NewSkill builds a "skill" tool over discovered skills.Skill values, returning nil when there are none so callers register nothing and the prompt stays byte-identical without the feature. The tool description lists available skills; the input schema's name property gets an enum of the discovered slugs patched on via withEnum/patchedSchema, since invopop/jsonschema can't reflect a list only known at startup. Repeated invocations return the same body with no deduplication. skill is added to approval's auto-approved read-only set: it only reads a file from a directory found at startup, no more reach than read_file already has. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Showing 4 changed files with 202 additions and 1 deletion.
| @@ -40,6 +40,9 @@ var readOnly = map[string]bool{ | ||
| 40 | 40 | "read_file": true, |
| 41 | 41 | "glob": true, |
| 42 | 42 | "grep": true, |
| 43 | + // skill reads one file from a directory found at startup, and cannot | |
| 44 | + // reach anything read_file could not already reach. | |
| 45 | + "skill": true, | |
| 43 | 46 | } |
| 44 | 47 | |
| 45 | 48 | // Policy decides whether a tool call proceeds, remembering what it is told. |
| @@ -15,7 +15,7 @@ func approve(t *testing.T, p *Policy, tool, input string) error { | ||
| 15 | 15 | func TestReadOnlyToolsNeverPrompt(t *testing.T) { |
| 16 | 16 | var asked int |
| 17 | 17 | p := New(func(context.Context, Request) (Decision, error) { asked++; return Deny, nil }) |
| 18 | - for _, tool := range []string{"read_file", "glob", "grep"} { | |
| 18 | + for _, tool := range []string{"read_file", "glob", "grep", "skill"} { | |
| 19 | 19 | if err := approve(t, p, tool, `{"path":"x"}`); err != nil { |
| 20 | 20 | t.Errorf("%s was denied: %v", tool, err) |
| 21 | 21 | } |
| @@ -0,0 +1,95 @@ | ||
| 1 | + package tools | |
| 2 | + | |
| 3 | + import ( | |
| 4 | + "context" | |
| 5 | + "encoding/json" | |
| 6 | + "fmt" | |
| 7 | + "sort" | |
| 8 | + "strings" | |
| 9 | + | |
| 10 | + "harness/internal/skills" | |
| 11 | + ) | |
| 12 | + | |
| 13 | + // SkillArgs is the skill input. The name is constrained to the discovered | |
| 14 | + // slugs by an enum patched into the schema, so a hallucinated name is a schema | |
| 15 | + // violation the model sees at once rather than a lookup failure it has to | |
| 16 | + // reason about. | |
| 17 | + type SkillArgs struct { | |
| 18 | + Name string `json:"name" jsonschema:"description=Which skill to load"` | |
| 19 | + Args string `json:"args,omitempty" jsonschema:"description=Arguments for the skill; whatever followed the skill name"` | |
| 20 | + } | |
| 21 | + | |
| 22 | + // NewSkill builds the skill tool over the skills discovered at startup. It | |
| 23 | + // returns nil when there are none, so the caller registers nothing and the | |
| 24 | + // prompt is byte-identical to a harness without the feature. | |
| 25 | + func NewSkill(found []skills.Skill) Tool { | |
| 26 | + if len(found) == 0 { | |
| 27 | + return nil | |
| 28 | + } | |
| 29 | + byName := make(map[string]skills.Skill, len(found)) | |
| 30 | + names := make([]string, 0, len(found)) | |
| 31 | + var listing strings.Builder | |
| 32 | + for _, s := range found { | |
| 33 | + byName[s.Name] = s | |
| 34 | + names = append(names, s.Name) | |
| 35 | + fmt.Fprintf(&listing, "\n %s: %s", s.Name, s.Description) | |
| 36 | + } | |
| 37 | + sort.Strings(names) | |
