| flow-next-v1.6.0 | ### Changed - **Tracker-sync is now opt-OUT, not opt-in โ hooking up the bridge activates the whole pipeline by default.** Previously every `tracker.perEvent.*` lifecycle touchpoint defaulted `off`, so after the discovery ceremony you had to opt each event in individually (fn-52 R1). That inverted the intent โ connecting a tracker means you want it kept in sync. Now the `/flow-next:tracker-sync` discovery ceremony, on confirmation, **activates every lifecycle event by default**: capture / inter | High | 6/4/2026 |
| flow-next-v1.3.4 | ### Fixed - **Review-output R-ID parser dropped single-letter suffixes (`R4a` / `R4b`)** โ `parse_unaddressed_rids` extracted R-IDs from a reviewer's `Unaddressed R-IDs:` summary line (`_extract_rids`) and from the `## Requirements coverage` table fallback with bare `\bR(\d+)\b`. fn-49.1 (1.2.1) taught the *spec* acceptance-criteria parser the `R\d+[a-z]?` suffix form but left this *review-output* path behind, so a reviewer reporting `Unaddressed R-IDs: [R4a, R4b]` parsed to drop the suffixed I | High | 5/27/2026 |
| flow-next-v1.1.11 | ### Fixed - **`flowctl init` no longer silently flips pre-1.1.3 users' `planSync.crossEpic` from on to off.** Caught while auditing `/flow-next:setup` against the dev repo (this fix's sibling 1.1.10 PR surfaced the `.flow/config.json` diff). Pre-1.1.3 users have `planSync.crossEpic: true` (the only key) in their on-disk config. 1.1.3 introduced `planSync.crossSpec: false` as the new canonical default, and the 1.1.3 read precedence is "canonical wins on presence". Without a pre-merge mirror, eve | High | 5/22/2026 |
| flow-next-v1.1.1 | ### Fixed - **`install-codex.sh` now copies `templates/spec.md` (and any sibling top-level templates) to `~/.codex/templates/`.** 1.1.0 shipped the canonical spec template at `plugins/flow-next/templates/spec.md` and wired the `/flow-next:interview` skill to read it at runtime via `${CLAUDE_PLUGIN_ROOT}/templates/spec.md`, but the Codex installer only copied skill-scoped templates (ralph-init). Codex users on 1.1.0 hit a missing-file when invoking `/flow-next:interview --scope=business` on a ne | High | 5/16/2026 |
| flow-next-v1.0.2 | ### Marketplace housekeeping - **Legacy `flow` plugin removed.** The original two-step planning + execution plugin (`plugins/flow/`) has been deleted from the marketplace; `flow-next` is now the only plugin shipped here. The legacy plugin had been unmaintained for ~10 months and never tagged a release โ keeping it side-by-side with `flow-next` confused new users about which to install. Marketplace metadata (`.claude-plugin/marketplace.json`), `scripts/install-codex.sh`, `scripts/bump.sh`, and ` | High | 5/9/2026 |
| flow-next-v0.42.0 | ### Added - **`/flow-next:make-pr` โ PR-as-cognitive-aid skill.** New eighteenth slash command closes the gap between "all tasks done" and "human reviews the PR." Five phases (pre-flight โ gather โ build body โ mermaid โ push + create) render a reviewable PR body from rich flow-next state: epic spec with R-IDs, per-task `done_summary` + evidence commits, `decisions` / bug / `architecture-patterns` memory, glossary changes, strategy alignment, deferred review findings, and the diff itself. Body | High | 5/7/2026 |
| flow-next-v0.41.0 | ### Changed - **CI smoke matrix expanded to 7 suites on ubuntu / macos / windows.** Beyond `ci_test.sh` (already in matrix), the workflow now runs `resolve-pr_smoke_test.sh`, `strategy_smoke_test.sh`, `audit_smoke_test.sh`, `glossary_smoke_test.sh`, `prospect_smoke_test.sh`, `impl-review_smoke_test.sh`, and `smoke_test.sh` on each OS leg. ~596 assertions per leg, ~260s runtime, matrix wall time ~4 min. `fail-fast: false` so one OS failure no longer cancels the others; `defaults.run.shell: bash` | High | 5/2/2026 |
| flow-next-v0.38.2 | ### Fixed - **`flowctl.py` subprocess calls now pin `encoding="utf-8"` instead of defaulting to the system locale.** On Windows, `subprocess.run(input=prompt, text=True, ...)` decodes through `locale.getencoding()` โ which is **cp1252** by default โ so any prompt containing characters outside the cp1252 range (Unicode in git diffs, prototype/documentation files, non-ASCII commit messages) raised `UnicodeEncodeError: 'charmap' codec can't encode characters ...`. Setting `PYTHONIOENCODING=utf-8` | High | 4/27/2026 |
