execution
|
||||
- `model`: override model for this skill
|
||||
- `context`: `fork` runs skill in isolated subagent
|
||||
- `agent`: which subagent type for `context: fork` — `Explore`, `Plan`, `general-purpose`, or custom
|
||||
- `hooks`: skill-scoped lifecycle hooks
|
||||
- `argument-hint`: autocomplete hint like `[issue-number]`
|
||||
|
||||
### String substitutions
|
||||
- `$ARGUMENTS` / `$ARGUMENTS[N]` / `$N` — argument placeholders
|
||||
- `${CLAUDE_SESSION_ID}` — current session ID
|
||||
- If `$ARGUMENTS` absent from content, args appended as `ARGUMENTS: <value>`
|
||||
|
||||
### Dynamic context injection
|
||||
- `` !`command` `` syntax runs shell commands at load time, output replaces placeholder
|
||||
- Runs before Claude sees content — pure preprocessing
|
||||
|
||||
### Subagent execution
|
||||
- `context: fork` = skill runs in isolated context without conversation history
|
||||
- Only useful for skills with explicit task instructions, not pure guidelines
|
||||
- `agent` field picks execution environment; defaults to `general-purpose`
|
||||
|
||||
### Invocation control matrix
|
||||
| Frontmatter | User invokes | Claude invokes | Description in context |
|
||||
|---|---|---|---|
|
||||
| (default) | yes | yes | yes |
|
||||
| `disable-model-invocation: true` | yes | no | no |
|
||||
| `user-invocable: false` | no | yes | yes |
|
||||
|
||||
### Context budget
|
||||
- Skill descriptions budget = 2% of context window, fallback 16k chars
|
||||
- Override with `SLASH_COMMAND_TOOL_CHAR_BUDGET` env var
|
||||
|
||||
### Practical tips
|
||||
- Keep SKILL.md under 500 lines
|
||||
- `description` drives auto-discovery — use keywords users would naturally say
|
||||
- Include "ultrathink" in skill content to enable extended thinking
|
||||
- `.claude/commands/` files still work but skills take precedence on name collision
|
||||
|
|
|
|||
Loading…
Reference in New Issue