| 38 | + | |
| 39 | + desc := strings.TrimSpace(` | |
| 40 | + Load a skill: a set of instructions for one kind of work. Invoke a skill when | |
| 41 | + its description matches what you are about to do — before starting, not after. | |
| 42 | + The instructions come back as the tool result; follow them. | |
| 43 | + | |
| 44 | + Available skills:`) + listing.String() | |
| 45 | + | |
| 46 | + t := New("skill", desc, func(_ context.Context, in SkillArgs) (string, error) { | |
| 47 | + s, ok := byName[in.Name] | |
| 48 | + if !ok { | |
| 49 | + return "", fmt.Errorf("no skill named %q; available: %s", in.Name, strings.Join(names, ", ")) | |
| 50 | + } | |
| 51 | + var b strings.Builder | |
| 52 | + fmt.Fprintf(&b, "Skill: %s\nDirectory: %s\n", s.Name, s.Dir) | |
| 53 | + b.WriteString("Relative paths below resolve against that directory.\n\n") | |
| 54 | + b.WriteString(s.Body) | |
| 55 | + if a := strings.TrimSpace(in.Args); a != "" { | |
| 56 | + b.WriteString("\n\nARGUMENTS: " + a) | |
| 57 | + } | |
| 58 | + return b.String(), nil | |
| 59 | + }) | |
| 60 | + return withEnum(t, "name", names) | |
| 61 | + } | |
| 62 | + | |
| 63 | + // withEnum patches an enum onto one string property of a reflected schema. | |
| 64 | + // invopop/jsonschema derives the schema from struct tags, which cannot carry a | |
| 65 | + // list only known at startup. | |
| 66 | + func withEnum(t Tool, property string, values []string) Tool { | |
| 67 | + var m map[string]any | |
| 68 | + if err := json.Unmarshal(t.InputSchema(), &m); err != nil { | |
| 69 | + return t | |
| 70 | + } | |
| 71 | + props, ok := m["properties"].(map[string]any) | |
| 72 | + if !ok { | |
| 73 | + return t | |
| 74 | + } | |
| 75 | + p, ok := props[property].(map[string]any) | |
| 76 | + if !ok { | |
| 77 | + return t | |
| 78 | + } | |
| 79 | + p["enum"] = values | |
| 80 | + patched, err := json.Marshal(m) | |
| 81 | + if err != nil { | |
| 82 | + return t | |
| 83 | + } | |
| 84 | + return patchedSchema{Tool: t, schema: patched} | |
| 85 | + } | |
| 86 | + | |
| 87 | + // patchedSchema wraps a Tool to override only its input schema. | |
| 88 | + type patchedSchema struct { | |
| 89 | + Tool | |
| 90 | + schema json.RawMessage | |
| 91 | + } | |
| 92 | + | |
| 93 | + // InputSchema returns the schema with the patched enum, overriding the | |
| 94 | + // embedded Tool's reflected one. | |
| 95 | + func (p patchedSchema) InputSchema() json.RawMessage { return p.schema } | |
| @@ -0,0 +1,103 @@ | ||
| 1 | + package tools | |
| 2 | + | |
| 3 | + import ( | |
| 4 | + "encoding/json" | |
| 5 | + "strings" | |
| 6 | + "testing" | |
| 7 | + | |
| 8 | + "harness/internal/skills" | |
| 9 | + ) | |
| 10 | + | |
| 11 | + func testSkills() []skills.Skill { | |
| 12 | + return []skills.Skill{ | |
| 13 | + {Name: "alpha", Description: "the first one", Body: "do the alpha thing", Dir: "/skills/alpha", Source: "personal"}, | |
| 14 | + {Name: "beta", Description: "the second one", Body: "do the beta thing", Dir: "/skills/beta", Source: "project"}, | |
| 15 | + } | |
| 16 | + } | |
| 17 | + | |
| 18 | + func TestSkillReturnsBodyWithDirectory(t *testing.T) { | |
| 19 | + tool := NewSkill(testSkills()) | |
| 20 | + out, err := run(t, tool, map[string]any{"name": "alpha"}) | |
| 21 | + if err != nil { | |
| 22 | + t.Fatal(err) | |
| 23 | + } | |
