[typist] Go type consistency audit: duplicate clusters and the any-to-typed migration

[github-actions[bot]]

Hey team — I just ran a type-consistency sweep over the Go code in pkg/ (839 files, 575 struct definitions, ~1900 any occurrences in 325 non-test files) and there are some really nice wins available. The big-picture story is two-fold: first, the pkg/cli log-analysis pipeline has three slightly-different shapes for "one workflow run plus its analyses" that have to be hand-copied between each other, and the pkg/workflow SafeOutput family has ~25 nearly-identical config structs. Second, there's already a half-finished migration from map[string]any → typed structs (WorkflowData.ParsedTools, WorkflowStep, RuntimesConfig) — finishing it would knock out a large fraction of the remaining any surface.

Nothing here is urgent; this is a "where would refactoring time pay off most" map. Below is the full breakdown with file:line citations, organized by impact.

Full Analysis Report

Duplicated Type Definitions

Summary Statistics

• Total struct types analyzed: 575 (in non-test .go files under pkg/)
• Duplicate clusters reported: 12
• Near-duplicates: 7
• Semantic duplicates: 4
• Deprecated/dead leftovers: 1

Cluster 1: SafeOutput entity-config family — Near duplicates

Type: Near duplicate (~25 structs differ only by docstring + a tiny field set)
Impact: High — largest concentration of duplicated declarations in the repo

Representative locations:

• pkg/workflow/add_labels.go:10 — AddLabelsConfig { BaseSafeOutputConfig; SafeOutputTargetConfig; SafeOutputFilterConfig; Allowed []string; Blocked []string }
• pkg/workflow/remove_labels.go:10 — RemoveLabelsConfig { ...same three embeds; Allowed []string; Blocked []string } (byte-identical apart from the type name)
• pkg/workflow/assign_to_user.go:10 — AssignToUserConfig { ...same three embeds; Allowed; Blocked; UnassignFirst *string }
• pkg/workflow/unassign_from_user.go:10, pkg/workflow/set_issue_type.go:10, pkg/workflow/set_issue_field.go:8, pkg/workflow/hide_comment.go:10, pkg/workflow/link_sub_issue.go:10, pkg/workflow/assign_milestone.go:10, pkg/workflow/mark_pull_request_as_ready_for_review.go:10, pkg/workflow/merge_pull_request.go:8, pkg/workflow/resolve_pr_review_thread.go:12, pkg/workflow/reply_to_pr_review_comment.go:11, pkg/workflow/dispatch_workflow.go:10, pkg/workflow/noop.go:10, pkg/workflow/call_workflow.go:13, pkg/workflow/create_check_run.go:14, pkg/workflow/create_agent_session.go:10, pkg/workflow/autofix_code_scanning_alert.go:10, pkg/workflow/add_reviewer.go:10, pkg/workflow/push_to_pull_request_branch.go:14, pkg/workflow/submit_pr_review.go:15, pkg/workflow/update_release.go:10

Recommendation: Introduce a single shared SafeOutputAllowBlockConfig embeddable (the three existing base embeds + Allowed []string + Blocked []string). Then AddLabelsConfig / RemoveLabelsConfig collapse to type AddLabelsConfig = SafeOutputAllowBlockConfig (or thin embeds). Estimated effort: 4-6 hours. Benefit: ~150 lines of boilerplate gone and the per-tool parse*Config helpers can share more.

