[typist] Typist — Go Type Consistency Analysis (duplicated types & untyped usages)

[github-actions[bot]]

🔤 Typist - Go Type Consistency Analysis

Analysis of repository: github/gh-aw

Executive Summary

Good news first: this codebase is already disciplined about cross-package type sharing. Rather than copy-pasting struct definitions, it leans on Go type aliases — e.g. type InputDefinition = types.InputDefinition, type SanitizeOptions = stringutil.SanitizeOptions, type LogMetrics = workflow.LogMetrics, the ActionPin/ActionYAMLInput family re-exported from pkg/actionpins, and fan-outs like MissingDataConfig/MissingToolConfig/ReportIncompleteConfig = IssueReportingConfig and CloseIssuesConfig/ClosePullRequestsConfig/CloseDiscussionsConfig = CloseEntityConfig. So most same-name-in-two-places hits are intentional single-source-of-truth aliases, not duplication. 🎉

Across ~990 non-test .go files under pkg/ (~1,015 top-level types), the genuine opportunities cluster into: three parallel shapes for GitHub rate-limit data, a family of per-tool scanner finding structs, several MCP-server-config representations, and two parallel audit diff vs. comparison delta vocabularies. On typing, any/interface{} appears in 160+ files — mostly legitimate YAML/frontmatter parsing, with a few spots where a concrete type would remove runtime assertions. Nothing on fire — a friendly punch-list. 🙂

Method: repo-wide type/any pattern extraction plus Serena-assisted spot verification. Clusters marked exact were read and confirmed; semantic ones are judgment calls for a human.

Full Analysis Report

Duplicated Type Definitions

• Types analyzed: ~1,015 (non-test, pkg/)
• Actionable clusters: 5 · Exact cross-package copies: 0 (sharing done via aliases ✅) · Near: 3 · Semantic: 2

⚠️ False positives (ignore): SpinnerWrapper, ProgressBar, RepositoryFeatures each appear twice only as //go:build wasm build-tag variants — the same name is required. Leave them.

Cluster 1 (near) — GitHub rate-limit data, 3 shapes in pkg/cli

• logs_github_rate_limit_usage.go:33 GitHubRateLimitEntry (Reset as string)
• logs_rate_limit.go:36 rateLimitResource (Reset as int64)
• update_org.go:48 orgRateLimitResponse

Same limit/remaining/used/reset concept modeled 3x, with a Reset type mismatch (string vs int64). Fix: one shared RateLimitBucket; reconcile the Reset encoding. Effort 2-3h.

Cluster 2 (near) — Per-tool scanner finding structs

• cli/zizmor.go:23 zizmorFinding · cli/poutine.go:24 poutineFinding · cli/runner_guard.go:21 runnerGuardFinding · cli/audit_report.go:63 Finding · workflow/markdown_security_scanner.go:57 SecurityFinding

Overlapping fields (rule id, message, severity, path, line). Fix: keep thin per-tool decode structs, add one internal ScanFinding, convert at each boundary. Effort 3-4h.

Cluster 3 (near) — Experiment records (exported vs. JSON twin)

• cli/experiments_command.go:41 ExperimentRunRecord vs cli/audit_report_experiments.go:42 experimentRunRecord; also ExperimentState vs experimentStateJSON.

Fix: if the twins only decode the same JSON, delete them and decode into the exported types. Effort 1-2h.

Cluster 4 (semantic) — MCP-server config shapes across packages

• types/mcp.go:6 BaseMCPServerConfig · workflow/tools_types.go:430 MCPServerConfig · parser/mcp.go:39 RegistryMCPServerConfig (+MCPServerInfo) · cli/mcp_registry.go:19 MCPRegistryServerForProcessing · cli/mcp_registry_types.go:26 ServerDetail · cli/mcp_config_file.go:22 VSCodeMCPServer

Layered-by-design, so not a full merge — but check whether the others can embed BaseMCPServerConfig instead of redeclaring overlapping fields. Effort 4-6h.

Cluster 5 (semantic) — Audit diff vs. comparison deltas

• cli/audit_diff.go — AuditDiff, FirewallDiff, MCPToolsDiff, ToolCallsDiff, TokenUsageDiff, and more
• cli/audit_comparison.go — AuditComparisonDelta, AuditComparisonIntDelta, AuditComparisonStringDelta, and more

Two parallel difference-between-two-runs vocabularies. Consider a generic Delta[T any] struct{ Before, After T; Changed bool } to collapse the int/string delta twins. Effort 3-5h.

Untyped Usages

• any/interface{} in 160+ non-test files; dominant pattern is map[string]any for YAML/frontmatter (largely idiomatic at the boundary).

Category 1 — map[string]any threaded deep (medium):

• cli/codemod_factory.go:13 PostTransformFunc(... frontmatter map[string]any, fieldValue any) — both untyped
• cli/outcome_eval_update.go:25 mutableStateLoader ... (map[string]any, bool, error)
• cli/devcontainer.go:45 DevcontainerFeatures map[string]any · workflow/mcp_config_types.go:66 MapToolConfig map[string]any

Decode into structs near the boundary where the shape is known; keep the map only at the raw-YAML edge.

Category 2 — any factory + untyped scope (low/medium):

• workflow/safe_output_handlers.go:17 NewConfig func() any (40+ uses). A marker interface would restore some compile-time safety.
• workflow/tools_types.go:302 type GitHubReposScope any — the comment documents two real shapes (string or slice); a small sum-type or decode-time validation fits.

Category 3 — untyped string constants (low): pkg/constants already defines strong types (JobName, StepID, EngineName, WorkflowID, URL, and more), but filename/artifact-name constants in job_constants.go (SafeOutputArtifactName, AgentOutputFilename, SarifFileName) stay untyped. Optional — type only to prevent mixing artifact names with arbitrary strings.

Prioritized Recommendations

1. Medium — shared RateLimitBucket, reconcile Reset encoding (Cluster 1). 2-3h.
2. Medium — internal ScanFinding + per-tool adapters (Cluster 2). 3-4h.
3. Low/Medium — remove experiment-record twins; early-decode known-shape frontmatter (Cluster 3 + Untyped Cat 1). 3-5h.
4. Human judgment — embed BaseMCPServerConfig (Cluster 4); generic Delta[T] (Cluster 5); marker interface for the safe-output factory (Cat 2).

Checklist

• Shared RateLimitBucket + Reset encoding reconciled
• Internal ScanFinding + adapters
• Delete experiment-record twins
• Evaluate BaseMCPServerConfig embedding
• Evaluate generic Delta[T]
• Early-decode known-shape frontmatter maps
• Run full test suite after each change
• Do not touch the wasm build-tag pairs (false positives)

Files analyzed: ~990 · Types: ~1,015 · Clusters: 5 (+3 false positives) · Untyped footprint: 160+ files · Date: 2026-07-02

Generated by ✍️ Typist - Go Type Analysis · 328.8 AIC · ⌖ 15.9 AIC · ⊞ 8.8K · ◷

• expires on Jul 3, 2026, 4:17 AM UTC-08:00