feat: compile-time validation of needs: ordering on custom safe-output jobs

pkg/constants/constants.go

@@ -629,6 +629,10 @@ const APMJobName JobName = "apm"
 const IndexingJobName JobName = "indexing"
 const PreActivationJobName JobName = "pre_activation"
 const DetectionJobName JobName = "detection"
+const SafeOutputsJobName JobName = "safe_outputs"
+const UploadAssetsJobName JobName = "upload_assets"
+const ConclusionJobName JobName = "conclusion"
+const UnlockJobName JobName = "unlock"
 const SafeOutputArtifactName = "safe-output"
 const AgentOutputArtifactName = "agent-output"
 

pkg/workflow/compiler.go

@@ -174,6 +174,12 @@ func (c *Compiler) validateWorkflowData(workflowData *WorkflowData, markdownPath
 		return formatCompilerError(markdownPath, "error", err.Error(), err)
 	}
 
+	// Validate safe-job needs: declarations against known generated job IDs
+	log.Printf("Validating safe-job needs declarations")
+	if err := validateSafeJobNeeds(workflowData); err != nil {
+		return formatCompilerError(markdownPath, "error", err.Error(), err)
+	}
+
 	// Emit warnings for push-to-pull-request-branch misconfiguration
 	log.Printf("Validating push-to-pull-request-branch configuration")
 	c.validatePushToPullRequestBranchWarnings(workflowData.SafeOutputs, workflowData.CheckoutConfigs)

