Documentation
¶
Index ¶
- Constants
- func DefaultAgentRoles() []string
- func DefaultAllowedRemoteResources() []string
- func EnsureDefaultAllowedRemoteResources(existing []string) []string
- func IsAgentExplicitlyDisabled(agents []AgentEntry, name string) bool
- func IsPerRepoYAML(data []byte) bool
- func PerRepoDefaultRoles() []string
- func ValidAgentNames() []string
- func ValidEffort(level string) bool
- func ValidEffortLevels() []string
- func ValidModelRef(ref string) bool
- func ValidProviders() []string
- func ValidRoles() []string
- func ValidRuntimes() []string
- func ValidateAgentEntries(agents []AgentEntry, allowlist []string) error
- type AgentEntry
- type AgentLister
- type AllowTargets
- type AllowedResourcesReader
- type CommentNotificationConfig
- type ConfigReader
- type ConfigWriter
- type CreateIssuesConfig
- type CreateIssuesReader
- type DispatchConfig
- type InferenceConfig
- type KillSwitchReader
- type LoadOpts
- type OpenAIWIFConfig
- type OrgConfigReader
- type OrgConfigWriter
- type PerRepoConfigReader
- type PerRepoConfigWriter
- func NewEmptyPerRepoOverlay() PerRepoConfigWriter
- func NewPerRepoConfig(roles []string, targetRepo string) PerRepoConfigWriter
- func NewPerRepoConfigFromOrg(orgCfg OrgConfigReader, repoName, targetRepo string) PerRepoConfigWriter
- func ParsePerRepoConfigWriterLayered(overlayData []byte, baseData []byte) (PerRepoConfigWriter, error)
- type PerRepoInferenceConfig
- type ReactionNotificationConfig
- type RepoConfig
- type RepoDefaults
- type StatusNotificationConfig
- type StatusNotificationsReader
Constants ¶
const ( // DefaultUpstreamRepo is the canonical fullsend repository for layered workflow calls. DefaultUpstreamRepo = "fullsend-ai/fullsend" // DefaultUpstreamRef is the default ref for layered upstream workflow calls. DefaultUpstreamRef = "main" // DefaultGHRunner is the default GitHub Actions runner image for scaffold workflows. DefaultGHRunner = "ubuntu-24.04" )
const ( // DefaultPerRepoMintURL is the hosted public mint used when no // mint_url is configured in config.yaml or config.base.yaml. DefaultPerRepoMintURL = "https://mint.fullsend.sh" // DefaultPerRepoInferenceProvider is the default inference backend. DefaultPerRepoInferenceProvider = "vertex" // DefaultPerRepoInferenceRegion is the default GCP region for // inference requests. DefaultPerRepoInferenceRegion = "global" )
Operational code defaults for per-repo configs. These are the terminal values in the overlay → base → code defaults chain (ADR 0069 Decision 2). Callers that need the canonical default value without constructing a full config can reference these constants directly.
const ( OverlayConfigFile = "config.yaml" BaseConfigFile = "config.base.yaml" )
Layer file names of a per-repo config directory (ADR 0069).
Variables ¶
This section is empty.
Functions ¶
func DefaultAgentRoles ¶
func DefaultAgentRoles() []string
DefaultAgentRoles returns the standard set of agent roles installed when no custom roles are specified. The fix stage reuses the coder app (role: coder) so it does not need a separate app or PEM.
func DefaultAllowedRemoteResources ¶ added in v0.24.0
func DefaultAllowedRemoteResources() []string
DefaultAllowedRemoteResources returns the standard allowlist prefixes for base composition and agent registration URLs.
func EnsureDefaultAllowedRemoteResources ¶ added in v0.31.0
EnsureDefaultAllowedRemoteResources returns a new slice with default allowed-remote-resources prefixes merged into the provided list. Nil input (field omitted from YAML) returns defaults alone. An explicit empty slice (allowed_remote_resources: []) is returned unchanged to preserve deny-all semantics. Non-empty input gets any missing defaults appended (preserving the caller's original ordering).
func IsAgentExplicitlyDisabled ¶ added in v0.33.0
func IsAgentExplicitlyDisabled(agents []AgentEntry, name string) bool
IsAgentExplicitlyDisabled returns true if the last config entry matching name has Enabled explicitly set to false. Iterates in reverse to respect last-writer-wins ordering.
func IsPerRepoYAML ¶ added in v0.31.0
IsPerRepoYAML probes raw YAML for structural markers that distinguish perRepoConfig from orgConfig. orgConfig has org-only top-level keys (dispatch, repos, defaults); perRepoConfig never does. The "inference" key is shared: orgConfig uses it for the provider name, perRepoConfig uses it for nested inference backend settings. Org configs always contain dispatch/repos/defaults (non-omitempty), so the remaining markers are sufficient for detection.