| flow-next-v0.31.0 | ### Added - **Unified review backend spec parser** โ `backend[:model[:effort]]` grammar accepted at every surface (env, config, per-task, per-epic, CLI flag). `parse_backend_spec()` + `BackendSpec` dataclass + `BACKEND_REGISTRY` (rp/codex/copilot/none) validate specs on store; invalid values rejected with helpful errors listing valid models/efforts. Legacy bare-backend values (`codex`, `copilot`, `rp`) still work unchanged. Unparseable strings on disk degrade to bare backend with a stderr warni | High | 4/22/2026 |
| flow-next-v0.29.4 | ### Fixed - **rp-cli 2.1.6: builder output missing tab/context ID** โ `cmd_rp_builder` and `cmd_rp_setup_review` now always pass `--raw-json` to builder (was conditional on `--response-type`). RP 2.1.6 removed the `Tab:`/`Context:` text line from plain-text output; IDs are only in JSON mode. JSON parse tried first, regex fallback for older RP versions. Closes #109. Thanks @berhanbero - **Python 3.12+ `datetime.utcnow()` deprecation** โ replaced with `datetime.now(timezone.utc)` in `now_iso()` a | High | 4/12/2026 |
| flow-next-v0.29.3 | ### Fixed - **RepoPrompt 2.x `oracle_send` support** โ `flowctl rp chat-send` now prefers RP 2.x `oracle_send` over legacy `chat_send`, falling back only on missing-tool errors. Strips `chat_name` and `selected_paths` fields that RP 2.1.x rejects. Real errors propagate immediately instead of being masked by fallback. Thanks @clairernovotny โ [#107](https://github.com/gmickel/flow-next/pull/107) - **Ralph receipt gate hardened** โ review receipts now require `type`, `id`, and `verdict` (SHIP/NEE | Medium | 4/12/2026 |
| flow-next-v0.29.2 | ### Fixed - **RepoPrompt 2.1.4 `Context:` builder output** โ `flowctl rp builder` / `flowctl rp setup-review` now accept the new `Context: <uuid>` text format and `context_id`/`context`/`contextId` JSON keys alongside the legacy `Tab:` / `tab_id` shapes. Downstream `--tab` flag unchanged; legacy paths still tried first for backward compat. CI regression coverage added. Thanks @clairernovotny โ [#106](https://github.com/gmickel/flow-next/pull/106) | High | 4/9/2026 |
| flow-next-v0.29.1 | ### Fixed - **RepoPrompt workspace leak in setup-review** โ Ralph sessions with `WORK_REVIEW=rp` could accumulate dozens of duplicate RepoPrompt workspaces/windows for the same repo when window matching fell through to `workspace create --new-window` on every retry. Now falls back through three layers: `bind_context` (RP's native repo-path matching, newest API) โ workspace inventory lookup by repo path โ last-resort creation. Hidden workspaces are reopened via `manage_workspaces switch` instead | High | 4/8/2026 |
| flow-next-v0.29.0 | ### Added - **DESIGN.md awareness** โ conditional design system integration when Google Stitch DESIGN.md exists - repo-scout detects and validates DESIGN.md (section headings + hex color heuristic) - Plan skill writes `## Design context` in frontend task specs with relevant tokens - Worker reads DESIGN.md sections in Phase 1.5 when design context present - Prime Pillar 4 DC7 criterion: DESIGN.md exists (informational) - docs-gap-scout scans for DESIGN.md and .stitch/DESIGN.md - Quality-auditor | Medium | 4/6/2026 |
| flow-next-v0.28.0 | ### Added - **Investigation targets** in task specs โ plan writes file paths (Required/Optional) workers must read before coding, reducing hallucination and ensuring pattern conformance - **Requirement coverage** traceability table in epic specs โ maps each requirement to covering task(s) with gap justification, maintained by plan-sync on drift - **Early proof point** in epic specs โ identifies which task validates the core approach and what to reconsider if it fails - **Bidirectional epic-revi | Medium | 4/5/2026 |
| flow-next-v0.27.0 | ### Added - **Native Codex plugin support** (`.codex-plugin/plugin.json`) โ Flow-Next is now a first-class Codex plugin discoverable via `/plugins` - **Codex marketplace discovery** (`.agents/plugins/marketplace.json`) โ repo works as a Codex marketplace source - **Pre-built Codex agents** as `.toml` files with subagent optimizations (`sandbox_mode`, `nickname_candidates`) - **Pre-built Codex skills** with platform-specific invocation patterns (`$flow-next-plan` instead of `/flow-next:plan`) - | Medium | 4/5/2026 |
