Documentation
¶
Index ¶
- Variables
- func DefaultIgnorePaths() []string
- func DefaultPhaseBudgets() map[string]float64
- func ResolveBudgetCaps(maxCostUSD *float64, maxDurationSeconds *int) (float64, int, error)
- type AutoDepthThreshold
- type BudgetConfig
- type CommentConfig
- type DepthProfile
- type ModelConfig
- type ReviewConfig
- type ScoringConfig
Constants ¶
This section is empty.
Variables ¶
var AutoDepthThresholds = []AutoDepthThreshold{
{Lines: 100, Depth: "quick"},
{Lines: 500, Depth: "standard"},
}
AutoDepthThresholds ports config.py AUTO_DEPTH_THRESHOLDS: <100 -> quick, 100-500 -> standard, >500 -> deep. Kept as an ordered slice (Python relied on insertion order of the {100:..,500:..} dict) so consumers can range in order.
var DepthProfiles = map[string]DepthProfile{
"quick": {MaxDimensions: 3, ModelTier: "budget"},
"standard": {MaxDimensions: 6, ModelTier: "standard"},
"deep": {MaxDimensions: 12, ModelTier: "premium"},
}
DepthProfiles ports config.py DEPTH_PROFILES. These are static (no env), so a package var is fine.
Functions ¶
func DefaultIgnorePaths ¶
func DefaultIgnorePaths() []string
DefaultIgnorePaths is the config.py glob ignore list. Returned fresh per call.
func DefaultPhaseBudgets ¶
DefaultPhaseBudgets is the per-phase USD allocation (config.py). Returned as a fresh map per call so callers cannot mutate a shared instance.
func ResolveBudgetCaps ¶
ResolveBudgetCaps ports app.py _resolve_budget_caps. An explicit argument always wins, and the defaults are 2.0 / 3600 (real reviews measure 60-70 minutes — the historical 300s default killed every fresh install mid-pipeline). pr-af also read PR_AF_MAX_COST_USD and PR_AF_MAX_DURATION_SECONDS between the two; codeaf always passes the run's own caps (internal/praf's reviewInput), so nothing reads them.
Types ¶
type AutoDepthThreshold ¶
AutoDepthThreshold maps a lines-changed ceiling to a depth. Ordered ascending; the first threshold whose Lines value the change count is under selects the depth. A change over the last threshold (>500) falls through to "deep".
type BudgetConfig ¶
type BudgetConfig struct {
MaxCostUSD float64 `json:"max_cost_usd"`
MaxDurationSeconds int `json:"max_duration_seconds"`
PhaseBudgets map[string]float64 `json:"phase_budgets"`
MaxConcurrentReviewers int `json:"max_concurrent_reviewers"`
MaxReferenceFollowsPerReviewer int `json:"max_reference_follows_per_reviewer"`
MaxChildSpawnsPerReviewer int `json:"max_child_spawns_per_reviewer"`
MaxCrossRefDeepDives int `json:"max_cross_ref_deep_dives"`
MaxCoverageIterations int `json:"max_coverage_iterations"`
MaxReviewDepth int `json:"max_review_depth"`
// EvidencePackReviewers pre-reads each dimension's target files and injects
// them so reviewers reason over a primed pack. Default ON: env
// PR_AF_EVIDENCE_PACK not in {"0","false","no"}.
EvidencePackReviewers bool `json:"evidence_pack_reviewers"`
}
BudgetConfig ports config.py BudgetConfig — global and per-phase budget caps. Numeric defaults are the §D "Config numeric tables" verbatim. Note max_duration_seconds is always overridden per call to the resolved value (3600 by default) via ReviewConfig.FromInput.
func DefaultBudgetConfig ¶
func DefaultBudgetConfig() BudgetConfig
DefaultBudgetConfig builds a BudgetConfig with the config.py defaults, reading PR_AF_EVIDENCE_PACK at call time.
type CommentConfig ¶
type CommentConfig struct {
MinSeverity string `json:"min_severity"` // Minimum severity to include
MaxComments int `json:"max_comments"` // Cap inline comments
IncludeSuggestions bool `json:"include_suggestions"`
IncludeDimensionAttribution bool `json:"include_dimension_attribution"`
IncludeConfidence bool `json:"include_confidence"`
SuggestionMode string `json:"suggestion_mode"` // comment | code
// PolishEnabled runs a parallel .ai() pass that rewrites each comment body
// right before posting; on any per-call failure the original body is kept.