Cluster 2: ProcessedRun / RunSummary / DownloadResult — Near duplicates

Type: Near duplicate (same conceptual record, three different shapes)
Impact: High — these are central to the pkg/cli logs pipeline and currently require hand-copying between shapes

Locations:

• pkg/cli/logs_models.go:90 — ProcessedRun { Run, AwContext, TaskDomain, BehaviorFingerprint, AgenticAssessments, AccessAnalysis, FirewallAnalysis, PolicyAnalysis, RedactedDomainsAnalysis, MissingTools, MissingData, Noops, MCPFailures, MCPToolUsage, TokenUsage, GitHubRateLimitUsage, JobDetails } (16 fields)
• pkg/cli/logs_models.go:203 — RunSummary { CLIVersion, RunID, ProcessedAt, ...same 16+ analyses; ArtifactsList }
• pkg/cli/logs_models.go:229 — DownloadResult { Run, Metrics, ...13 of the same analyses; Error, Skipped, Cached, LogsPath }

13 of the analysis fields appear verbatim in all three structs.

Recommendation: Extract type RunAnalyses struct { AwContext *AwContext; TaskDomain *TaskDomainInfo; BehaviorFingerprint *BehaviorFingerprint; AgenticAssessments []AgenticAssessment; AccessAnalysis *DomainAnalysis; FirewallAnalysis *FirewallAnalysis; RedactedDomainsAnalysis *RedactedDomainsAnalysis; MissingTools []MissingToolReport; MissingData []MissingDataReport; Noops []NoopReport; MCPFailures []MCPFailureReport; MCPToolUsage *MCPToolUsageData; TokenUsage *TokenUsageSummary; GitHubRateLimitUsage *GitHubRateLimitUsage; JobDetails []JobInfoWithDuration } and embed it. The three outer types keep only the role-specific fields (CLIVersion/ProcessedAt for RunSummary; Error/Skipped/Cached/LogsPath for DownloadResult). Estimated effort: 2-3 hours. Benefit: the manual field-copying between the three structs disappears.

Cluster 3: MCPServerStats / MCPServerHealthDetail / MCPServerCrossRunHealth — Semantic duplicates

Locations:

• pkg/cli/audit_report.go:197 — MCPServerStats { ServerName, RequestCount, ToolCallCount, TotalInputSize, TotalOutputSize, AvgDuration, ErrorCount }
• pkg/cli/audit_expanded.go:88 — MCPServerHealthDetail { ServerName, RequestCount, ToolCalls, ErrorCount, ErrorRate, ErrorRateStr, AvgLatency, Status }
• pkg/cli/audit_cross_run.go:89 — MCPServerCrossRunHealth { ServerName, RunsConnected, TotalRuns, TotalCalls, TotalErrors, ErrorRate, Unreliable }

Recommendation: Define a small MCPServerCoreStats { ServerName; Requests; ToolCalls; Errors } shared by pkg/cli/audit_* files and embed it in the three role-specific extensions. Impact: medium — clearer code in three audit reports, easier to keep them in sync.

Cluster 4: WorkflowTrialResult / TrialArtifacts — Near duplicates

Locations:

• pkg/cli/trial_types.go:6 — WorkflowTrialResult { WorkflowName; RunID; SafeOutputs map[string]any; AgenticRunInfo map[string]any; AdditionalArtifacts map[string]any; Timestamp time.Time }
• pkg/cli/trial_support.go:19 — TrialArtifacts { SafeOutputs map[string]any; AgenticRunInfo map[string]any; AdditionalArtifacts map[string]any }

Recommendation: type WorkflowTrialResult struct { TrialArtifacts; WorkflowName; RunID; Timestamp }. 30-minute change.

Cluster 5: PRInfo / PullRequest — Semantic duplicates

Locations:

• pkg/cli/pr_command.go:27 — PRInfo { Number, Title, Body, State, HeadSHA, BaseBranch, HeadBranch, SourceRepo, TargetRepo, AuthorLogin }
• pkg/cli/pr_automerge.go:21 — PullRequest { Number, Title, IsDraft, Mergeable, CreatedAt, UpdatedAt }

Both represent a GitHub PR in the same package, both unmarshal gh pr --json output.

Recommendation: Merge into a single PullRequest with the union of fields + omitempty JSON tags so each command requests only what it needs from gh pr --json.