| flow-next-v0.26.1 | Release flow-next-v0.26.1 | Low | 3/22/2026 |
| flow-next-v0.26.0 | ### Changed - **Codex model defaults: gpt-5.4 across the board** โ review/oracle model upgraded from `gpt-5.2` to `gpt-5.4` (high reasoning). Agent intelligent tier upgraded from `gpt-5.3-codex` to `gpt-5.4` (high reasoning). Fast scouts remain on `gpt-5.3-codex-spark`. ### Fixed - **Codex docs: removed incorrect Ralph support claim** โ "What works" section incorrectly listed Ralph autonomous mode. Ralph requires plugin hooks (guard hooks, receipt gating) which Codex doesn't support. Expande | Low | 3/6/2026 |
| flow-next-v0.25.0 | ### Fixed - **Codex reviews: embed files on all platforms** โ removed `os.name == "nt"` gate that restricted file embedding to Windows only. On Unix/macOS, Codex wasted its entire turn budget reading files via `sed`/`rg` before producing a verdict (observed 114 shell commands, 3.68M tokens, no verdict on complex epics). Now always embeds changed files with budget-aware fallback: disk reads allowed when embed budget is exceeded. Default `FLOW_CODEX_EMBED_MAX_BYTES` raised from 100KB to 500KB. ( | Low | 3/1/2026 |
| flow-next-v0.24.0 | ### Added - **Spec-driven workflow** โ "create a spec for X" now has guidance in the CLAUDE.md/AGENTS.md snippet installed by `/flow-next:setup`. Creates an epic with structured spec template (Goal & Context, Architecture & Data Models, API Contracts, Edge Cases & Constraints, Acceptance Criteria, Boundaries, Decision Context). Then choose `/flow-next:plan` (task breakdown) or `/flow-next:interview` (refine spec). - **README: spec-driven entry point** โ new "Spec-driven" workflow section in "W | Low | 2/21/2026 |
| flow-next-v0.23.0 | ### Added - **Browser skill: comprehensive update from upstream agent-browser** โ synced with latest `vercel-labs/agent-browser` skill. New features: version check on use, command chaining guidance, `snapshot -i -C` (cursor-interactive), `click --new-tab`, diff commands (snapshot/screenshot/url comparison), annotated screenshots (`--annotate` vision mode), safe JS eval (`--stdin`/`-b`), config file support, session persistence with encryption (`--session-name`), `--auto-connect` for existing C | Low | 2/20/2026 |
| flow-next-v0.22.3 | ### Fixed - **RP `--create` fails on empty default window** โ when only an empty RP window exists (no folder loaded), `setup-review --create` reused it instead of creating a workspace with the repo folder, causing "No workspace open" from the builder. Now falls through to workspace creation. | Low | 2/19/2026 |
| flow-next-v0.22.2 | ### Fixed - **Codex: ensure `multi_agent` at TOML root** โ `generate_config_entries` appended `multi_agent = true` at end of config.toml, which landed inside a preceding table instead of at root scope. Now prepended before any `[table]` header. - **Codex: deduplicate `[agents]` table** โ installer always emitted a fresh `[agents]` declaration; if user already had one, the resulting file was invalid TOML. Now checks before declaring. - **Codex: patch prime workflow for multi-agent** โ `Task flo | Low | 2/19/2026 |
| flow-next-v0.22.1 | Release flow-next-v0.22.1 | Low | 2/19/2026 |
| flow-next-v0.22.0 | ### Fixed - **Fix receipt-reset false positive on codex reviews** โ PostToolUse receipt-write detection matched codex commands containing `--receipt` path + `>` chars in stdout (from `<verdict>` tags), causing `chat_send_succeeded` and `codex_review_succeeded` to reset immediately after being set. Receipt-write detection now uses proper shell redirect pattern matching (same regexes as PreToolUse) instead of naive substring checks. (thanks @clairernovotny for reporting) ### Added - **Block se | Low | 2/17/2026 |
| flow-next-v0.21.0 | ### Changed - **Upgrade scout agents from Haiku to Sonnet 4.6** โ All 11 lightweight scout agents (build, claude-md, docs-gap, env, epic, memory, observability, security, testing, tooling, workflow) now use `claude-sonnet-4-6` (pinned) instead of `haiku`. Sonnet 4.6 brings improved reasoning, instruction following, and a training data cutoff of Jan 2026. Requires Claude Code 2.1.45+. | Low | 2/17/2026 |
| flow-next-v0.20.21 | ### Changed - **github-scout now opt-in** โ Disabled by default (`scouts.github: false`). Enable via `/flow-next:setup` or `flowctl config set scouts.github true`. Reduces planning cost and removes `gh` CLI requirement for users who don't need cross-repo search. | Low | 2/10/2026 |
