[deep-report] Fix MCP tool-call status=unknown telemetry capture in workflow audit #34055

Context

Agentic Workflow Audit #33873 (2026-05-21 24h) notes:

All tool calls show status=unknown in summary (telemetry capture issue).

This affects the audit-workflows MCP-tool table: 92 safeoutputs calls, 29 github calls, 19 agenticworkflows calls, 10 serena, 8 sentry, 6 mcpscripts, 1 tavily — none have a real status. Downstream agents using these aggregates (cost-per-tool, error-rate-per-tool, retry analytics) lose all signal.

Likely root cause

The MCP tool-call status field in pkg/cli/logs_models.go (or sibling parser) is being populated by a path that does not see the real per-call return code. Possibilities:

1. Parser reads status from a JSON path that no longer exists in current gateway log shape.
2. The gateway emits the status under a renamed key, drifted post-OTel-collector migration.
3. The aggregation step strips the field before serialization.

Proposed Quick Win

1. Locate the parser that aggregates per-call MCP stats (likely in pkg/cli/logs_*.go).
2. Print one raw call record from a recent run to identify where status actually lives.
3. Patch the field path.
4. Re-run audit-workflows and verify the per-tool table shows success/error counts.

Suggested Agent

Copilot coding agent or Sergo.

Estimated Effort

< 2 hours of agent time.

Acceptance

• Next audit-workflows discussion shows real status values per MCP server
• Tool-success ratios surface for the 24h period

Referenced from DeepReport Intelligence Briefing — 2026-05-22.

Generated by 🔬 DeepReport - Intelligence Gathering Agent · ● 18.7M · ◷

• expires on May 24, 2026, 3:50 PM UTC