Cluster 6: AccessLogSummary / FirewallLogSummary — Near duplicates

Locations (same file):

• pkg/cli/logs_report_firewall.go:12 — AccessLogSummary { TotalRequests, AllowedCount, BlockedCount, AllowedDomains, BlockedDomains, ByWorkflow map[string]*DomainAnalysis }
• pkg/cli/logs_report_firewall.go:22 — FirewallLogSummary { TotalRequests, AllowedRequests, BlockedRequests, AllowedDomains, BlockedDomains, RequestsByDomain, ByWorkflow map[string]*FirewallAnalysis }

Recommendation: A shared LogDomainSummary parameterized by the per-workflow value type, or simply normalize the field names (AllowedRequests/BlockedRequests is more descriptive) and share most fields.

Cluster 7: Audit Diff *Summary cluster — Near duplicates

Locations (same file):

• pkg/cli/audit_diff.go:50 — FirewallDiffSummary { NewDomainCount, RemovedDomainCount, StatusChangeCount, VolumeChangeCount, HasAnomalies, AnomalyCount }
• pkg/cli/audit_diff.go:240 — MCPToolsDiffSummary { NewToolCount, RemovedToolCount, ChangedToolCount, HasAnomalies, AnomalyCount }
• pkg/cli/audit_diff.go:299 — ToolCallsDiffSummary { NewToolCount, RemovedToolCount, ChangedToolCount, Run1TotalCalls, Run2TotalCalls }

Recommendation: Generic DiffSummary[T] struct { New, Removed, Changed int; HasAnomalies bool; AnomalyCount int } and per-type extensions for totals. Standardize on "New/Removed/Changed" instead of "NewDomain/NewTool".

Cluster 8: Engine identity (AuditEngineConfig / EngineConfig / AwInfo) — Semantic duplicates

Locations:

• pkg/cli/audit_expanded.go:21 — AuditEngineConfig { EngineID, EngineName, Model, Version, CLIVersion, FirewallVersion, MCPServers, TriggerEvent, Repository }
• pkg/workflow/engine.go:29 — EngineConfig (workflow runtime config)
• pkg/workflow/engine_definition.go:178 — ResolvedEngineTarget

Field names diverge between layers (ID vs EngineID), forcing manual mapping in extractEngineConfigWithInferredEngine.

Recommendation: Introduce EngineIdentity { EngineID, EngineName, Model, Version } in pkg/types, embed in both workflow.EngineConfig and cli.AuditEngineConfig. Eliminates the manual mapping.

Cluster 9: Workflow representation triple — Near duplicates

Locations (all in pkg/cli):

• pkg/cli/add_workflow_resolution.go:21 — ResolvedWorkflow { Spec *WorkflowSpec; Content []byte; SourceInfo *FetchedWorkflow; ... } (the struct comment explicitly notes Content is a "convenience accessor" duplicating SourceInfo.Content)
• pkg/cli/fetch.go:35 — FetchedWorkflow { Content []byte; CommitSHA; IsLocal; SourcePath; ConvertedFromJSON; JSONConversionWarnings }
• pkg/cli/spec.go:33 — WorkflowSpec
• pkg/cli/jsonworkflow_to_markdown.go:137 — GeneratedWorkflow
• pkg/cli/workflows.go:61 — GitHubWorkflow { ID, Name, Path, State }

Recommendation: Drop the duplicated Content []byte from ResolvedWorkflow (always read through SourceInfo); rename GitHubWorkflow → GitHubWorkflowListItem (different concept: GitHub API listing payload).

Cluster 10: ModelTokenUsage / ModelTokenUsageRow — Near duplicates

Locations:

• pkg/cli/token_usage.go:66 — ModelTokenUsage { Provider, InputTokens, OutputTokens, CacheReadTokens, CacheWriteTokens, Requests, DurationMs, ResponseBytes, EffectiveTokens }
• pkg/cli/token_usage.go:79 — ModelTokenUsageRow { Model, Provider, InputTokens, OutputTokens, CacheReadTokens, CacheWriteTokens, EffectiveTokens, Requests, AvgDuration }

Recommendation: Drop ModelTokenUsageRow; render the map directly with a display helper that formats DurationMs at print time.

