Overview

Architecture improvements for the Smart Output Filtering system (parent epic #426). Addresses gaps discovered during review: command matching ambiguity, no pipeline composition, missing fallback strategy, insufficient configuration, and security risks.

These improvements should be implemented before or during Phase 2 of #426.

Issues

Phase 1a: Core Architecture (P0, blocks #428)

• feat: standardized CommandMatcher enum for output filters #439 — Standardized CommandMatcher enum
• feat: FilterConfidence enum for output filter results #440 — FilterConfidence enum for filter results
• feat: move filter metrics from Phase 3 to Phase 1 #442 — Move filter metrics to Phase 1
• feat: security pattern whitelist for output filters #443 — Security pattern whitelist

Phase 1b: Configuration (P1, parallel with #428-#430)

• feat: enhanced per-filter TOML configuration with command overrides #444 — Enhanced per-filter TOML configuration with command overrides

Phase 2: Pipeline Composition (P1, before #431-#432)

• feat: composable FilterPipeline for chained output filters #441 — Composable FilterPipeline for chained filters

Dependency Graph

#439 CommandMatcher ──┐
#440 FilterConfidence ├──► #442 Metrics ──► #428 Test filter
#443 Security ────────┘                 ──► #429 Git filter
                                        ──► #430 Clippy filter
#444 Config (parallel)

#441 Pipeline ──► #431, #432 (Phase 2 filters)

Impact

• ~90 cfg annotations removed (M26), now focusing on filter quality
• Estimated ~50+ cfg annotations worth of filter trait improvements
• Validates claimed 60-90% token savings via early metrics

References

• Parent epic: Epic: smart tool output filtering for token savings (M26) #426
• Plan: .local/plan/m26.1-output-filtering-improvements.md
• Related issues: feat: command-aware output filter framework in ShellExecutor #427-feat: token savings tracking and metrics #434