eval-magic v0.3.0

What's Changed

• Merge pull request #28 from slowdini/dev by @slowdini in #29
• chore(docs): document run mode intention by @slowdini in #30
• chore(dx): agent memory files by @slowdini in #46
• feat(cli): init command for fresh evals/eval.json by @slowdini in #47
• refactor(pipeline): enumerate per-condition run slots by @slowdini in #48
• feat(run): N runs per condition cell (--runs + per-eval override) by @slowdini in #49
• feat(cli): simplify and improve param default values by @slowdini in #50
• feat(cli): warn that guard is armed between iterations by @slowdini in #51
• feat(cli): record run provenance in manifest, handle notes.md on promote by @slowdini in #52
• feat(opencode): add foundational harness selection and staging by @slowdini in #53
• fix(cli): sharpen run/ingest diagnostics and next-step commands by @slowdini in #54
• fix(claude): provide correct handling and guidance for claude skill-staging gotcha by @slowdini in #55
• feat(claude): auto-resolve suagents dir by @slowdini in #56
• fix(codex): fix codex run friction by @slowdini in #57
• refactor(adapters): HarnessAdapter trait + run-mode foundation (PR1 of #58) by @slowdini in #59
• refactor(run): organize dispatch + ingest around DispatchMechanism (PR2 of #58) by @slowdini in #60
• docs(parity): reframe parity around run mode; comment sweep (PR3 of #58) by @slowdini in #61

Full Changelog: v0.2.1...v0.3.0