[typist] Typist Go Type Consistency Analysis

[github-actions[bot]]

Analysis of repository: github/gh-aw

Executive Summary

I scanned all 1,205 non-test .go files under pkg/ (1,129 type definitions across ~33 packages), looking for duplicated type definitions and untyped (interface{}/any/untyped-constant) usage. The good news first: this codebase is already quite disciplined about avoiding type duplication. Most same-named types across packages (ActionPin, ContainerPin, InputDefinition, SanitizeOptions, LogMetrics, ToolCallInfo, etc.) turned out to be intentional type X = otherpkg.X re-export aliases pointing at one canonical definition, not copy-pasted redundancy — so I excluded those from the findings below.

The real duplication is small but concrete, concentrated in pkg/cli: JobStep/JobStepData are 100%-identical structs that require an explicit conversion (JobStepData(step)) every time data crosses between them, and four independently-defined "log entry" structs (AccessLogEntry, FirewallLogEntry, AuditLogEntry, GatewayLogEntry) all model the same concept — a parsed log line — with no shared base type. On the untyped-usage side, the most actionable findings aren't interface{} (that's rare here, and appropriately reserved for genuine YAML/JSON tree-walking) but a handful of untyped numeric/duration constants in pkg/constants whose units (bytes, minutes, hours, credits) are documented only in comments, not enforced by the compiler — plus a couple of any-returning helper functions in pkg/parser/schema_suggestions.go that always return one concrete type.

Full Analysis Report

Duplicated Type Definitions

Summary Statistics

• Total files analyzed: 1,205 (non-test .go under pkg/)
• Total types analyzed: 1,129
• Candidate clusters found: 13 (name-collision or synonym matches)
• Confirmed re-export aliases (excluded): 9 clusters (ActionYAMLInput, ActionPin, ActionPinsData, ContainerPin, SHAResolver, InputDefinition, SanitizeOptions, LogMetrics, ToolCallInfo) — each is a type X = otherpkg.X alias in pkg/workflow/pkg/cli pointing at one canonical definition in pkg/actionpins, pkg/types, or pkg/stringutil. Verified directly, e.g. pkg/workflow/action_pins.go:25 is literally type ActionPin = actionpins.ActionPin.
• Genuine exact duplicates: 1 (JobStep/JobStepData)
• Genuine near duplicates: 1 confirmed (JobInfo/JobData), 1 questionable (ExperimentInfo/ExperimentData — see note)
• Semantic duplicates: 1 cluster (4 log-entry structs)

Cluster 1: JobStep / JobStepData — exact duplicate

Impact: Medium — same shape defined twice, forcing a manual conversion at the one place they meet

Locations:

1. pkg/cli/logs_models.go — type JobStep struct { Name, Status, Conclusion string }
2. pkg/cli/audit_report.go — type JobStepData struct { Name, Status, Conclusion string }

Evidence of the pain point (pkg/cli/audit_report.go:385):

Steps: sliceutil.Map(jobDetail.Steps, func(step JobStep) JobStepData {
    return JobStepData(step)
}),

Because the two types are structurally identical but nominally distinct, buildAuditJobs has to map over every step just to do a bare struct conversion — this line of code exists purely to satisfy the type system, not because the data differs.

Recommendation: Drop JobStepData and use JobStep (with its existing json tags, if any need adding) in audit_report.go directly. Delete the sliceutil.Map/conversion call.

• Estimated effort: 30–60 minutes
• Benefits: Removes a no-op conversion step and one duplicate type

---

Cluster 2: JobInfo / JobData — near duplicate (Info/Data synonym)

Impact: Medium — JobInfo is the more widely used and embedded type (JobInfoWithDuration embeds it; referenced in logs_github_api.go, logs_summary.go, and 8+ test files), while JobData in audit_report.go duplicates Name/Status/Conclusion but adds a Duration string field (pre-formatted) instead of the StartedAt/CompletedAt time.Time pair JobInfo uses, and nests []JobStepData instead of []JobStep.

Recommendation: Since JobData is really "a JobInfo rendered for the audit report, with duration pre-formatted as a string," consider deriving it from JobInfo via a constructor function (func toJobData(j JobInfo) JobData) rather than an independently maintained struct, or add a FormattedDuration() string method to JobInfo and drop JobData entirely once JobStepData (Cluster 1) is removed.

