Audit log, metrics, and slow-query diagnostic boundaries

[Widthdom]

Context

Dogfooding the locally built cdidx surfaced enough logging and metrics touchpoints to justify a focused audit. These paths can affect support-safe diagnostics, local privacy, retention behavior, and performance investigations.

Representative search inventory:

dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search CDIDX_LOG --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search GlobalToolLog --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search CDIDX_METRICS --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search SlowQuery --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search LogRetention --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search LogRotation --format count --limit 100

Observed baseline:

• CDIDX_LOG: 12 hits.
• GlobalToolLog: 87 hits.
• CDIDX_METRICS: 21 hits.
• SlowQuery: 12 hits.
• LogRetention: 0 hits.
• LogRotation: 0 hits.

Audit questions

• Are log paths, file modes, and fallback paths consistent across global-tool, workspace, and test execution?
• Do logs and metrics avoid leaking secrets, raw environment values, full user paths, request payloads, or plugin/hook diagnostics beyond support-safe bounds?
• Is retention/rotation behavior explicit enough for long-running MCP, watch, and repeated CLI use?
• Are slow-query diagnostics bounded, redacted, and actionable without requiring users to share a full local DB or verbose environment dump?

Acceptance criteria

• Logging and metrics outputs have clear redaction and support-safety boundaries.
• Long-running or repeated invocations do not grow logs unbounded without documented cleanup guidance.
• Slow-query diagnostics are useful for performance triage while avoiding sensitive local data exposure.
• Tests cover representative CLI and MCP-adjacent logging paths where feasible.

[Widthdom]

Additional dogfood findings: profile and slow-query controls

A later pass found more profiling and slow-query diagnostics adjacent to the logging/metrics audit.

Representative inventory:

dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search profile --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search slow-query-ms --format count --limit 100
dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll search --exact-substring --query slow-query-ms --format count --limit 100

Observed additions:

• profile: 145 hits across 46 files.
• slow-query-ms: 23 hits across 15 files and emitted a punctuation-heavy-query hint recommending --exact-substring.
• exact-substring slow-query-ms: 19 hits across 12 files.

Follow-up angle: profiling and slow-query paths should be audited with the same support-safety lens as logs and metrics. In particular, SQL/profile details, file paths, query text, and timing diagnostics need bounded output and redaction rules that work in CLI and MCP contexts.

[Widthdom]

Additional dogfood coverage from the 2026-07-07 terminal/end pass.

Diagnostics/profile/logging surface:

• search --query "CommandErrorWriter" --format count --json --metrics /private/tmp/.../metrics-round3.jsonl wrote a JSONL metrics record with timestamp, tool, source, elapsed_ms, and exit_code.
• search --query "CommandErrorWriter" --format count --json --profile --slow-query-ms 0 emitted the normal count result followed by a large profile payload containing phase timings, row counts, query plan, and full SQL text.

Related code-search counts:

• --metrics: 107 matches across 31 files.
• --profile: 165 / 46.
• --slow-query-ms: 23 / 15.
• log-format: 23 / 13.
• log-retain: 20 / 11.
• log-max-size: 30 / 12.
• debug-unsafe: 26 / 12.

Follow-up angle: profile output is very useful for troubleshooting, but raw SQL text can dominate JSON output. A compact/redacted profile mode, or a separate profile size bound, would make it easier to attach diagnostics to issues.

[Widthdom]

Work started on branch fix-issue4313-4319-4320-4332-4342-4359.