v0.2.0 --- initial release with 2.0 design

What's Changed

• fix: correct factory ceo examples to use positional prompt by @akashgit in #96
• fix: use descriptive experiment branch names by @shivchander in #97
• ci: add Codecov coverage reporting by @akashgit in #98
• fix: add CODECOV_TOKEN to coverage upload by @akashgit in #99
• fix: drop OBSIDIAN_VAULT_PATH fallback to prevent personal vault writes by @akashgit in #100
• ci: run CI on push to main for Codecov baseline by @akashgit in #101
• feat: experiment-level checkpoint/resume for CEO agent by @shivchander in #102
• fix: strip calendar-time estimates from agent pipeline by @akashgit in #104
• feat: unified backlog for improve-mode convergence by @akashgit in #105
• docs: update docs and README for unified backlog by @akashgit in #106
• feat: redefine --focus as targeted mode by @akashgit in #107
• docs: update docs and diagrams for targeted mode (--focus) by @akashgit in #108
• docs: fix --prompt usage across README and docs by @akashgit in #111
• feat: add Distiller agent and --interactive ideation mode by @akashgit in #110
• docs: restructure README and docs around Build / Improve / Focus workflows by @akashgit in #112
• feat: add session summary report at end of factory cycles by @akashgit in #114
• Fix factory_effectiveness scoring — wire finalize scores + auto-discovery by @akashgit in #118
• Fix research_grounding cold-start — citation backfill + extraction by @akashgit in #120
• fix: CEO agent must not auto-merge PRs by @akashgit in #121
• fix: interactive mode treats idea string as file path by @akashgit in #125
• docs: add meta mode cadence guidance by @Maxusmusti in #116
• feat: add Claude Code plugin/marketplace with /factory:implement by @xukai92 in #117
• feat: add Bob Shell runner support as alternative CLI backend by @colehurwitz in #113
• release: v0.2.0 by @akashgit in #130
• Add operational hypothesis support to Strategist and CEO prompts by @akashgit in #145
• fix: CEO must propagate and enforce operational execution steps by @akashgit in #148
• fix: add backlog item verification gates to CEO prompt by @akashgit in #149
• fix: move backlog verification before finalize to prevent stale metadata by @akashgit in #151
• feat: add research mode models, config parsing, and CLI wiring by @akashgit in #153
• feat: add research run infrastructure — runner, parser, store by @akashgit in #155
• feat: add research mode context to Strategist prompt by @akashgit in #157
• feat: CEO prompt — research mode cycle protocol by @akashgit in #159
• feat: add Researcher step (R1.5) to research mode cycle by @akashgit in #160
• feat: ground truth leakage guards for research mode by @akashgit in #161
• feat: research ideation mode for --mode research by @akashgit in #162
• feat: raise per-cycle limit to 8, remove daily limit by @colehurwitz in #136
• feat: add --no-github flag to disable GitHub operations by @lukeinglis in #163
• Fix Live Events panel expanding without bounds by @lukeinglis in #169
• docs: lifecycle narrative, research mode, updated diagrams by @akashgit in #172
• docs: revamp README with workflow walkthroughs and SVG diagrams by @akashgit in #174
• feat: overhaul ACE data pipeline — performance reports, registry, vault removal by @akashgit in #176
• docs: update for vault→archive migration by @akashgit in #177
• docs: add 'Built with the Factory' showcase section by @akashgit in #178
• feat: add real-time factory visualizer to web dashboard by @abhi1092 in #180
• Fix research_grounding eval to use .factory/archive/ as knowledge source by @abhi1092 in #182
• fix: make heartbeat loop interruptible by Ctrl+C/SIGTERM during sleep by @RohanAwhad in #193
• feat: expose factory agents as plugin subagents by @xukai92 in #139
• feat: finalize gate + hard constraints by @RobotSail in #195
• fix: force-add agents/ in plugin workflow to bypass .gitignore by @xukai92 in #197
• feat: add pipeline skills for dynamic multi-agent orchestration by @xukai92 in #198
• feat: sprint-based resume with scrum master agent by @shivchander in #137
• feat: user-to-CEO message channel by @RobotSail in #194
• feat: close plugin-vs-CLI gaps (#199) by @xukai92 in #201
• feat: expand --focus to accept issue refs (closes #202) by @akashgit in #213
• feat: use --append-system-prompt for headless, strengthen CEO identity by @akashgit in #221
• Add --force flag to factory finalize to bypass precheck gate by @akashgit in #232
• feat: add --dir parameter to factory ceo for explicit project directory naming by @johncohn in #228
• feat: smart project directory naming from descriptions by @akashgit in #233
