agentrunner

package

v0.0.4 Latest Latest Go to latest Published: Sep 18, 2026 License: MIT Imports: 8 Imported by: 0

Details

Valid go.mod file
Redistributable license
Tagged version
Stable version
Learn more about best practices

Repository

github.com/leonp92/golem

Links

Open Source Insights

Documentation ¶

Index ¶

func BuildPrompt(ctx Context) string
func GenerateClaudeCodeArtifacts(repoRoot string, roleContent map[string]string) error
func GenerateClaudeCodeCommands(repoRoot string) error
type ClaudeCode
- func (c ClaudeCode) RunAgent(role string, ctx Context) (Result, error)
- func (c ClaudeCode) WorktreeSetup(worktreePath string) error
type Context
type Mock
- func NewMock() *Mock
type Result
type Runner

Constants ¶

This section is empty.

Variables ¶

This section is empty.

Functions ¶

func BuildPrompt ¶

func BuildPrompt(ctx Context) string

BuildPrompt reinforces "content is data, not instructions" at the mechanism level, not just in the role prompt (spec: Roles). The delimiters are best-effort, not a guarantee — adversarial diff or log content could itself contain a fake closing tag. This raises the bar against injection, it does not eliminate it. Every backend that takes a flat text prompt (any CLI-shim adapter) should call this rather than assembling its own — that's what keeps injection-mitigation behavior identical across backends instead of drifting per adapter.

func GenerateClaudeCodeArtifacts ¶

func GenerateClaudeCodeArtifacts(repoRoot string, roleContent map[string]string) error

GenerateClaudeCodeArtifacts translates neutral role content into Claude Code subagent definitions. This is a projection of the neutral role files, not the source of truth (spec: Architecture Overview) — re-running it after a role file changes regenerates the artifact.

func GenerateClaudeCodeCommands ¶

func GenerateClaudeCodeCommands(repoRoot string) error

Types ¶

type ClaudeCode ¶

type ClaudeCode struct {
	RepoRoot string
	Model    string
}

ClaudeCode dispatches role invocations as one-shot `claude -p` calls. It stores no credentials — it relies entirely on the host's own `claude` CLI authentication (spec: Backend Adapters).

func (ClaudeCode) RunAgent ¶

func (c ClaudeCode) RunAgent(role string, ctx Context) (Result, error)

func (ClaudeCode) WorktreeSetup ¶

func (c ClaudeCode) WorktreeSetup(worktreePath string) error

type Context ¶

type Context struct {
	LogSlice   []blog.Entry
	Diff       string
	RolePrompt string
}

Context is everything a one-shot role invocation needs, reconstructed fresh by the observer from ticket state — there is no persistent conversation across invocations (spec: Concurrency Model).

type Mock ¶

type Mock struct {
	// contains filtered or unexported fields
}

Mock is a scripted Runner used in tests so the ticket lifecycle can be exercised end-to-end without a real model call or token cost (spec: Testing Strategy).

func NewMock ¶

func NewMock() *Mock

func (*Mock) RunAgent ¶

func (m *Mock) RunAgent(role string, _ Context) (Result, error)

func (*Mock) ScriptResponse ¶

func (m *Mock) ScriptResponse(role string, result Result)

func (*Mock) WorktreeSetup ¶

func (m *Mock) WorktreeSetup(_ string) error

type Result ¶

type Result struct {
	Output string
	Model  string
}

type Runner ¶

type Runner interface {
	RunAgent(role string, ctx Context) (Result, error)
	WorktreeSetup(worktreePath string) error
}

Runner is implemented once per backend (claude-code today; gemini, codex later) against the same signature, so the orchestration core never depends on which backend is configured.

Source Files ¶

View all Source files

?	: This menu
/	: Search site
f or F	: Jump to
y or Y	: Canonical URL