func PerRepoDefaultRoles ¶ added in v0.8.0
func PerRepoDefaultRoles() []string
PerRepoDefaultRoles returns agent roles for per-repo installation. The "fullsend" dispatch role is excluded because per-repo mode uses the target repo's shim workflow for dispatch instead of a separate app.
func ValidAgentNames ¶ added in v0.38.0
func ValidAgentNames() []string
ValidAgentNames returns the built-in agents fullsend dispatches by name — the names an agents: entry may tune without a source. These are the names passed to `fullsend run <agent>` and used by workflow stages. They are NOT harness role: values — "code" and "fix" both carry role: coder.
func ValidEffort ¶ added in v0.38.0
ValidEffort reports whether level is an accepted effort value.
func ValidEffortLevels ¶ added in v0.38.0
func ValidEffortLevels() []string
ValidEffortLevels returns the accepted effort values, in documentation order.
func ValidModelRef ¶ added in v0.38.0
ValidModelRef reports whether ref is a well-formed model reference (single segment or provider/id form). Exported for use by harness validation and per-run override validation.
func ValidProviders ¶ added in v0.0.3
func ValidProviders() []string
ValidProviders returns the set of recognized inference providers.
func ValidRoles ¶
func ValidRoles() []string
ValidRoles returns the set of agent roles accepted in `.fullsend` `roles:` / `defaults.roles` config. This is intentionally narrower than mintcore's canonical roles: mint-only dogfood roles (e.g. scribe) can be registered with `fullsend mint add-role` before scaffold/workflow wiring lands, and must not silently pass config validation.
func ValidRuntimes ¶ added in v0.30.0
func ValidRuntimes() []string
ValidRuntimes returns the set of recognized agent runtimes. "pi" (#6464) and "codex" (#6920) are both opt-in per repo, per agent, or as a repos.yaml default; "dummy" and "dummy-playback" are for behaviour test orgs only.
func ValidateAgentEntries ¶ added in v0.25.0
func ValidateAgentEntries(agents []AgentEntry, allowlist []string) error
Types ¶
type AgentEntry ¶
type AgentEntry struct {
Name string `yaml:"name,omitempty"`
Source string `yaml:"source,omitempty"`
Ref string `yaml:"ref,omitempty"`
Enabled *bool `yaml:"enabled,omitempty"`
Runtime string `yaml:"runtime,omitempty"`
Model string `yaml:"model,omitempty"`
Effort string `yaml:"effort,omitempty"`
}
AgentEntry represents a registered agent source in config. It supports both string shorthand (just the source URL/path) and object form (with an explicit name override).
Enabled controls whether the agent participates in the merged agent set. When nil (omitted) the agent defaults to enabled. When explicitly set to false the agent is suppressed — this allows disabling built-in scaffold agents without removing their role. A suppression-only entry (Enabled=false, no Source) is valid.
Ref records the original branch or tag ref that was resolved to a commit SHA when the agent was adopted via `agent add`. When present, `agent update` re-resolves against this ref instead of the repo's default branch, so agents adopted from non-default branches stay on their intended branch across updates. Empty for entries that predate this field or were adopted with an explicit commit SHA.
Runtime, Model and Effort tune how the agent runs (ADR 0091): they override the repo-wide runtime: key and the harness model:/effort: for this agent, beneath the per-run --runtime/--model/--effort flags and FULLSEND_* variables. An enabled entry may carry them without a Source — an override-only entry — when Name is a built-in agent (ValidAgentNames) or matches a sourced entry in a parent layer; the built-in keeps resolving through the agents-repo fallback.
func AgentSettingsFor ¶ added in v0.38.0
func AgentSettingsFor(agents []AgentEntry, name string) (AgentEntry, bool)
AgentSettingsFor returns the entry for name (case-insensitive) from an effective (merged) agent list, and whether one exists. Callers read Runtime/Model/Effort from it.
func UpsertAgentSettings ¶ added in v0.38.0
func UpsertAgentSettings(agents []AgentEntry, name, runtime, model, effort string) []AgentEntry
UpsertAgentSettings sets runtime/model/effort for name on a layer's local agent list: on the entry with that name when present, else as a new override-only entry. An empty value clears that setting. Returns the updated list.
func (AgentEntry) DerivedName ¶ added in v0.24.0
func (a AgentEntry) DerivedName() string
DerivedName returns the explicit Name if set, otherwise derives one from the Source filename (e.g. "triage.yaml" → "triage").