Cluster 11: TokenUsageEntry / TokenUsageSummary / TokenUsageDiff token fields — Near duplicates

Locations:

• pkg/cli/token_usage.go:23 — TokenUsageEntry { InputTokens, OutputTokens, CacheReadTokens, CacheWriteTokens, EffectiveTokens, ... }
• pkg/cli/token_usage.go:51 — TokenUsageSummary { TotalInputTokens, TotalOutputTokens, TotalCacheReadTokens, TotalCacheWriteTokens, TotalEffectiveTokens, ... }
• pkg/cli/audit_diff.go:250 — TokenUsageDiff { Run1InputTokens, Run2InputTokens, Run1OutputTokens, Run2OutputTokens, Run1CacheReadTokens, Run2CacheReadTokens, Run1CacheWriteTokens, Run2CacheWriteTokens, Run1EffectiveTokens, Run2EffectiveTokens, ... }

Recommendation: type TokenCounts struct { Input, Output, CacheRead, CacheWrite, Effective int }. Embed in entry, use as value in summary totals, use a Run1, Run2 TokenCounts pair in the diff. Adding a sixth token class today requires editing three places — after this change, one.

Cluster 12: AddCommentConfig (deprecated empty) — Dead-code leftover

Location: pkg/workflow/add_comment.go:10 — type AddCommentConfig struct{} with comment saying "deprecated, use AddCommentsConfig"

Recommendation: Delete after verifying no callers, or alias type AddCommentConfig = AddCommentsConfig. 5-minute change.

---

Untyped Usages

Summary Statistics

• interface{} usages: 0 (the codebase uses any exclusively — good)
• any usages (non-test): ~1900 across 325 files
• map[string]any usages (non-test): dominant pattern; top-50 files account for >700 occurrences

Files with heaviest untyped usage:

1. pkg/parser/import_field_extractor.go — 74 any
2. pkg/workflow/compiler_safe_outputs_handlers.go — 52 any
3. pkg/parser/schema_suggestions.go — 45 any
4. pkg/parser/mcp.go — 42 any
5. pkg/workflow/tools.go, pkg/workflow/role_checks.go — 32 any each
6. pkg/workflow/trigger_parser.go — 31 any

Important context: The project has already invested in pkg/typeutil/convert.go (Lookup/Parse/Convert helpers) and pkg/workflow/tools_types.go (a Tools struct) explicitly to migrate away from map[string]any. The recommendations below align with that migration.

Category 1: any in function parameters (runtime type is consistent)

Example 1: getGitHub* accessor family

• Location: pkg/workflow/mcp_github_config.go:138, 163, 182, 194, 204, 264, 284, 414, 532
• Current: func getGitHubToken(githubTool any) string, getGitHubLockdown(githubTool any) bool, getGitHubType(githubTool any) string, ... (about a dozen accessors)
• Actual usage: every body opens with if toolConfig, ok := githubTool.(map[string]any); ok { ... } — the value is always the GitHub tool sub-map from frontmatter
• Suggested: func getGitHubToken(githubTool *GitHubToolConfig) string — WorkflowData.ParsedTools.GitHub is already typed
• Benefit: eliminates ~10 repeated type assertions; matches the existing typed pipeline

Example 2: PermissionsParser

• Location: pkg/workflow/permissions_parser.go:244 — func NewPermissionsParserFromValue(permissionsValue any) *PermissionsParser
• Actual usage: callers pass the permissions sub-map; GitHubActionsPermissionsConfig / GitHubAppPermissionsConfig already exist in pkg/workflow/frontmatter_types.go
• Suggested: accept the typed structs directly
• Benefit: replaces dynamic field probing with field access; the typed structs are already wired through FrontmatterConfig.PermissionsTyped