| 24 | + for _, want := range []string{"Skill: alpha", "Directory: /skills/alpha", "do the alpha thing"} { | |
| 25 | + if !strings.Contains(out, want) { | |
| 26 | + t.Errorf("output missing %q:\n%s", want, out) | |
| 27 | + } | |
| 28 | + } | |
| 29 | + } | |
| 30 | + | |
| 31 | + func TestSkillAppendsArguments(t *testing.T) { | |
| 32 | + tool := NewSkill(testSkills()) | |
| 33 | + out, err := run(t, tool, map[string]any{ | |
| 34 | + "name": "beta", "args": "fix the flaky test", | |
| 35 | + }) | |
| 36 | + if err != nil { | |
| 37 | + t.Fatal(err) | |
| 38 | + } | |
| 39 | + if !strings.Contains(out, "ARGUMENTS: fix the flaky test") { | |
| 40 | + t.Errorf("arguments missing:\n%s", out) | |
| 41 | + } | |
| 42 | + } | |
| 43 | + | |
| 44 | + func TestSkillOmitsArgumentsLineWhenEmpty(t *testing.T) { | |
| 45 | + tool := NewSkill(testSkills()) | |
| 46 | + out, _ := run(t, tool, map[string]any{"name": "alpha"}) | |
| 47 | + if strings.Contains(out, "ARGUMENTS:") { | |
| 48 | + t.Errorf("empty args produced an ARGUMENTS line:\n%s", out) | |
| 49 | + } | |
| 50 | + } | |
| 51 | + | |
| 52 | + func TestSkillUnknownNameListsTheRealOnes(t *testing.T) { | |
| 53 | + tool := NewSkill(testSkills()) | |
| 54 | + _, err := run(t, tool, map[string]any{"name": "nope"}) | |
| 55 | + if err == nil { | |
| 56 | + t.Fatal("unknown skill was accepted") | |
| 57 | + } | |
| 58 | + if !strings.Contains(err.Error(), "alpha") || !strings.Contains(err.Error(), "beta") { | |
| 59 | + t.Errorf("error does not list the available skills: %v", err) | |
| 60 | + } | |
| 61 | + } | |
| 62 | + | |
| 63 | + func TestSkillSchemaEnumeratesSlugs(t *testing.T) { | |
| 64 | + tool := NewSkill(testSkills()) | |
| 65 | + var schema struct { | |
| 66 | + Properties struct { | |
| 67 | + Name struct { | |
| 68 | + Enum []string `json:"enum"` | |
| 69 | + } `json:"name"` | |
| 70 | + } `json:"properties"` | |
| 71 | + } | |
| 72 | + if err := json.Unmarshal(tool.InputSchema(), &schema); err != nil { | |
| 73 | + t.Fatal(err) | |
| 74 | + } | |
| 75 | + if strings.Join(schema.Properties.Name.Enum, ",") != "alpha,beta" { | |
| 76 | + t.Errorf("enum = %v, want [alpha beta]", schema.Properties.Name.Enum) | |
| 77 | + } | |
| 78 | + } | |
| 79 | + | |
| 80 | + func TestSkillDescriptionListsSkills(t *testing.T) { | |
| 81 | + tool := NewSkill(testSkills()) | |
| 82 | + d := tool.Description() | |
| 83 | + if !strings.Contains(d, "alpha: the first one") || !strings.Contains(d, "beta: the second one") { | |
| 84 | + t.Errorf("tool description does not list the skills:\n%s", d) | |
| 85 | + } | |
| 86 | + } | |
| 87 | + | |
| 88 | + func TestNewSkillWithNoneIsNil(t *testing.T) { | |
| 89 | + if NewSkill(nil) != nil { | |
| 90 | + t.Error("NewSkill(nil) must return nil so the caller registers nothing") | |
| 91 | + } | |
| 92 | + } | |
| 93 | + | |
| 94 | + func TestSkillRepeatInvocationReturnsBodyAgain(t *testing.T) { | |
| 95 | + // Deduping would save tokens and cost the model the ability to re-read | |
| 96 | + // instructions that compaction dropped. | |
| 97 | + tool := NewSkill(testSkills()) | |
| 98 | + first, _ := run(t, tool, map[string]any{"name": "alpha"}) | |
| 99 | + second, _ := run(t, tool, map[string]any{"name": "alpha"}) | |
| 100 | + if first != second || first == "" { | |
| 101 | + t.Error("the second invocation did not return the same body") | |
| 102 | + } | |
| 103 | + } | |