• docs: add SDG Hub to Built with Factory showcase by @shivchander in #214
• docs: add GSD comparison section to README by @gx-ai-architect in #203
• docs: update all docs for scrum master agent and sprint-based resume by @shivchander in #204
• feat: consolidate factory configuration into config.toml by @akashgit in #241
• Fold Scrum Master into CEO prompt (exp #59) by @akashgit in #243
• feat: review-until-clean pipeline loop and draft PR flow (#240) by @akashgit in #255
• feat: support --mode interactive on existing projects by @akashgit in #254
• fix: unify on bare 'factory' CLI + fix path resolution (#239) by @akashgit in #252
• Branch isolation via git worktrees (Issue #240 Part 1) by @akashgit in #257
• Enforce mandatory review pipeline — Sacred Rule 9 + prompt hardening by @akashgit in #267
• feat: add factory backfill-archive CLI command by @akashgit in #259
• feat: research mode dashboard view (Issue #264) by @akashgit in #268
• Fix worktree creation: detect default branch instead of hardcoding 'main' by @akashgit in #271
• Fix interactive mode on existing projects — banner, CEO mode, Phase 0e research by @akashgit in #275
• Fix committed .factory symlink and add defensive mkdir helper by @akashgit in #278
• fix: interactive mode crash on file input and empty-repo worktree creation by @mairin in #280
• fix: persist spec for raw-idea interactive mode by @akashgit in #281
• fix: provide git identity in _ensure_repo for CI by @akashgit in #282
• fix: correct HypothesisBudget fields and harden read_config by @akashgit in #290
• docs: document interactive mode file input in README by @akashgit in #284
• README overhaul: restructure around 3 workflows, cut to ~200 lines by @akashgit in #293
• feat: add Codex runner for OpenAI Codex CLI support by @lukeinglis in #291
• Unify Phase 0 and Phase 0e — full ideation flow for existing projects by @akashgit in #298
• Add eval_spec: testing instructions for Evaluator context injection by @akashgit in #303
• feat: redesign welcome wizard as conversational agent by @akashgit in #305
• Welcome Wizard — interactive CLI entry point by @akashgit in #301
• Add OpenSkies Airline Corpus to Built with the Factory by @lukeinglis in #295
• Fix wizard ENAMETOOLONG crash and add explicit --mode flags by @akashgit in #311
• feat: promote eval_spec to scored dimensions + within-tier weight config by @akashgit in #315
• CEO: add progress tracking task list to every cycle by @akashgit in #322
• CEO identity drift defense — re-anchor at agent transitions by @akashgit in #318
• Fix inner/outer loop semantics and add validated interactive extraction by @akashgit in #328
• Implement inner/outer loop Python infrastructure by @akashgit in #336
• Fix inner/outer loop semantics and add validated interactive extraction (issue #326) by @akashgit in #338
• Implement inner/outer loop Python infrastructure (issue #333) by @akashgit in #339
• Fix review pipeline counter starvation: separate 2d-review and 2h-final counters by @akashgit in #331
• Enforce spec depth in Distiller — grounding protocol + quantified constraints by @akashgit in #341
• Wizard: redirect long input to file before classification by @akashgit in #334
• Add session naming to factory-spawned Claude Code sessions by @akashgit in #337
• feat: LLM-driven user profiling system by @akashgit in #343
• Add Refinement Stage — Refiner agent + --refine CLI flag + CEO Mode: Refine by @akashgit in #319
• Add explicit inner/outer loop controls to research mode (issue #313) by @akashgit in #323
• Fix inner/outer loop field name mismatch in Distiller/CEO prompts by @akashgit in #353
• docs: document inner/outer loop config for research mode by @akashgit in #355
• fix: derive smart session names from user intent by @akashgit in #348
• docs: document post-cycle refinement loop and Refiner agent by @akashgit in #357
• Clean PR mode: strip non-essential artifacts from PRs by @akashgit in #363
• docs: document Clean PR Mode by @akashgit in #371
• fix: defer project directory creation until just before worktree setup by @akashgit in #361
• docs: rebrand to re:factory and add logo by @colehurwitz in #377
• chore: update README.md to remove double re:factory title by @colehurwitz in #389
• feat: Codex CLI plugin support (manifest, agent generation, skills) by @lukeinglis in #395
• fix: stop treating the factory's own 'implementation' label as an unbuilt-repo signal (#378, #382) by @colehurwitz in #383
• Bob runner: strip ANSI escapes on live-terminal write to stop flicker by @colehurwitz in #381
• feat: token usage auditing and profiling across factory workflow by @akashgit in #412