• Estimated effort: 1–2 hours (touches audit report rendering + tests)
• Benefits: One source of truth for "what a job looks like"

---

Cluster 3: ExperimentInfo / ExperimentData — name-only near duplicate ⚠️

Note: this pair was flagged by the automated name-synonym pass (Info/Data suffix match) but the fields don't actually overlap: ExperimentInfo (pkg/cli/experiments_command.go) is {WorkflowID, Branch string; Experiments, TotalRuns int; LastRun string}, while ExperimentData (pkg/cli/audit_report_experiments.go) is {Assignments map[string]string; CumulativeCounts map[string]map[string]int} — a completely different shape for a different purpose (variant-assignment bookkeeping vs. a list-view summary). This is very likely just naming-convention coincidence, not duplication. No action needed beyond, optionally, renaming one of them to avoid the misleading impression of overlap (e.g. ExperimentAssignmentData).

---

Cluster 4: Log-entry structs — semantic duplicate

Impact: Low/Medium — four structs, one per log source, all modeling "a parsed line from a log," with no shared base and largely overlapping intent:

1. pkg/cli/access_log.go — AccessLogEntry {Timestamp, Duration, ClientIP, Status, Size, Method, URL, User, Hierarchy, Type string}
2. pkg/cli/firewall_log.go — FirewallLogEntry {Timestamp, ClientIPPort, Domain, DestIPPort, Proto, Method, Status, Decision, URL, UserAgent string}
3. pkg/cli/firewall_policy.go — AuditLogEntry {Schema string, Timestamp float64, Client, Host, Dest, Method, Decision, URL string, Status int}
4. pkg/cli/gateway_logs_types.go — GatewayLogEntry (19 fields — timestamp, level, type, event, server/tool identifiers, duration, sizes, status, error, tags, author info)

Recommendation: These genuinely differ in source format (access log vs. firewall log vs. audit JSON vs. gateway JSONL) so full unification isn't warranted — but a shared LogEntryCommon { Timestamp string; Method, URL, Status string } embed (mirroring the pattern the codebase already uses for AnalysisBase/DomainBuckets in pkg/cli/firewall_policy.go, which the team's own comment says exists to "eliminate code duplication") would cut the repeated field declarations and give parsers/renderers a common interface to work against.

• Estimated effort: 3–4 hours
• Benefits: Consistent handling in any code that treats "a log entry" generically (filtering, redaction, export)

---

Untyped Usages

Summary Statistics

• interface{} usages: rare and mostly confined to test/linter-testdata files (no high-value production hits found)
• any usages requiring attention: 5 (4 function signatures, 1 struct field)
• Untyped constants flagged: 8, concentrated in pkg/constants

Category 1: any-returning helpers that always return one concrete type

Impact: Medium — these discard compile-time guarantees for callers, in code that's otherwise fully within the maintainers' control (not truly polymorphic).

generateStringExample — pkg/parser/schema_suggestions.go:386

func generateStringExample(schema map[string]any) any {
    if enum, ok := schema["enum"].([]any); ok && len(enum) > 0 {
        if str, ok := enum[0].(string); ok {
            return str
        }
    }
    return "string"
}

Every return path is a string. Fix: func generateStringExample(schema map[string]any) string.

generateArrayExample — pkg/parser/schema_suggestions.go:405

Both return paths produce []any. Fix: func generateArrayExample(schema map[string]any) []any.

generateObjectExample — pkg/parser/schema_suggestions.go:417

Always constructs and returns map[string]any. Fix: func generateObjectExample(schema map[string]any) map[string]any.

LayoutEmphasisBox — pkg/console/layout_wasm.go:16

func LayoutEmphasisBox(content string, color any) string {

The color parameter is typed any but is never read in the function body — it's dead. Fix: remove the parameter, or type it concretely (e.g. lipgloss.Color) and actually apply it.

---

Category 2: any-typed struct field with unclear contract

FormField.Value — pkg/console/console_types.go:50

type FormField struct {
    ...
    Value any // Pointer to the value to store the result
    ...
}

The comment implies a small closed set of pointer kinds, but the field is any, and it appears to have no live (non-wasm-stub) consumer that unwraps it via type assertion. Fix: either delete the unused path, or replace with a small tagged interface (FormValue implemented by *StringValue/*BoolValue) keyed off the existing Type discriminator.

---

Category 3: Untyped constants with an implied unit

Impact: Medium — the unit (bytes, minutes, hours, credits) is documented only in a comment, not enforced by the type system, so a future mix-up (e.g. passing a byte count where a duration is expected) would compile silently.