func (AgentEntry) HasSettings ¶ added in v0.38.0
func (a AgentEntry) HasSettings() bool
HasSettings reports whether the entry tunes runtime, model or effort.
func (AgentEntry) IsEnabled ¶ added in v0.33.0
func (a AgentEntry) IsEnabled() bool
IsEnabled returns whether the agent entry is enabled. A nil Enabled pointer (field omitted) defaults to true.
func (AgentEntry) IsOverrideOnly ¶ added in v0.38.0
func (a AgentEntry) IsOverrideOnly() bool
IsOverrideOnly reports whether the entry only tunes an agent defined elsewhere: enabled, no source, at least one setting.
func (*AgentEntry) UnmarshalYAML ¶ added in v0.24.0
func (a *AgentEntry) UnmarshalYAML(value *yaml.Node) error
UnmarshalYAML implements yaml.Unmarshaler so that a plain string is treated as a source-only entry, while a mapping decodes normally. Old-format entries (role/name/slug identity tuples from pre-ADR-0058 config) are detected and rejected with a clear error message.
type AgentLister ¶ added in v0.31.0
type AgentLister interface {
AgentEntries() []AgentEntry
}
AgentLister provides read access to registered agent entries.
type AllowTargets ¶ added in v0.18.0
type AllowTargets struct {
Orgs []string `yaml:"orgs,omitempty"`
Repos []string `yaml:"repos,omitempty"`
}
AllowTargets defines which orgs and repos agents may create issues in.
type AllowedResourcesReader ¶ added in v0.31.0
type AllowedResourcesReader interface {
AllowedResources() []string
}
AllowedResourcesReader provides read access to the allowed remote resources list.
type CommentNotificationConfig ¶ added in v0.15.0
type CommentNotificationConfig struct {
Start string `yaml:"start,omitempty"`
Completion string `yaml:"completion,omitempty"`
}
CommentNotificationConfig controls start/completion comments. Valid start values: "enabled" (default), "disabled". Valid completion values: "enabled" (default), "on_failure", "disabled".
type ConfigReader ¶ added in v0.31.0
type ConfigReader interface {
AgentLister
KillSwitchReader
AllowedResourcesReader
CreateIssuesReader
StatusNotificationsReader
ConfigVersion() string
IsOrgMode() bool
}
ConfigReader is the common read interface for fields shared by both orgConfig and perRepoConfig. Consumer packages should depend on this interface rather than accessing struct fields directly.
func LoadConfig ¶ added in v0.31.0
func LoadConfig(dir string, opts LoadOpts) (ConfigReader, error)
LoadConfig reads config.yaml (and config.base.yaml if present) from dir, returning a ConfigReader. For per-repo configs the parent chain is wired as overlay (config.yaml) → base (config.base.yaml) → code defaults. This is the preferred entry point for consumer packages that only need read access.
type ConfigWriter ¶ added in v0.31.0
type ConfigWriter interface {
ConfigReader
SetKillSwitch(bool)
SetAgents([]AgentEntry)
SetAllowedRemoteResources([]string)
SetStatusNotifications(*StatusNotificationConfig)
Marshal() ([]byte, error)
Validate() error
}
ConfigWriter extends ConfigReader with mutation methods shared by both config modes.
func LoadConfigWriter ¶ added in v0.32.0
func LoadConfigWriter(dir string, opts LoadOpts) (ConfigWriter, error)
LoadConfigWriter reads config.yaml (and config.base.yaml if present) from dir, returning a ConfigWriter. For per-repo configs the parent chain is wired as overlay (config.yaml) → base (config.base.yaml) → code defaults. Only the overlay layer is mutable; base and defaults are read-only via the parent pointer. This is the preferred entry point for consumer packages that need read-write access (e.g. CLI commands that modify and write-back config).
func ParsePerRepoConfigWriter ¶ added in v0.33.0
func ParsePerRepoConfigWriter(data []byte) (ConfigWriter, error)
ParsePerRepoConfigWriter parses YAML bytes into a ConfigWriter for callers that need to modify the config after parsing.
type CreateIssuesConfig ¶ added in v0.18.0
type CreateIssuesConfig struct {
AllowTargets AllowTargets `yaml:"allow_targets"`
}
CreateIssuesConfig controls cross-repo issue creation by agents.
type CreateIssuesReader ¶ added in v0.31.0
type CreateIssuesReader interface {
IssueCreationConfig() *CreateIssuesConfig
}
CreateIssuesReader provides read access to cross-repo issue creation configuration.
type DispatchConfig ¶
type DispatchConfig struct {
Platform string `yaml:"platform"`
Mode string `yaml:"mode,omitempty"` // "oidc-mint"
MintURL string `yaml:"mint_url,omitempty"` // informational, set when mode=oidc-mint
}
DispatchConfig configures how agent work is dispatched.
type InferenceConfig ¶ added in v0.0.3
type InferenceConfig struct {
Provider string `yaml:"provider"`
}
InferenceConfig configures the inference provider used by agents (org-mode config).
type KillSwitchReader ¶ added in v0.31.0
type KillSwitchReader interface {
IsKillSwitchActive() bool
}
KillSwitchReader provides read access to the kill switch state.
type LoadOpts ¶ added in v0.31.0
type LoadOpts struct {
// MissingOK returns a default config when config.yaml is absent.