Example 3: Job-dependency helpers

• Location: pkg/workflow/compiler_jobs.go:55, 72, 92 — jobDependsOnPreActivation(jobConfig map[string]any) bool, jobDependsOnActivation, jobDependsOnAgent
• Actual usage: each function does jobConfig["needs"] then asserts to either []any or string
• Suggested: parse all custom job configs once into type JobConfig struct { Needs StringOrSlice; ... }
• Benefit: removes triple type-assert boilerplate in three near-identical functions

Example 4: Dependabot value normalizers

• Location: pkg/workflow/dependabot.go:543, 570, 601 — dependabotToAnySlice(value any) ([]any, bool), dependabotToStringAnyMap(value any) (map[string]any, bool), isYAMLNullOrEmptyScalar(value any) bool
• Suggested: define a DependabotUpdate struct (matching the documented schema) and unmarshal directly via yaml.Unmarshal once; have helpers operate on the struct
• Benefit: each consumer drops the second (value, ok) plumbing path

Example 5: OTLP header normalizer

• Location: pkg/workflow/observability_otlp.go:30 — normalizeOTLPHeadersForEndpoint(raw any, endpoint string) string
• Actual usage: switch on two cases (string, map[string]any)
• Suggested: type OTLPHeaders struct { Raw string; Map map[string]string } (or a sum type) parsed once at the frontmatter boundary
• Benefit: pushes the string-vs-map decision to a single parse site

Example 6: Workflow inputs extractor

• Location: pkg/workflow/workflow_inputs_extractor.go:40 — extractInputsFromParsedWorkflow(workflow map[string]any, trigger string) map[string]any
• Actual usage: 4 successive type assertions to walk workflow["on"][trigger]["inputs"], returning empty map on each failure
• Suggested: use a typed OnSection/WorkflowInputSpec (the project already has parser.FrontmatterResult and FrontmatterConfig.On)
• Benefit: removes the 4-level "type-assert-or-bail" pattern; would also fix the silent-empty error behavior

Category 2: map[string]any struct fields

Example 1: WorkflowData carries 5 untyped maps

• Location: pkg/workflow/compiler_types.go:483, 524, 537, 540, 551, 555
• Current: Tools map[string]any, Jobs map[string]any, Features map[string]any, Runtimes map[string]any, RawFrontmatter map[string]any, ResolvedMCPServers map[string]any
• Suggested:
  • Tools already has a parallel ParsedTools *Tools field, so Tools is a candidate for removal once readers migrate to ParsedTools
  • Runtimes already has *RuntimesConfig defined in frontmatter_types.go:21
  • Features only holds bool+string — map[string]FeatureValue (a typed sum) would be enough
  • Jobs should become map[string]JobConfig
• Benefit: this is the central state object — many downstream any usages stem from this one struct holding untyped maps

Example 2: Step lists held as []map[string]any

• Location: pkg/workflow/compiler_types.go:499, 568, 746 — WorkflowData.OnSteps, EngineConfigSteps, SecretMaskingConfig.Steps
• Suggested: []WorkflowStep — WorkflowStep already exists in pkg/workflow/step_types.go:18 with MapToStep/ToMap converters
• Benefit: the type already exists; using it closes the loop on the existing migration

Example 3: TriggerIR filters

• Location: pkg/workflow/trigger_parser.go:23, 29 — Filters map[string]any; AdditionalEvents map[string]any
• Actual usage: Filters carries branches/paths/tags/labels — a fixed set
• Suggested: type TriggerFilters struct { Branches []string; BranchesIgnore []string; Paths []string; PathsIgnore []string; Tags []string; TagsIgnore []string; Types []string; Labels []string }
• Benefit: parsing the 31 any references in trigger_parser.go becomes field access

