[typist] Typist: Go type consistency analysis (duplicates and untyped usages)

[github-actions[bot]]

🔤 Typist — Go Type Consistency Analysis

Analysis of repository: github/gh-aw

Executive Summary

Good news first: this codebase already has strong type-consolidation discipline. I scanned 1,383 non-test .go files under pkg/ (≈1,406 top-level type declarations) for duplicated types, and nearly every apparent "duplicate" turned out to already be intentionally resolved — most cross-package name collisions (ActionPin, ActionYAMLInput, ActionPinsData, ContainerPin, SHAResolver, InputDefinition, SanitizeOptions, LogMetrics, ToolCallInfo) are type X = otherpkg.X aliases, and the ValidationError "duplicate" is actually a deliberate shared-interface pattern (pkg/validationerror) that parser.ValidationError and workflow.WorkflowValidationError both implement by embedding a common Payload. Only two name-alike pairs remain as loose ends, and both turn out to have little-to-no real field overlap — likely coincidental naming rather than duplication.

On the untyped-usage side, interface{}/any usage is already rare and mostly legitimate (JSON/YAML interop, or dead code behind build tags). After filtering out ~19,800 false-positive matches of the English word "any" in comments/strings, only 3 genuine any-as-type usages and 2 interface{} usages remain worth a look. The most actionable opportunity is a small cluster of untyped numeric constants (timeouts expressed as bare int in hours/minutes/seconds) that would benefit from named unit types — cheap wins that prevent unit-confusion bugs.

Full Analysis Report

Duplicated Type Definitions

Summary Statistics

• Total files analyzed: 1,383
• Total types analyzed: ~1,406
• Clusters flagged by initial pass: 12
• Already resolved via type alias (not a problem): 9
• Intentional shared-interface design (not a problem): 1
• Near-duplicate names, low real overlap (low priority): 2

Already consolidated — no action needed

These name collisions are not duplication bugs; they're type X = otherpkg.X aliases that let pkg/workflow and pkg/cli reference shared types from pkg/actionpins and pkg/types without an extra import, confirmed by reading the alias declarations directly:

Type	Canonical package	Alias location
ActionYAMLInput, ActionPin, ActionPinsData, ContainerPin, SHAResolver	pkg/actionpins	pkg/workflow/action_pins.go
InputDefinition	pkg/types	pkg/workflow/inputs.go
SanitizeOptions	pkg/stringutil	pkg/workflow/strings.go
LogMetrics, ToolCallInfo	pkg/workflow	pkg/cli/logs_models.go

Recommendation: none — this is the pattern the team should keep using when a type needs to be shared across packages.

ValidationError — intentional shared-interface pattern

pkg/validationerror.ValidationError is a documented interface (error + ValidationField/Value/Reason/Suggestion); pkg/parser.ValidationError and pkg/workflow.WorkflowValidationError both implement it by embedding validationerror.Payload. The package doc comment explicitly says this is by design so errors.As can uniformly detect structured validation errors regardless of producing package. No action needed.

Near-duplicate names with low real overlap (low priority)

• JobInfo (pkg/cli/logs_models.go, 23 fields) vs JobData (pkg/cli/audit_report.go, 5 fields) — same package, Info/Data naming pattern, but only 3 of 23 fields actually match (Name, Status, Conclusion) — ~13% overlap. JobInfo mirrors the raw GitHub Actions Jobs API response; JobData is a slimmed-down report-rendering DTO. Likely intentional separation of concerns, not a duplicate to merge — but worth a short doc comment on each explaining why both exist, since the names invite confusion.
• ExperimentInfo (pkg/cli/experiments_state.go) vs ExperimentData (pkg/cli/audit_report_experiments.go) — 0% field overlap (WorkflowID/Branch/Experiments/TotalRuns/LastRun vs Assignments/CumulativeCounts). This looks like a pure naming coincidence rather than duplication — recommend renaming one (e.g. ExperimentSummary / ExperimentAssignmentCounts) purely for readability, not correctness.

