Optimize workflow prompts: reduce token usage by 70% in shared files

[Copilot]

Token usage spiked 4x to 1.71M across 10 workflow runs. Root cause: verbose prompts with redundant examples, lengthy bash scripts, and repetitive instructions.

Changes

Shared files (high-impact, imported by 30-45 workflows each):

• trending-charts-simple.md: 313 → 78 lines (75% ↓)
  • Removed 3 complete Python example patterns
  • Condensed to essential snippets only
• reporting.md: 80 → 14 lines (82% ↓)
  • Simplified formatting guidelines
  • Removed redundant examples

High-volume workflows (daily/per-event):

• audit-workflows.md: 374 → 163 lines (56% ↓)
• daily-code-metrics.md: 563 → 93 lines (83% ↓)
• dev-hawk.md: 143 → 98 lines (31% ↓)

All 121 workflow lock files recompiled: -5,316 lines net across affected sections.

Token impact

Shared file multiplier effect:

• trending-charts-simple.md: 235 lines × 32 imports = 7,520 lines eliminated
• reporting.md: 66 lines × 45 imports = 2,970 lines eliminated

Estimated savings: ~42K tokens/run from shared files alone, ~450K tokens across 10 runs.

Before/After example (trending-charts-simple.md):

- ## Quick Start Pattern 1: Daily Metrics Tracking
- 
- Track daily metrics and visualize trends over time:
- 
- ```python
- #!/usr/bin/env python3
- """Daily metrics trending"""
- import pandas as pd
- [... 60 lines of example code ...]
- ```
- 
- ## Quick Start Pattern 2: Moving Averages
- [... another 40 lines ...]
- 
- ## Quick Start Pattern 3: Comparative Trends
- [... another 40 lines ...]

+ ## Store Historical Data (JSON Lines)
+ 
+ ```python
+ import json
+ from datetime import datetime
+ 
+ with open('/tmp/gh-aw/cache-memory/trending/<metric>/history.jsonl', 'a') as f:
+     f.write(json.dumps({"timestamp": datetime.now().isoformat(), "value": 42}) + '\n')
+ ```

Projected results

From 1.71M baseline → ~1.26M tokens (26% reduction). Target: <850K (50%). Additional optimizations available if monitoring shows gap remains: output summarization, MCP server instruction files (slack.md: 219 lines, drain3.md: 196 lines), remaining high-line-count workflows.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>[plan] Optimize prompt token usage in workflow runs</issue_title>
<issue_description>## Objective

Reduce token consumption in workflow runs by trimming verbose prompts and enabling summarization where appropriate.

Context

Token usage spiked ~4x to 1.71M tokens across the last 10 runs, suggesting inefficient prompt construction or unnecessarily verbose outputs.

Approach

1. Analyze the last 10 workflow runs to identify top token consumers:
   • Use gh aw logs or audit commands to extract token usage per step
   • Identify prompts/steps with highest token counts
2. Review prompt construction in high-cost steps:
   • Look for redundant context or instructions
   • Check for unnecessary verbosity in system prompts
   • Identify opportunities for summarization
3. Implement optimizations:
   • Trim verbose instructions where possible
   • Enable output summarization for intermediate steps
   • Remove redundant context from prompts
4. Validate changes don't reduce quality

Files to Modify

• .github/workflows/*.md - Workflow source files with prompts
• Any agent configuration files affecting prompt construction

Acceptance Criteria

• Token usage reduced by at least 50% (target: <850k for 10 runs)
• Output quality maintained (no regression in agent effectiveness)
• Top 3 token-consuming steps identified and optimized
• Changes documented with before/after token counts

Estimated Effort

Medium (3-5 hours)
Related to #6857

AI generated by Plan Command for discussion #6855

Comments on the Issue (you are @copilot in this section)

• Fixes [plan] Optimize prompt token usage in workflow runs #6859

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.