• fix: increase timeout durations for eval and agent invocations by @akashgit in #415
• fix: wizard _quick_classify bypasses LLM for long inputs by @akashgit in #410
• fix: use sys.executable instead of bare 'python' in hygiene eval by @akashgit in #419
• feat: tmux persist mode for CLI runners by @xukai92 in #285
• fix: write agent prompts to temp files to avoid ARG_MAX limit by @Maxusmusti in #422
• ci: rename plugin branch to plugins and document plugin install by @xukai92 in #397
• feat: add --mode review --pr N for on-demand CEO PR review by @xukai92 in #421
• ci: debug ceo-review workflow for CI testing by @xukai92 in #434
• chore: test ceo-review workflow by @xukai92 in #435
• ci: auto-approve PR when CEO review verdict is KEEP by @xukai92 in #438
• fix: address PR #443 review — interactive mode + Codex auth by @akashgit in #468
• fix: ceo-review falls back to PR comment when formal review fails by @akashgit in #469
• feat: runner abstraction v2 — universal multi-runner factory by @akashgit in #443
• fix: auto-approve checks formal reviews, not stale comments by @akashgit in #477
• fix: skip test_available_runners_detected in CI by @akashgit in #481
• ci: allow Cole and Shiv to trigger @ceo-review by @akashgit in #482
• ci: allow osilkin98 and gx-ai-architect to trigger @ceo-review by @akashgit in #495
• fix: detect OpenCode npm vs Go binary incompatibility at runtime by @akashgit in #502
• fix: Bob RunnerMeta.check_auth() supports file-based auth by @akashgit in #503
• fix: replace tmux wait-for with Stop-hook sentinel for completion signaling by @shivchander in #500
• fix: address 6 confirmed review issues on PR #514 by @colehurwitz in #520
• feat: parallelize researcher spawning with --review-tag (fixes #524) by @akashgit in #528
• feat: replace wall-clock timeout with inactivity-based timeout (issue #525) by @akashgit in #531
• feat: merge Distiller into Strategist — dual-mode agent by @akashgit in #530
• fix: align Strategist ideation output with phased build-plan format by @akashgit in #534
• Fix 6 CI test failures: review_tag mocks, prompt assertion updates by @colehurwitz in #541
• fix: eliminate eval duplication for 4x speedup (#536) by @colehurwitz in #538
• Implement 6 stubbed eval dimensions + 3 combined methods for Go/Node/Rust by @colehurwitz in #548
• fix: eval language evaluator gaps — silent failures and scoring blindness by @colehurwitz in #551
• refactor: interactive_run() reuses build_command() across all runners by @gx-ai-architect in #552
• fix: make test_timeout configurable to handle long test suites by @colehurwitz in #559
• feat: enrich factory lifecycle events for external consumers by @colehurwitz in #557
• refactor: rename interactive mode to design by @xukai92 in #563
• docs: add abstract meta-harness specification by @xukai92 in #423
• Add Langfuse tracing for re-factory by @gx-ai-architect in #577
• Session persistence: capture and view CEO + specialist agent conversations by @gx-ai-architect in #569
• feat: add benchmark CI pipeline for pre-release validation by @RobotSail in #590
• feat: add PR comments and release notes to benchmark workflow by @RobotSail in #592
• fix: benchmark workflow skips all jobs on PR trigger by @RobotSail in #594
• fix: wire up pr_number input for workflow_dispatch by @RobotSail in #598
• fix: add missing pr_number input to workflow_dispatch by @RobotSail in #599
• fix: disable benchmark CI from running on every PR by @RobotSail in #601
• feat: add ProgramBench support to benchmark CI pipeline by @RobotSail in #600
• feat: --bg flag for background agent dispatch (rework of PR #424) by @xukai92 in #571
• Fix 4 mypy errors causing CI lint failure on main by @xukai92 in #605
• feat: run factory CEO as TerminalBench solver + fix GCP auth for CI by @RobotSail in #603
• feat: run factory CEO as solver for SWE-bench and FeatureBench by @RobotSail in #609
• fix: install factory CLI in benchmark GHA workflow by @RobotSail in #611
• feat: run factory CEO as ProgramBench solver by @RobotSail in #610
• fix: use git fsck instead of reflog for worktree recovery by @RobotSail in #612
• Unify Design Mode and Build Mode into a single Plan Loop by @akashgit in #615
• fix: make GCP credentials readable by agent user in ProgramBench container by @RobotSail in #617
• fix: replace :ro bind mount with docker cp for GCP credentials in ProgramBench by @RobotSail in #620
• feat: add --solver flag to benchmark CI for factory vs claude-code comparison by @RobotSail in #624
• fix: pass --extra-instruction-path for claude-code solver in TerminalBench by @RobotSail in #629
