Documentation
¶
Overview ¶
Package skills exposes three LLM-callable tools that surface Agent Skills to a chat model through progressive disclosure. It is a thin adapter over the skills module's github.com/Tangerg/scope/skills.Source capability: the base module parses, validates, and serves skill content; this package maps that onto the shared `tool.Tool` contract.
Each disclosure step has one tool and one input shape:
- list_skills — every skill's name + description
- load_skill — one skill's full instruction body, by name
- read_skill_resource — one bundled file under a skill
Scripts bundled with a skill are NOT executed here — the model runs them with its own shell/file tools after reading the instructions.
Index ¶
Constants ¶
const (
DefaultMaxOutputBytes = int64(256 * 1024)
)
Exported defaults keep constructor behavior visible and overridable.
Variables ¶
var ErrNilSource = errors.New("skills: source must not be nil")
Functions ¶
func NewTools ¶
func NewTools(source skillsrc.ResourceSource, config Config) ([]toolcontract.Tool, error)
NewTools returns the listing, loading, and resource-reading tools together because they are only useful as a set: a model that can load a skill but not read the resources its instructions reference will follow a broken pointer.
Types ¶
type Config ¶
type Config struct {
MaxOutputBytes int64
}
Config bounds how much skill content may enter a model's context. The limit lives here rather than in the repository because the same skill set is reasonable or oversized depending on the model it is fed to.
type LoadSkillRequest ¶
type LoadSkillRequest struct {
Name string `` /* 136-byte string literal not displayed */
}
LoadSkillRequest names a skill exactly. Loading is by name rather than by a query so a model cannot broaden what it loads, and so the same prompt resolves to the same skill on every run.
type ReadSkillResourceRequest ¶
type ReadSkillResourceRequest struct {
Name string `` /* 129-byte string literal not displayed */
Path string `` /* 159-byte string literal not displayed */
}
ReadSkillResourceRequest carries the skill name alongside the resource path, so a read is confined to the skill that referenced it. A path alone would let one skill read another's files.
Source Files