MissingOK bool
}
LoadOpts controls how LoadConfig handles a missing config.yaml.
type OpenAIWIFConfig ¶ added in v0.38.0
type OpenAIWIFConfig struct {
Audience string `yaml:"audience,omitempty"`
IdentityProviderID string `yaml:"identity_provider_id,omitempty"`
ServiceAccountID string `yaml:"service_account_id,omitempty"`
}
OpenAIWIFConfig identifies the OpenAI Workload Identity provider and service-account mapping a run exchanges its GitHub OIDC token with.
func (OpenAIWIFConfig) IsZero ¶ added in v0.38.0
func (c OpenAIWIFConfig) IsZero() bool
IsZero reports whether no identifier is set.
func (OpenAIWIFConfig) Missing ¶ added in v0.38.0
func (c OpenAIWIFConfig) Missing() []string
Missing lists the identifiers still unset, in a fixed order, so a partially configured block can be reported precisely.
func (OpenAIWIFConfig) Trimmed ¶ added in v0.38.0
func (c OpenAIWIFConfig) Trimmed() OpenAIWIFConfig
Trimmed returns the identifiers with surrounding whitespace removed, so a whitespace-only YAML value counts as unset.
type OrgConfigReader ¶ added in v0.31.0
type OrgConfigReader interface {
ConfigReader
DispatchSettings() DispatchConfig
InferenceSettings() InferenceConfig
OrgRepoDefaults() RepoDefaults
RepoMap() map[string]RepoConfig
EnabledRepos() []string
DisabledRepos() []string
}
OrgConfigReader extends ConfigReader with org-mode-specific fields.
func ParseOrgConfig ¶
func ParseOrgConfig(data []byte) (OrgConfigReader, error)
ParseOrgConfig parses YAML bytes into an OrgConfigReader.
type OrgConfigWriter ¶ added in v0.31.0
type OrgConfigWriter interface {
OrgConfigReader
ConfigWriter
SetDispatch(DispatchConfig)
SetInference(InferenceConfig)
SetDefaultRuntime(string)
SetRepo(name string, rc RepoConfig)
}
OrgConfigWriter extends OrgConfigReader and ConfigWriter with org-specific mutation methods.
func NewOrgConfig ¶
func NewOrgConfig(allRepos, enabledRepos, roles []string, inferenceProvider, org string) OrgConfigWriter
NewOrgConfig creates a new orgConfig with sensible defaults. The returned OrgConfigWriter provides full read-write access.
func ParseOrgConfigWriter ¶ added in v0.33.0
func ParseOrgConfigWriter(data []byte) (OrgConfigWriter, error)
ParseOrgConfigWriter parses YAML bytes into an OrgConfigWriter for callers that need to modify the config after parsing.
type PerRepoConfigReader ¶ added in v0.31.0
type PerRepoConfigReader interface {
ConfigReader
ConfigRoles() []string
ConfigRuntime() string
ConfigForge() string
ConfigTracker() string
ConfigMintURL() string
ConfigInferenceProvider() string
ConfigInferenceProject() string
ConfigInferenceRegion() string
ConfigInferenceWIFProvider() string
ConfigInferenceOpenAI() OpenAIWIFConfig
}
PerRepoConfigReader extends ConfigReader with per-repo-specific fields. Methods are prefixed with "Config" to avoid conflicts with the struct field names (Roles and Runtime).
func ParsePerRepoConfig ¶ added in v0.8.0
func ParsePerRepoConfig(data []byte) (PerRepoConfigReader, error)
ParsePerRepoConfig parses YAML bytes into a PerRepoConfigReader.
type PerRepoConfigWriter ¶ added in v0.33.0
type PerRepoConfigWriter interface {
PerRepoConfigReader
ConfigWriter
SetRoles([]string)
SetRuntime(string)
SetMintURL(string)
SetInferenceProvider(string)
SetInferenceProject(string)
SetInferenceRegion(string)
SetInferenceWIFProvider(string)
SetInferenceOpenAI(OpenAIWIFConfig)
}
PerRepoConfigWriter extends PerRepoConfigReader and ConfigWriter with per-repo-specific mutation methods.