---

Untyped Usages

Summary Statistics

• Raw interface{} matches (non-test): ~33 across 7 files (most in pkg/linters/**/testdata fixtures or already-typed contexts) → 2 genuine findings
• Raw any matches (non-test): ~19,808 across 1,582 files (overwhelmingly the English word in comments/docs) → 3 genuine type-usage findings
• Untyped constants worth typing: 6

Category 1: interface{} / any findings

#	Location	Issue	Suggested fix	Impact
1	pkg/console/layout_wasm.go:16	func LayoutEmphasisBox(content string, color any) string — parameter unused, zero callers anywhere in pkg	Delete the dead parameter, or give it a concrete Color-like type if intended for future use	Dead code in a js/wasm-tagged stub; zero functional risk
2	pkg/console/console_types.go:50	FormField.Value any — but FormField/RunForm appear unused by the real (non-wasm) form implementation (pkg/console/prompt_form.go uses huh/v2 directly)	Delete FormField/RunForm if truly vestigial, or replace any with a discriminated union keyed by Type	Low risk — likely dead API surface
3	pkg/cli/mcp_tools_privileged.go:458	auditArgs.RunID any / RunIDOrURL any — jsonschema docs say these only ever accept string-or-number	Introduce a small StringOrInt type with custom UnmarshalJSON shared across "run ID or URL" MCP args	Low-medium — works today, but a shared type would remove repeated ad-hoc type switches

Category 2: Untyped numeric constants (unit-confusion risk)

#	Location	Current	Suggested
1	pkg/workflow/cache_integrity.go:17	const defaultCacheIntegrityLevel = "none" — duplicates the existing GitHubIntegrityLevel enum value GitHubIntegrityNone declared in pkg/workflow/tools_types.go:300	const defaultCacheIntegrityLevel GitHubIntegrityLevel = GitHubIntegrityNone; change cacheIntegrityLevel() to return GitHubIntegrityLevel instead of string
2	pkg/cli/run_workflow_execution.go:25	const workflowCompletionWaitTimeoutMinutes = 6 * 60 (passed as bare int to WaitForWorkflowCompletion(..., timeoutMinutes int, ...))	type Minutes int (or use time.Duration at the call site) so callers can't pass seconds/hours by mistake
3	pkg/workflow/publish_code_coverage.go:18	const defaultCodeCoverageWaitForProcessingTimeout = 160 (seconds, per comment only)	type Seconds int
4	pkg/workflow/maintenance_workflow.go:128	const defaultNoOpIssueExpirationHours = 24 * 30, compared/aggregated with other bare-int *Hours constants in the same function	Shared type Hours int for this and finding #5
5	pkg/workflow/repo_config.go:63	const DefaultActionFailureIssueExpiresHours = 24 * 7 (exported, referenced across files/tests)	Same type Hours int
6	pkg/constants/job_constants.go:310-311	DefaultRateLimitWindow = 60 (minutes, per comment; also user-configurable from workflow frontmatter as a plain int)	type Minutes int for the window; DefaultRateLimitMax is a plain count and is fine as-is

The highest-confidence item is #1 — it's a plain duplicate of an already-existing typed enum value in the same package, so a future edit could typo "None" here without the compiler catching it, unlike every other integrity-level call site which already uses GitHubIntegrityNone.

---

Recommendations, prioritized

Priority 1 — quick, safe win: Fix pkg/workflow/cache_integrity.go:17 to reuse GitHubIntegrityNone instead of a bare string literal. Single-file change, ~15 minutes.

Priority 2 — unit-safety cleanup: Introduce type Hours int / type Minutes int / type Seconds int in an appropriate shared location (e.g. pkg/workflow or a small pkg/timeutil) and apply to the 5 remaining untyped-constant findings above. Each is independent and low-risk; ~2-3 hours total including tests.

Priority 3 — cosmetic / optional: Rename ExperimentInfo/ExperimentData to less confusable names, and add doc comments clarifying why JobInfo and JobData are intentionally separate. No functional change.

Priority 4 — cleanup, if confirmed dead: Remove pkg/console/layout_wasm.go's unused color any parameter and the unused FormField/RunForm in pkg/console/console_types.go/form_wasm.go, or wire them up properly if they're meant to be used.

Not recommended: consolidating the 9 already-aliased types (ActionPin family, InputDefinition, SanitizeOptions, LogMetrics, ToolCallInfo) or touching ValidationError — both are already the correct pattern.

Analysis Metadata

• Total Go files analyzed: 1,383 (non-test, under pkg/)
• Total type definitions surveyed: ~1,406
• Duplicate clusters flagged / confirmed real: 12 flagged, 2 low-priority near-duplicates confirmed, 10 resolved as intentional (aliases or shared-interface design)
• Untyped usage locations flagged / confirmed real: ~19,841 raw matches, 11 confirmed genuine (2 interface{}, 3 any, 6 untyped constants)
• Detection method: Serena semantic analysis (get_symbols_overview) + targeted grep, manually verified against actual call sites
• Analysis date: 2026-09-29

Generated by ✍️ Typist - Go Type Analysis · claude · agent · 732.6 AIC · ⌖ 6.37 AIC · ⊞ 6.3K · ◷

• expires on Sep 30, 2026, 3:48 AM UTC-08:00