| flow-next-v0.20.20 | ### Fixed - **Review skills: prevent double context build** โ Reordered RP workflow in impl-review, plan-review, and epic-review to run context-gathering before setup-review. Builder now runs once with a real summary instead of a placeholder. Added guardrails against re-running setup-review. | Low | 2/7/2026 |
| flow-next-v0.20.19 | Release flow-next-v0.20.19 | Low | 2/3/2026 |
| flow-next-v0.20.16 | ### Fixed - **Full cross-platform variable support** โ Hooks and skills now use `${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}` bash fallback pattern. Works on both Claude Code and Factory Droid without duplication. Hook matchers use `Bash|Execute` regex for both platforms. > **Note:** v0.20.10โ0.20.16 added Factory Droid compatibility. If you experience issues on Claude Code, downgrade to v0.20.9: `claude plugins install flow-next@0.20.9` | Low | 2/3/2026 |
| flow-next-v0.20.15 | ### Fixed - **Restore read-only scout permissions** โ v0.20.14 inadvertently gave all agents Edit/Write access. Now scouts use `disallowedTools: Edit, Write, Task` to maintain read-only restrictions while staying cross-platform compatible (no whitelist of tool names that differ between Claude Code and Droid). | Low | 2/3/2026 |
| flow-next-v0.20.14 | ### Fixed - **Full Droid compatibility** โ Removed explicit `tools:` field from all agents. Both platforms now inherit their native tools automatically. Fixes "partially loaded" issue on Factory Droid caused by unknown tool names (`WebFetch`/`FetchUrl`, `Bash`/`Execute`). | Low | 2/3/2026 |
| flow-next-v0.20.13 | ### Fixed - **Droid Bash/Execute compatibility** โ Added `Execute` alongside `Bash` in 18 agents. Droid uses `Execute`, Claude Code uses `Bash` โ now both work. | Low | 2/3/2026 |
| flow-next-v0.20.12 | ### Fixed - **Droid agent tool compatibility** โ Added `FetchUrl` alongside `WebFetch` in 7 agents (context-scout, docs-scout, flow-gap-analyst, github-scout, practice-scout, quality-auditor, repo-scout). Droid uses `FetchUrl`, Claude Code uses `WebFetch` โ now both work. | Low | 2/3/2026 |
| flow-next-v0.20.11 | ### Changed - **Marketplace reorder** โ flow-next now listed first (Droid auto-installs first plugin when adding marketplace) | Low | 2/3/2026 |
| flow-next-v0.20.10 | ### Fixed - **Factory Droid compatibility** โ Plugin version checks now work on both Claude Code (`.claude-plugin/`) and Factory Droid (`.factory-plugin/`). Skills gracefully handle either directory structure. | Low | 2/3/2026 |
| flow-next-v0.20.9 | ### Fixed - **Cleaner Ralph branch names** โ Branch format changed from `ralph-20260203T143000Z-hostname-email-pid-rand` to `ralph-20260203-143000-rand`. Removes PII (hostname, email) and noise (PID) from git history. Full verbose ID preserved in logs for debugging. Thanks to [@aleparreira](https://github.com/aleparreira) for the report! (#90) ### Added - **ZSH-safe file truncation helper** โ Added `truncate_file()` function using `: > "$file"` pattern for portable file truncation across bas | Low | 2/3/2026 |
| flow-next-v0.20.8 | ### Fixed - **Double context builder in reviews** โ SKILL.md files for epic-review, impl-review, and plan-review no longer contain duplicate executable code. Now explicitly direct agent to workflow.md as single source of truth. Fixes issue where agent would run setup-review and chat-send twice. ### Changed - **Codex install script improvements**: - Agents now installed to `~/.codex/agents/` with frontmatter converted to Codex format (`profile`, `approval_policy`, `sandbox_mode`) - `flow- | Low | 2/3/2026 |
| flow-next-v0.20.7 | ### Fixed - **Epic ID collision prevention** โ `scan_max_epic_id` now scans both `epics/*.json` and `specs/*.md` to catch orphaned specs created outside flowctl. Prevents reusing numeric IDs when specs exist without matching epic JSON. - **Collision detection in validate** โ `flowctl validate --all` now detects and reports epic ID collisions (multiple epics with same `fn-N` prefix) as errors. - **Orphaned spec warnings** โ `flowctl validate --all` warns about specs without matching epic JSON f | Low | 2/2/2026 |
| flow-next-v0.20.6 | Release flow-next-v0.20.6 | Low | 2/2/2026 |