Example 4: Import-input maps

• Location: pkg/parser/import_field_extractor.go:60, pkg/parser/import_bfs.go:53, 59 — importAccumulator.importInputs, nestedImportEntry.inputs, importBFSState.visitedInputs all map[string]any
• Suggested: define type ImportInputValues map[string]InputValue where InputValue is a small sum (string/int/bool); marshalling already JSON-encodes them in marshalImportInputValue
• Benefit: import_field_extractor.go (the rejig docs #1 hotspot at 74 any) and import_bfs.go (22 any) both rely on this — typing the inputs would cascade

Example 5: SafeOutputs handler-config builders

• Location: pkg/workflow/compiler_safe_outputs_handlers.go:6 (handler registry) and pkg/workflow/compiler_safe_outputs_builder.go:112 — type handlerBuilder func(*SafeOutputsConfig) map[string]any
• Current: each handler builds and returns a map[string]any keyed by string field names
• Suggested: a fixed HandlerConfig struct (or a union of well-known structs); the JSON shape stays the same via json.Marshal
• Benefit: largest single concentration (52 any in one file)

Example 6: GitHub MCP guard policies

• Location: pkg/workflow/mcp_github_config.go:284, 414 — getGitHubGuardPolicies(githubTool any) map[string]any, deriveSafeOutputsGuardPolicyFromGitHub(githubTool any) map[string]any
• Suggested: model GuardPolicy { Rules []GuardRule } explicitly
• Benefit: removes nested assertions from both sites and the renderer

Category 3: any struct fields

Example 1: WorkflowStep.ContinueOnError

• Location: pkg/workflow/step_types.go:28 — ContinueOnError any // Can be bool or string expression
• Suggested: type BoolOrExpr struct { Bool *bool; Expr string } with custom YAML/JSON unmarshal — there's precedent: TemplatableInt32 and TemplatableBool already exist for the same pattern
• Benefit: removes runtime checks at every consumer; matches the existing Templatable* family

Example 2: FrontmatterConfig.Checkout

• Location: see pkg/workflow/frontmatter_parsing.go:54 (config.Checkout.(bool))
• Current: Checkout any (single obj, array of objs, or false-to-disable)
• Suggested: CheckoutValue union type with custom UnmarshalJSON — the parser already calls ParseCheckoutConfigs(config.Checkout), so the union can be parsed at unmarshal time
• Benefit: the special boolean-false case becomes a typed field rather than a runtime .(bool) branch

Example 3: ThreatDetectionConfig.Steps and PostSteps

• Location: pkg/workflow/threat_detection.go:19, 20 — both []any
• Suggested: []WorkflowStep

Example 4: SafeOutputsConfig top-level steps

• Location: pkg/workflow/compiler_types.go:692 Steps []any and :746 Steps []map[string]any
• Suggested: []WorkflowStep
• Benefit: every YAML emitter currently re-asserts each entry to map[string]any before marshalling

Category 4: Untyped constants with semantic meaning

The constants package is already well-typed (pkg/constants/constants.go uses time.Duration, fs.FileMode, semantic types like LineLength, CommandPrefix, WorkflowID). Remaining candidates:

Example 1: Network ports as int literals

• Location: pkg/constants/constants.go:100-130 — DefaultMCPGatewayPort = 8080, ClaudeLLMGatewayPort = 10000, etc.
• Suggested: type NetworkPort uint16 and const DefaultMCPGatewayPort NetworkPort = 8080
• Benefit: prevents mixing with arbitrary ints; range-checks at compile time

Example 2: Budget integers

• Location: pkg/constants/constants.go:263, 266 — DefaultMaxEffectiveTokens int64 = 25000000, DefaultMaxRuns = 500
• Suggested: type TokenBudget int64, type RunCount int
• Benefit: aligns with existing semantic-type pattern in the same file

