Skip to content

DISTRIBUTION_CHECKLIST.md

Latest commit

 

History

History
50 lines (35 loc) · 1.69 KB

DISTRIBUTION_CHECKLIST.md

File metadata and controls

50 lines (35 loc) · 1.69 KB
Raw

Distribution checklist

This is the local launch checklist for DM-Code-Agent v2. It is a plan, not a record of completed external posting.

GitHub release

  • Tag v2.0.0 after CI passes.
  • Release title: DM-Code-Agent v2: auditable local code agent with Reflexion and critic review.
  • Attach links to:
    • README.md
    • docs/research-log/06-final-writeup.md
    • bench_reports/swebench_lite_baseline.md
    • bench_reports/economics.md
  • State clearly that the SWE-bench Lite number is Tier-1 host verifier, not leaderboard comparable.

Blog / long-form post

  • Publish the final write-up from 06-final-writeup.md.
  • Keep the headline focused on the engineering artifact, not on unverified score improvements.
  • Include the frozen baseline caveat near the first SWE-bench mention.

Short posts

  • X / Twitter: one thread with architecture, trace screenshot/GIF, and the frozen baseline note.
  • Jike / Weibo: short Chinese summary with the research-log index.
  • Reddit / Hacker News: post only after a clean install smoke test from a fresh clone.

Awesome lists

  • awesome-llm
  • awesome-ai-agents
  • awesome-mcp

Suggested description:

Local-first Python code agent with JSONL trace/replay, MCP tools, hidden-test benchmarks, SWE-bench Lite Tier-1 harness, and default-off Reflexion/Critic/Self-Consistency modules.

Demo asset

The intended 90-second demo flow:

  1. Fresh checkout and pip install -e ".[dev]".
  2. dm-agent-bench --suite maintenance --list.
  3. Show one deterministic eval or benchmark manifest.
  4. Show trace replay.
  5. Show dm-agent-economics regenerating bench_reports/economics.md.

Do not present a new SWE-bench score unless a permitted real evaluation has been run.