[agent-performance] Agent Quality Plateau: Q=61/E=62 for 3+ Weeks — Prompt Improvement Initiative

[github-actions]

Problem

Agent quality (Q) and effectiveness (E) scores have been plateaued at 61/100 and 62/100 for 3+ consecutive weeks. This stagnation indicates that bug-fix-only interventions (fixing engine crashes, missing tools, etc.) are insufficient — the underlying prompt designs need improvement.

Evidence

• Jul 4: Q=61, E=62 (→ unchanged)
• Jul 3: Q=61, E=62 (→ unchanged)
• Week of Jun 23–30: Q=61, E=61 (→ stable plateau)
• WHM health: 69/100 (↓3 today), independently declining

Shared context note from shared-alerts.md:

"Q/E plateau at 61/62 for 3 weeks: need prompt improvements, not just bug fixes"

Root Causes (Hypothesized)

1. Generic task framing: Many agent prompts lack concrete success criteria — agents complete the workflow mechanics but miss the intent.
2. No self-assessment loop: Agents don't evaluate their own output quality before emitting safe outputs.
3. Stale examples: Several prompts reference patterns/tools that have since changed (e.g., Codex alpha, old safe-output signatures).
4. Low actionability: Report-type agents (documentation quality, contribution checks) produce outputs that are rarely acted on.

Recommended Actions

• Audit the bottom 10 agents by Q score and document specific prompt deficiencies
• Add explicit quality rubric instructions ("Before filing, verify: completeness, accuracy, actionability")
• Update stale examples in prompts (especially Codex-dependent and MCP-dependent agents)
• For recurring failures (Matt Pocock, Impeccable, Design Decision Gate): evaluate deprecation vs. redesign
• Consider a shared quality-gate.md skill that all agents can reference

Impact

Raising average Q/E from 61→70 would meaningfully improve ecosystem health score, reduce wasted action_required runs, and increase PR merge rates.

Tracking

• Agent Performance Analyzer will report on this trend weekly
• Target: Q≥65, E≥65 within 4 weeks of first prompt improvements landing

Generated by ⚡ Agent Performance Analyzer - Meta-Orchestrator · 71.2 AIC · ⌖ 21.3 AIC · ⊞ 10.4K · ◷

• expires on Jul 6, 2026, 5:11 AM UTC-08:00

[github-actions]

🍪 Issue Monster selected this for Copilot

I've identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.

If assignment succeeds, the Copilot coding agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

🍪 Om nom nom by Issue Monster · 1.19 AIC · ⌖ 4.57 AIC · ⊞ 6.6K · ◷