PolishEnabled bool `json:"polish_enabled"`
// MergeGateEnabled runs a parallel .ai() pass classifying each finding as
// blocking vs advisory. Default ON; failures default to advisory.
MergeGateEnabled bool `json:"merge_gate_enabled"`
// PostWorthinessGate: an experienced-reviewer precision pass. DEFAULT OFF;
// enabled by PR_AF_POSTWORTHINESS_GATE in {"1","true","yes"}.
PostWorthinessGate bool `json:"post_worthiness_gate"`
SeverityEmojis map[string]string `json:"severity_emojis"`
}
CommentConfig ports config.py CommentConfig — comment formatting and posting preferences (§D numeric tables). PostWorthinessGate is env-driven (default OFF); every other default is a plain constant.
func DefaultCommentConfig ¶
func DefaultCommentConfig() CommentConfig
DefaultCommentConfig builds the config.py comment defaults, reading PR_AF_POSTWORTHINESS_GATE at call time.
type DepthProfile ¶
type DepthProfile struct {
MaxDimensions int `json:"max_dimensions"`
ModelTier string `json:"model_tier"` // budget | standard | premium
}
DepthProfile ports config.py DepthProfile — a pre-built review-depth profile.
type ModelConfig ¶
type ModelConfig struct {
IntakeGate string `json:"intake_gate"` // .ai() fast classification
IntakeFallback string `json:"intake_fallback"` // .harness() when not confident
AnatomySemantic string `json:"anatomy_semantic"` // Narrative understanding
Planner string `json:"planner"` // THE critical agent
Reviewer string `json:"reviewer"` // Deep code analysis
CrossRef string `json:"cross_ref"` // Interaction detection
Adversary string `json:"adversary"` // Challenge quality
CoverageGate string `json:"coverage_gate"` // Simple completeness check
DedupGate string `json:"dedup_gate"` // Near-duplicate detection
}
ModelConfig ports config.py ModelConfig — model routing per agent. Budget models for gates/classification; premium models for planning/reviewing/ challenging (plan quality = review quality, so the planner gets premium).
func DefaultModelConfig ¶
func DefaultModelConfig() ModelConfig
DefaultModelConfig builds the config.py model-routing defaults.
type ReviewConfig ¶
type ReviewConfig struct {
Budget BudgetConfig `json:"budget"`
Model ModelConfig `json:"models"`
Scoring ScoringConfig `json:"scoring"`
Comments CommentConfig `json:"comments"`
IgnorePaths []string `json:"ignore_paths"`
Hints []string `json:"hints"`
DepthRules []map[string]any `json:"depth_rules"`
}
ReviewConfig ports config.py ReviewConfig — the top-level config combining all sub-configs. The Go `Model` field carries the json tag "models" to match the Python field name.
func DefaultReviewConfig ¶
func DefaultReviewConfig() ReviewConfig
DefaultReviewConfig assembles a ReviewConfig from the sub-config defaults (the Go analog of pydantic's ReviewConfig() with its default_factory fields).
func (ReviewConfig) FromInput ¶
func (ReviewConfig) FromInput(in schemas.ReviewInput) (ReviewConfig, error)
FromInput ports ReviewConfig.from_input — it merges per-call API overrides into the defaults. The value receiver is ignored (matching the Python classmethod semantics: it always starts from a fresh default config).
Budget caps: Python resolves max_cost_usd / max_duration_seconds in review() BEFORE constructing the ReviewInput, so from_input always writes a concrete resolved value over the BudgetConfig defaults (the raw BudgetConfig duration default is dead — the per-call value is 3600 by default). The Go ReviewInput carries these as *float64 / *int, so FromInput runs the same ResolveBudgetCaps cascade (explicit > env > 2.0/3600) to reproduce that effective value exactly.
type ScoringConfig ¶
type ScoringConfig struct {
BaseWeights map[string]float64 `json:"base_weights"`
Multipliers map[string]float64 `json:"multipliers"`
ConfidenceThresholds map[string]float64 `json:"confidence_thresholds"`
}
ScoringConfig ports config.py ScoringConfig — deterministic scoring weights and multipliers (§D numeric tables). LLMs reason about issues; code computes scores, so the same findings always produce the same scores. Maps are returned fresh per call so a caller cannot mutate a shared default.
func DefaultScoringConfig ¶
func DefaultScoringConfig() ScoringConfig
DefaultScoringConfig builds the config.py scoring defaults.
Source Files