[deep-report] Add prompt-writing guidance for Copilot-assigned issues (concise, concrete prompts merge more per 6-week trend data) #53253

Description

The recurring "Copilot PR Prompt Pattern Analysis" report has now shown the same pattern across multiple cycles (discussion #53203, Copilot PR Prompt Analysis - 2026-08-16, consistent with its own historical trend table back to 2026-07-01): merged PRs average 151 words per task prompt and reference concrete files/subsystems (pkg, engine, specific package names), while closed/unsuccessful PRs average 229 words — roughly 50% longer — and skew toward open-ended, exploratory language without a concrete acceptance criterion (e.g. "Thanks for asking me to work on this. I will get started on it and keep this PR's description up to date as I form a plan..."). The report explicitly recommends flagging "very long, exploratory prompts (>200 words) without a concrete acceptance criterion" as a correlate of lower merge rates, and separately notes automated CVE/dependency-tracker PRs should be excluded from this signal since their close rate reflects triage policy, not prompt quality.

Expected Impact

A short, concrete guidance section (in CONTRIBUTING.md or wherever Copilot-assignable issues are documented) on writing task descriptions for Copilot SWE-agent-assigned issues — concise, file/subsystem-specific, with an explicit acceptance criterion — should measurably improve merge rate for agent-assigned work, based on data already collected across ~6 weeks of this recurring report.

Suggested Agent

New Agent, or whichever agent maintains contributor-facing docs.

Estimated Effort

Quick (< 1 hour) — a short doc addition, not a code change.

Data Source

DeepReport Intelligence analysis, 2026-08-17 cycle (discussion #53203, cross-referenced against its historical trend table 2026-07-01 → 2026-08-16).

Generated by 🔬 Deep Report · agent · 239.4 AIC · ⌖ 7.81 AIC · ⊞ 11.8K · ◷

• expires on Aug 18, 2026, 4:42 PM UTC-08:00