chore(contributing): refresh the PR template and guard intentional issue references #349

Problem and evidence

Read-only review of main f5e7c49554ff02771990a151e00b52f2abf3df0a on 2026-10-29 found that the PR template:

• defaults to an automatic closing reference;
• asks contributors to run the frozen legacy cc_peer.py, rather than the current session-peer CLI;
• names a nonexistent root CLAUDE.md and records only Claude Code/OS versions;
• omits the generated-standalone check and current subsystem validation guidance.

This can give reviewers irrelevant validation evidence and confuse partial work with completion. The latter has occurred twice: maintainer comments on #161 explicitly record accidental closure from negated closing phrases in PR #197 and PR #224. Those PR bodies were corrected and #161 is currently open; the issue is prevention, not reopening it again.

Expected behavior / scope

Refresh contributor guidance to match current CI. Use a non-closing reference by default, with an explicit intentional-completion alternative. Explain that negating an automatic-closing phrase is unsafe. Document relevant Python/Node/agent versions, supported subsystem checks and honest not-run/not-applicable entries; do not require real-host or live-agent actions for every PR.

Evaluate a read-only PR-body warning for ambiguous/negated closing references. It must not execute PR body text or untrusted code, change issue state, or prohibit legitimate intentional closure. A warning is a review aid, not proof of GitHub's complete closing behavior.

Acceptance checks

• Template points to current CLI, existing documentation and python tools/generate_session_peer.py --check.
• Relevant core/adapter/Codex/distribution/docs and optional MCP/Relay/Control checks can be reported with explicit skipped/not-run limits.
• Partial work uses non-closing references; intentional completion is a deliberate choice.
• If a body checker is added, fixture tests cover ordinary references, intentional closure, negated phrases, multiple references and body edits, with minimal permissions and no untrusted execution.
• Template/guidance regression coverage is added where appropriate.

Validation and related work

Source, issue timeline/comments, and open/closed issues/PRs were reviewed. #1/#2 introduced the old conventions; they do not track this refresh. No tests or live messaging were run in this review. No source, settings or existing issue states were changed.