Constant	File	Current	Suggested
DefaultMCPGatewayPayloadSizeThreshold	pkg/constants/constants.go:245	= 524288	ByteSize = 524288
DefaultMaxDailyAICredits	pkg/constants/constants.go:329	= "5000" (string!)	typed int64, format to string only at the template-embedding site — note its siblings DefaultMaxAICredits/DefaultDetectionMaxAICredits are already typed int64
DefaultMaxRuns	pkg/constants/constants.go:332	= 500	int64 = 500 (match sibling constants)
MaxSymlinkDepth	pkg/constants/constants.go:615	= 5	int = 5
DefaultRateLimitWindow	pkg/constants/job_constants.go:260	= 60 (minutes, bare int)	time.Duration = 60 * time.Minute, matching how DefaultToolTimeout already uses time.Duration a few dozen lines earlier
DefaultActionFailureIssueExpiresHours	pkg/workflow/repo_config.go:63	= 24 * 7 (hours)	time.Duration = 24 * 7 * time.Hour
maxScannerBufferSize	pkg/cli/gateway_logs_types.go:15	= 1024 * 1024 (reused in 6+ files)	int = 1024 * 1024, consider promoting to pkg/constants given its reuse
maxRedirectDepth	pkg/cli/update_redirects.go:18	= 20	int = 20; same recursion-depth-limit pattern as MaxSymlinkDepth

Also worth noting: defaultCacheIntegrityLevel in pkg/workflow/cache_integrity.go:17 is a bare "none" string standing in for a proper enum type (GitHubIntegrityLevel) that already exists in the same package with typed constants like GitHubIntegrityApproved. Because it's untyped, two call sites do string(github.MinIntegrity) conversions just to compare against it. Adding a GitHubIntegrityNone typed constant would remove both conversions.

---

Refactoring Recommendations

Priority 1: Remove the JobStep/JobStepData conversion (Cluster 1)

Delete JobStepData, use JobStep directly in audit_report.go, drop the sliceutil.Map conversion.
Estimated effort: 30–60 minutes · Impact: Medium

Priority 2: Fix any-returning schema-suggestion helpers

Narrow generateStringExample/generateArrayExample/generateObjectExample return types in pkg/parser/schema_suggestions.go; remove the dead color any parameter from LayoutEmphasisBox.
Estimated effort: 1–2 hours · Impact: Medium (mostly clarity + removes dead code)

Priority 3: Type the constants in pkg/constants that carry an implied unit

Introduce ByteSize/reuse time.Duration for the eight constants listed above; fix DefaultMaxDailyAICredits specifically since it's a string standing in for a number.
Estimated effort: 2–3 hours (small diffs, but touches call sites doing manual parsing/formatting) · Impact: Medium — prevents unit-confusion bugs at compile time

Priority 4 (optional, lower confidence): Consolidate JobInfo/JobData and give the four log-entry structs a shared base

Estimated effort: 4–6 hours combined · Impact: Low/Medium — nice-to-have consistency, not urgent

---

Implementation Checklist

• Remove JobStepData, use JobStep directly (Priority 1)
• Narrow return types of generateStringExample/generateArrayExample/generateObjectExample; remove dead color any param from LayoutEmphasisBox (Priority 2)
• Add typed units (ByteSize, time.Duration) to the 8 flagged constants in pkg/constants / pkg/workflow/repo_config.go / pkg/cli (Priority 3)
• Fix defaultCacheIntegrityLevel to use the existing GitHubIntegrityLevel enum
• (Optional) Consolidate JobInfo/JobData; add a shared base for the 4 log-entry structs
• Rename ExperimentData if the Info/Data naming coincidence is considered confusing (no structural change needed)
• Run full test suite after each change

Analysis Metadata

• Total Go Files Analyzed: 1,205 (non-test, under pkg/)
• Total Type Definitions: 1,129
• Genuine Duplicate Clusters: 4 (1 exact, 1 confirmed near, 1 semantic, 1 questionable/name-only)
• Confirmed Re-export Aliases (not duplication): 9
• Untyped Usage Locations Flagged: 13 (5 any usages, 8 untyped constants)
• Detection Method: Serena semantic analysis (get_symbols_overview, find_referencing_symbols) + targeted pattern matching, cross-referenced to confirm real vs. alias duplication
• Analysis Date: 2026-08-11

Generated by ✍️ Typist - Go Type Analysis · agent · 481.3 AIC · ⌖ 49.2 AIC · ⊞ 8.7K · ◷

• expires on Aug 12, 2026, 3:50 AM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Typist - Go Type Analysis.

A newer discussion is available at Discussion #52283.