pkg/workflow/safe_jobs_needs_validation.go

@@ -0,0 +1,193 @@
+package workflow
+
+import (
+	"fmt"
+	"sort"
+	"strings"
+
+	"github.com/github/gh-aw/pkg/constants"
+	"github.com/github/gh-aw/pkg/logger"
+	"github.com/github/gh-aw/pkg/stringutil"
+)
+
+var safeJobsNeedsValidationLog = logger.New("workflow:safe_jobs_needs_validation")
+
+// validateSafeJobNeeds validates the needs: declarations on custom safe-output jobs.
+//
+// For each custom safe-job, every entry in its needs: list must refer to a job that
+// will actually exist in the compiled workflow. Valid targets are:
+//
+//   - "agent"        — main agent job (always present)
+//   - "detection"    — threat-detection job (only when threat detection is enabled)
+//   - "safe_outputs" — consolidated safe-outputs job (always present in safe-outputs context)
+//   - "upload_assets"— upload-assets job (only when upload-asset is configured)
+//   - "unlock"       — unlock job (only when lock-for-agent is enabled)
+//   - other custom safe-job names (normalised to underscore format)
+//
+// Additionally, cycles between custom safe-jobs are detected and reported as errors.
+func validateSafeJobNeeds(data *WorkflowData) error {
+	if data.SafeOutputs == nil || len(data.SafeOutputs.Jobs) == 0 {
+		return nil
+	}
+
+	safeJobsNeedsValidationLog.Printf("Validating needs: declarations for %d safe-jobs", len(data.SafeOutputs.Jobs))
+
+	validIDs := computeValidSafeJobNeeds(data)
+
+	for originalName, jobConfig := range data.SafeOutputs.Jobs {
+		if jobConfig == nil || len(jobConfig.Needs) == 0 {
+			continue
+		}
+
+		normalizedJobName := stringutil.NormalizeSafeOutputIdentifier(originalName)
+		for _, need := range jobConfig.Needs {
+			normalizedNeed := stringutil.NormalizeSafeOutputIdentifier(need)
+			if !validIDs[normalizedNeed] {
+				return fmt.Errorf(
+					"safe-outputs.jobs.%s: unknown needs target %q\n\nValid dependency targets for custom safe-jobs are:\n%s\n\n"+
+						"Custom safe-jobs cannot depend on workflow control jobs such as 'conclusion' or 'activation'",
+					originalName,
+					need,
+					formatValidNeedsTargets(validIDs),
+				)
+			}
+			// Prevent a job from listing itself as a dependency
+			if normalizedNeed == normalizedJobName {
+				return fmt.Errorf(
+					"safe-outputs.jobs.%s: a job cannot depend on itself in needs",
+					originalName,
+				)
+			}
+		}
+	}
+
+	// Detect cycles between custom safe-jobs
+	if err := detectSafeJobCycles(data.SafeOutputs.Jobs); err != nil {
+		return err
+	}
+
+	safeJobsNeedsValidationLog.Print("safe-job needs: validation passed")
+	return nil
+}
+
+// computeValidSafeJobNeeds returns the set of job IDs that custom safe-jobs are
+// allowed to depend on, based on the workflow configuration.
+func computeValidSafeJobNeeds(data *WorkflowData) map[string]bool {
+	valid := map[string]bool{
+		string(constants.AgentJobName): true, // agent is always present
+	}
+
+	if data.SafeOutputs == nil {
+		return valid
+	}
+
+	// safe_outputs job is always present when safe-outputs is configured
+	valid[string(constants.SafeOutputsJobName)] = true
+
+	// detection job exists when threat detection is enabled
+	if IsDetectionJobEnabled(data.SafeOutputs) {
+		valid[string(constants.DetectionJobName)] = true
+	}
+
+	// upload_assets job exists when upload-asset is configured
+	if data.SafeOutputs.UploadAssets != nil {
+		valid[string(constants.UploadAssetsJobName)] = true
+	}
+
+	// unlock job exists when lock-for-agent is enabled
+	if data.LockForAgent {
+		valid[string(constants.UnlockJobName)] = true
+	}
+
+	// other custom safe-job names (normalized) are also valid targets
+	for jobName := range data.SafeOutputs.Jobs {
+		normalized := stringutil.NormalizeSafeOutputIdentifier(jobName)
+		valid[normalized] = true
+	}
+
+	return valid
+}
+
+// formatValidNeedsTargets returns a human-readable, sorted list of valid need targets.
+func formatValidNeedsTargets(validIDs map[string]bool) string {
+	targets := make([]string, 0, len(validIDs))
+	for id := range validIDs {
+		targets = append(targets, "  - "+id)
+	}
+	sort.Strings(targets)
+	return strings.Join(targets, "\n")
+}
+
+// detectSafeJobCycles checks for dependency cycles among custom safe-jobs using DFS.
+func detectSafeJobCycles(jobs map[string]*SafeJobConfig) error {
+	if len(jobs) == 0 {
+		return nil
+	}
+
+	// Build normalized name mapping
+	normalized := make(map[string]*SafeJobConfig, len(jobs))
+	originalNames := make(map[string]string, len(jobs))
+	for name, cfg := range jobs {
+		n := stringutil.NormalizeSafeOutputIdentifier(name)
+		normalized[n] = cfg
+		originalNames[n] = name
+	}
+
+	const (
+		unvisited = 0
+		visiting  = 1
+		visited   = 2
+	)
+	state := make(map[string]int, len(normalized))
+
+	var dfs func(node string, path []string) error
+	dfs = func(node string, path []string) error {
+		if state[node] == visited {
+			return nil
+		}
+		if state[node] == visiting {
+			// Build the cycle description using original names where available
+			cycleNodes := make([]string, 0, len(path)+1)
+			for _, p := range path {
+				if orig, ok := originalNames[p]; ok {
+					cycleNodes = append(cycleNodes, orig)
+				} else {
+					cycleNodes = append(cycleNodes, p)
+				}
+			}
+			origNode := node
+			if orig, ok := originalNames[node]; ok {
+				origNode = orig
+			}
+			cycleNodes = append(cycleNodes, origNode)
+			return fmt.Errorf(
+				"safe-outputs.jobs: dependency cycle detected: %s",
+				strings.Join(cycleNodes, " → "),
+			)
+		}
+
+		state[node] = visiting
+		cfg, exists := normalized[node]
+		if exists && cfg != nil {
+			for _, dep := range cfg.Needs {
+				depNorm := stringutil.NormalizeSafeOutputIdentifier(dep)
+				// Only recurse into other custom safe-jobs; skip generated jobs
+				if _, isSafeJob := normalized[depNorm]; isSafeJob {
+					if err := dfs(depNorm, append(path, node)); err != nil {
+						return err
+					}
+				}
+			}
+		}
+		state[node] = visited
+		return nil
+	}
+
+	for node := range normalized {
+		if err := dfs(node, nil); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}