• feat: add cost and token tracking to benchmark results by @RobotSail in #635
• fix: read correct cost field name from Claude Code JSON output by @RobotSail in #638
• feat: add benchmark result history tracking and baseline comparison by @RobotSail in #639
• fix: YAML syntax error in benchmark workflow (inline Python breaking parser) by @RobotSail in #640
• Remove dead smoke_test_command parameter from run_precheck by @akashgit in #633
• feat: add benchmark dashboard and remove broken history commit by @RobotSail in #647
• Archivist async redesign: structured output, haiku model, reduced invocations by @akashgit in #636
• feat: benchmark data branch storage and dashboard charts by @RobotSail in #648
• fix: benchmark-data branch creation in report job by @RobotSail in #650
• fix: debug and fix benchmark data append to benchmark-data branch by @RobotSail in #652
• fix: move artifact download after checkout so results/ survives by @RobotSail in #654
• feat: dashboard UI overhaul and nightly scheduled benchmark runs by @RobotSail in #655
• fix: dashboard data pipeline — solver field, cost extraction, chart labels by @RobotSail in #658
• revamp: benchmark dashboard with main-branch filtering, PR comparison, full-width charts by @RobotSail in #661
• revamp: benchmark dashboard layout — averages first, grouped history by @RobotSail in #663
• fix: replace average duration/cost bar charts with line charts over time by @RobotSail in #664
• revamp: complete rewrite of benchmark dashboard (v4 spec) by @RobotSail in #667
• feat: add methodology section to benchmark PR comments by @RobotSail in #668
• feat: add non-blocking benchmark status check to PRs by @RobotSail in #669
• fix: benchmark-eval check only passes when factory jobs ran by @RobotSail in #671
• fix: enhance PR comment with dimension-level comparison vs main baseline by @RobotSail in #672
• fix: benchmark-eval status check requires all 4 factory benchmarks by @RobotSail in #673
• feat: increase default benchmark timeout to 7200s by @RobotSail in #674
• fix: move worktrees to .factory-worktrees/ to eliminate ELOOP by @akashgit in #676
• feat: SPEC.md-aware Plan Loop — Strategist reads SPEC.md, plan grounded in spec diff by @xukai92 in #622
• fix: benchmark CI — terminalbench resolved:false, $0 costs, nightly schedule by @RobotSail in #679
• fix: remove stale factory eval patterns from CEO prompt, add QA delegation tests by @akashgit in #683
• fix: enable claude-code solver in nightly benchmark schedule by @RobotSail in #678
• fix: ProgramBench resolved threshold requires 100% test pass rate by @RobotSail in #687
• fix: create project directory before git worktree prune by @RobotSail in #688
• Integrate Sentrux architectural quality check into CI by @xukai92 in #660
• ci: release benchmark notes + on-merge PR testing by @RobotSail in #695
• fix: design mode shows 'Mode: design' instead of 'Mode: build' in CEO task string by @xukai92 in #692
• perf: add pagination to benchmark dashboard run history by @RobotSail in #719
• fix: streaming CEO tracing + trace name corruption by @gx-ai-architect in #722
• Standardize review mode to reuse Improve pipeline QA steps by @xukai92 in #701
• fix: switch Claude runner to stream-json output to prevent watchdog kills by @gx-ai-architect in #725
• fix: FeatureBench harness — improve --focus with pre-seeded factory state by @RobotSail in #734
• feat: add architecture hygiene dimension via Sentrux integration by @xukai92 in #739
• feat: CI eval baseline pipeline + factory baseline CLI command by @RobotSail in #740
• fix: increase asyncio StreamReader limit to 1MB by @RobotSail in #742
• Feat/workflow graph engine by @xukai92 in #744

New Contributors

• @Maxusmusti made their first contribution in #116
• @xukai92 made their first contribution in #117
• @colehurwitz made their first contribution in #113
• @lukeinglis made their first contribution in #163
• @abhi1092 made their first contribution in #180
• @RohanAwhad made their first contribution in #193
• @johncohn made their first contribution in #228
• @gx-ai-architect made their first contribution in #203
• @mairin made their first contribution in #280

Full Changelog: https://github.com/akashgit/remote-factory/commits/v0.2.0

Benchmark Results

Benchmark	Solver	Instance	Result	Score	Cost	Duration
swebench	factory	sympy__sympy-20590	✅	1	$3.65	539s
terminalbench	factory	fix-git	❌	0	$0.00	1005s
featurebench	factory	pypa__packaging.013f3b03.test_metadata.e00b5801.lv1	❌	0	$5.43	1074s
programbench	factory	abishekvashok__cmatrix.5c082c6	❌	0	$9.93	1852s

📊 View benchmark dashboard