func NewEmptyPerRepoOverlay ¶ added in v0.36.0
func NewEmptyPerRepoOverlay() PerRepoConfigWriter
NewEmptyPerRepoOverlay creates an empty per-repo config suitable for use as a stub overlay when a preset base layer is provided. No roles, agents, or version are populated; the overlay inherits everything from the base layer via the parent fallback chain.
func NewPerRepoConfig ¶ added in v0.8.0
func NewPerRepoConfig(roles []string, targetRepo string) PerRepoConfigWriter
NewPerRepoConfig creates a new perRepoConfig with the given roles. The returned ConfigWriter provides read-write access to shared config fields; use PerRepoConfigReader type assertion for per-repo-specific methods.
func NewPerRepoConfigFromOrg ¶ added in v0.34.0
func NewPerRepoConfigFromOrg(orgCfg OrgConfigReader, repoName, targetRepo string) PerRepoConfigWriter
NewPerRepoConfigFromOrg creates a per-repo config by mapping portable fields from an org config. Per-repo role overrides (repos.<name>.roles) take precedence over defaults.roles. Non-portable fields (max_implementation_retries, auto_merge) are not carried over — callers should warn separately.
func ParsePerRepoConfigWriterLayered ¶ added in v0.38.0
func ParsePerRepoConfigWriterLayered(overlayData []byte, baseData []byte) (PerRepoConfigWriter, error)
ParsePerRepoConfigWriterLayered parses overlay and optional base YAML bytes into a PerRepoConfigWriter with the parent chain overlay → base → code defaults. When baseData is nil the result is equivalent to ParsePerRepoConfigWriter (parent = code defaults only). This is the preferred entry point when both layers are available as raw bytes (e.g. fetched from a forge API) rather than on the local filesystem (where LoadConfigWriter should be used instead).
type PerRepoInferenceConfig ¶ added in v0.36.0
type PerRepoInferenceConfig struct {
Provider string `yaml:"provider,omitempty"`
Project string `yaml:"project,omitempty"`
Region string `yaml:"region,omitempty"`
WIFProvider string `yaml:"wif_provider,omitempty"`
// OpenAI holds the OpenAI Workload Identity Federation identifiers
// (ADR 0092). They are not secrets: the mapping they name only
// trusts a GitHub OIDC token whose claims match, and the run prints
// them. The FULLSEND_OPENAI_* runner variables, when set, take
// precedence over this block.
OpenAI *OpenAIWIFConfig `yaml:"openai,omitempty"`
}
PerRepoInferenceConfig groups inference backend settings for per-repo configs. The Provider field identifies the inference backend (currently only "vertex"); Project, Region, and WIFProvider hold provider-specific connection details.
type ReactionNotificationConfig ¶ added in v0.38.0
type ReactionNotificationConfig struct {
Start string `yaml:"start,omitempty"`
Completion string `yaml:"completion,omitempty"`
}
ReactionNotificationConfig controls start/completion emoji reactions, an alternative to comments that doesn't generate a GitHub notification. Unlike comments, both fields default to "disabled" — reactions are an opt-in addition rather than a default-on behavior. Valid start values: "enabled", "disabled" (default). Valid completion values: "enabled", "on_failure", "disabled" (default).
type RepoConfig ¶
RepoConfig holds per-repo configuration. StatusNotifications is intentionally absent here — notification style is an org-wide UX decision (consistent appearance across all repos), unlike roles and auto_merge which are operationally per-repo.
type RepoDefaults ¶
type RepoDefaults struct {
Roles []string `yaml:"roles"`
Runtime string `yaml:"runtime,omitempty"`
MaxImplementationRetries int `yaml:"max_implementation_retries"`
AutoMerge bool `yaml:"auto_merge"`
StatusNotifications *StatusNotificationConfig `yaml:"status_notifications,omitempty"`
}
RepoDefaults holds default settings applied to all repos.
type StatusNotificationConfig ¶ added in v0.15.0
type StatusNotificationConfig struct {
Comment CommentNotificationConfig `yaml:"comment,omitempty"`
Reaction ReactionNotificationConfig `yaml:"reaction,omitempty"`
}
StatusNotificationConfig controls status comments and reactions posted on issues/PRs when agents start and complete.
type StatusNotificationsReader ¶ added in v0.36.0
type StatusNotificationsReader interface {
StatusNotifications() *StatusNotificationConfig
}
StatusNotificationsReader provides read access to status notification configuration (comment start & completion).