Example 3: MCP type strings

• Location: pkg/parser/mcp.go:21 — var ValidMCPTypes = []string{"stdio", "http", "local"} plus a stringly-typed IsMCPType switch
• Suggested: type MCPType string; const ( MCPTypeStdio MCPType = "stdio"; ... )
• Benefit: removes magic strings; turns IsMCPType(s string) into a type-safe enum

Example 4: AI engine identifier

• Location: pkg/workflow/compiler_types.go:486 — AI string // "claude" or "codex"
• Suggested: type EngineID string; const ( EngineClaude EngineID = "claude"; EngineCodex EngineID = "codex"; ... )
• Benefit: this string drives engine dispatch across many files; today the matches are stringly-typed

Acceptable any uses (intentional — do not change)

• pkg/parser/schema_suggestions.go, pkg/parser/schema_compiler.go — operate on parsed JSON schema (arbitrary user-supplied structure)
• pkg/parser/yaml_import.go, frontmatter_hash.go, frontmatter_content.go — root YAML/JSON deserialization sinks
• pkg/typeutil/convert.go — purpose-built helpers for heterogeneous values; this is the place where any legitimately lives
• pkg/workflow/yaml.go:371 formatYAMLValue — generic YAML formatter, must accept any node value
• pkg/linters/* — AST analyzers receive ast.Node, intrinsically polymorphic
• pkg/cli/codemod_* (~40 files) — codemod framework operates on raw YAML nodes; any here is structural, not domain
• pkg/workflow/expression_extraction.go:505 marshalImportInputValue(value any) string — generic JSON marshaller for arbitrary input values

Recommended Action Plan (Prioritized)

Priority 1 — Highest leverage

1. Finish the WorkflowData.ParsedTools migration (pkg/workflow/compiler_types.go:483). Remove the parallel Tools map[string]any field. This unlocks typing the entire getGitHub* accessor family (pkg/workflow/mcp_github_config.go) and similar Claude/Codex/Gemini tool expanders.
2. Extract RunAnalyses embeddable to deduplicate ProcessedRun / RunSummary / DownloadResult in pkg/cli/logs_models.go. ~2-3 hours, removes hand-copying between three structs.
3. Type import-input values (pkg/parser/import_field_extractor.go, the single hottest any file at 74 occurrences). Cascades to import_bfs.go.

Priority 2 — Significant cleanup

4. Collapse SafeOutput entity-config family into a SafeOutputAllowBlockConfig base + thin per-entity wrappers. ~25 structs in pkg/workflow/ lose their duplicated bodies.
5. Migrate step lists to []WorkflowStep everywhere []map[string]any is currently used (compiler_types.go:499, 568, 746, threat_detection.go:19, 20). The type and converters already exist.
6. Type TriggerIR.Filters as TriggerFilters struct in pkg/workflow/trigger_parser.go.
7. Unify token-count fields via shared TokenCounts (Cluster 11) — eliminates triple maintenance.

Priority 3 — Polish

8. Consolidate MCPServer*Stats types (Cluster 3).
9. Merge PRInfo + PullRequest (Cluster 5).
10. Add typed enums for MCPType and EngineID.
11. Add NetworkPort / TokenBudget semantic types to constants.
12. Delete deprecated AddCommentConfig (Cluster 12).

Analysis Metadata

• Go files analyzed: 839 non-test files under pkg/
• Struct type definitions: 575
• Duplicate clusters reported: 12 (7 near, 4 semantic, 1 dead-code)
• any usages: ~1900 across 325 files
• Detection method: ripgrep over pkg/ + targeted file reads, cross-referenced against the existing typed migration scaffolding (pkg/typeutil, pkg/workflow/tools_types.go, pkg/workflow/step_types.go)
• Analysis date: 2026-05-25

Generated by ✍️ Typist - Go Type Analysis · opus47 44.1M · ◷

• expires on May 26, 2026, 12:50 PM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-05-26T12:50:43.916Z.

Closed by Workflow