Cost-Per-Issue Attribution Dashboard with Optimization Recommendations

[github-actions]

Strategic Improvement

Cost intelligence exists but lacks issue-granular tracking. Build a system that attributes token usage and cost to individual issues, stages, and repos, then surfaces optimization opportunities like "refactor issues cost 3x more than bugfix issues" or "review stage costs $2/issue, consider Haiku routing". Enables data-driven cost optimization decisions and feeds the model router (#382).

Acceptance Criteria

• Extend sw-cost.sh to track cost per issue ID, per pipeline stage, per repo
• Store in ~/.shipwright/cost-attribution.json with issue/stage/repo/model breakdown
• Dashboard command (shipwright cost breakdown) showing cost by dimension
• Identify cost outliers: issues or stages >2 standard deviations above mean
• Automated recommendations: suggest model routing changes for expensive stages
• Historical trends: cost per issue over last 30 days with trend line
• Integration with budget system: alert when issue type approaches budget limit

Context

• Priority: P3
• Complexity: standard
• Generated by: Strategic Intelligence Agent
• Strategy alignment: P3: Cost Efficiency & P4: Observability

[github-actions]

⚓ Shipwright Pipeline Starting

Field	Value
Template	standard
Complexity	standard
Max Iterations	20
Triage	Extends cost intelligence with issue-granular attribution and automated recommendations; builds on existing cost/dashboard infrastructure with moderate statistical analysis and rule-generation work.
Runner	ubuntu-latest
Risk Level	unknown
Workflow	24282380172

The autonomous pipeline is processing this issue. Updates will be posted as stages complete.

[github-actions]

[github-actions]

[github-actions]

[github-actions]

📋 Implementation Plan

Click to expand full plan (24 lines)

The implementation plan has been written to .claude/plan.md. Here's a summary of what was analyzed and planned:

Key Finding: The existing cost_record() function already captures issue/stage/model data — the ingestion pipeline is complete. What's missing is the aggregation, analysis, and presentation layer.

Approach: Extend sw-cost.sh (2 files modified + 1 new test file) rather than creating a separate script. This minimizes blast radius while keeping the cost domain cohesive.

10 tasks covering:

1. SQLite query functions (db_cost_by_issue, db_cost_attribution_summary)
2. Attribution build function with multi-dimensional aggregation
3. Outlier detection (mean + 2σ) using awk
4. Rule-based recommendation engine (model routing, stage cost, budget proximity)
5. 30-day trend analysis with daily windows
6. Dashboard display (terminal + JSON modes)
7. CLI wiring (shipwright cost breakdown)
8. Budget integration with per-issue-type warnings
9. Comprehensive test suite (~15 tests)
10. Registration and regression verification

Key design decisions:

• No schema changes needed — existing cost_entries table has all required columns
• cost-attribution.json is a derived cache (fully idempotent, self-healing)
• All data operations use jq with --arg escaping (no string interpolation)
• Bash 3.2 compatible (no associative arrays)
• Atomic file writes via tmp + mv pattern

---

Generated by shipwright pipeline at 2026-04-11T12:35:41Z

[github-actions]

[github-actions]

📐 Architecture Decision Record

Click to expand ADR (20 lines)

The Architecture Decision Record has been written to scripts/skills/generated/cost-attribution-analysis.md. Here's a summary of the key decisions:

Approach: Extend sw-cost.sh (+~400 lines) and sw-db.sh (+~60 lines) rather than creating a separate script. This keeps the cost domain cohesive with minimal blast radius (2 files modified + 1 new test file).

5 Components with clean dependency direction (CLI -> Analysis -> Aggregation -> Storage -> Data):

1. Storage Layer (sw-db.sh): db_cost_by_issue(), db_cost_attribution_summary() -- SQLite queries with JSON fallback
2. Aggregation Layer (sw-cost.sh): cost_attribution_build() -- idempotent cache generation to cost-attribution.json
3. Analysis Engine (sw-cost.sh): cost_outlier_detect() (mean + 2 sigma via awk), cost_recommendations() (rule engine), cost_trend_by_issue() (30-day daily windows)
4. CLI Layer: shipwright cost breakdown [--period N] [--json] [--issue N] [--recommendations]
5. Data Sources: Existing costs.json + cost_entries table -- no schema changes needed

Critical design decisions:

• Statistical guards: N >= 5 for outliers, N >= 10 for recommendations (prevents spurious signals)
• awk for stddev (jq lacks sqrt(), awk is more portable than bc)
• Rule-based recommendations over ML (explainable, predictable, works with small datasets)
• Derived cache (cost-attribution.json) is fully idempotent and self-healing on deletion

3 alternatives rejected: separate script (fragments domain), dashboard server integration (wrong blast radius for v1), SQLite-only (breaks backward compat).

---

Generated by shipwright pipeline design stage at 2026-04-11T12:41:37Z

[github-actions]

[github-actions]

🔨 Build started — shipwright loop with 20 max iterations, 1 agent(s), model: opus

[github-actions]

[github-actions]

❌ Pipeline failed at stage build after 12m 18s.

No log available