| flow-next-v0.20.5 | ### Fixed - **Duplicate skill/command listings** โ Skills that have command stubs now set `user-invocable: false` to hide from `/` menu. Commands remain the user-facing entry points; skills still work when Claude invokes them. | Low | 1/31/2026 |
| flow-next-v0.20.4 | ### Added - **`epic set-title` command** โ Rename epics by updating title and slug: `flowctl epic set-title fn-1-old --title "New Title"`. Renames all related files, updates task references and `depends_on_epics` in other epics. | Low | 1/31/2026 |
| flow-next-v0.20.3 | ### Changed - **Readable epic IDs** โ Epic IDs now use slugified titles instead of random suffixes. `fn-23-zgk` โ `fn-23-readable-epic-ids`. Random 3-char suffix only used as fallback for empty/special-char titles. Existing IDs remain fully compatible. ### Updated - All error messages and CLI help strings to show new slug format examples - TUI regex patterns to accept slug-based IDs - Skill docs with new ID format examples | Low | 1/31/2026 |
| flow-next-v0.20.2 | ### Added - **`task set-deps` command** โ Set multiple task dependencies in one call: `flowctl task set-deps fn-1.3 --deps fn-1.1,fn-1.2`. Convenience wrapper for `dep add` that matches the `--deps` syntax from `task create`. | Low | 1/30/2026 |
| flow-next-v0.20.1 | ### Added - **Epic dependency visualization skill** โ New `flow-next-deps` skill shows epic dependency graphs, blocking chains, and execution phases. Triggers on "what's blocking", "execution order", "critical path", "which epics can run in parallel". Uses flowctl for data access with jq-based phase computation. Thanks [@clairernovotny](https://github.com/clairernovotny)! (PR #85) ### Fixed - **Skill count sync** โ Updated manifest descriptions to reflect actual counts (20 subagents, 11 comm | Low | 1/30/2026 |
| flow-next-v0.20.0 | ### Added - **Epic-completion review gate** โ New `/flow-next:epic-review` skill runs when all epic tasks complete, before epic closes. Two-phase review (extract requirements โ verify coverage) catches gaps that per-task impl-review misses: decomposition gaps, cross-task requirements, scope drift. Supports RepoPrompt and Codex backends. Closes #83. - **flowctl commands** โ `codex completion-review` for LLM-driven epic review, `epic set-completion-review-status` for manual status control, `--r | Low | 1/30/2026 |
| flow-next-v0.19.1 | ### Fixed - **Plan skill scout enforcement** โ Added CRITICAL block requiring ALL scouts to run in parallel during planning. Previously, agents would skip scouts "because they seem most relevant", causing incomplete plans missing external docs, epic dependencies, and practice pitfalls. - **Task dependency guidance** โ Updated steps.md to document existing `--deps` flag on `task create`. Removes incorrect guidance that said flag didn't exist. Shows preferred inline dependency declaration vs se | Low | 1/30/2026 |
| flow-next-v0.19.0 | ### Changed - **Worker review enforcement** โ Phase 4 header now reads "MANDATORY if REVIEW_MODE != none" with clearer instruction that worker must invoke `/flow-next:impl-review` and receive SHIP verdict before proceeding to Phase 5. Addresses issue where worker would skip review phase entirely. - **Stop hook guidance improved** โ When worker tries to stop without completing review, the ralph-guard hook now tells the worker to invoke the review skill (`/flow-next:impl-review` or `/flow-next: | Low | 1/28/2026 |
| flow-next-v0.18.27 | ### Added - **`--config` flag for Ralph** โ Specify alternate config file: `ralph.sh --config my-codex-config.env`. Enables different configs for different platforms/review backends without editing config.env. Closes #82. | Low | 1/28/2026 |
| flow-next-v0.18.26 | ### Added - **Version check warning in Ralph** โ Ralph now checks if local setup version differs from plugin version at startup. Shows warning: "Plugin updated to vX.Y.Z. Run /flow-next:setup to refresh local scripts (current: vA.B.C)." Non-blocking, warn only. | Low | 1/28/2026 |
| flow-next-v0.18.25 | ### Fixed - **Block Explore auto-delegation in Ralph mode** โ Worker subagent has `disallowedTools: Task` but enforcement is inconsistent (known Claude Code bugs #21295, #21296). When Explore was auto-spawned, it failed with READ-ONLY constraint and couldn't write receipts, causing infinite retry loops. Now explicitly block `Task(Explore)` at CLI level in ralph.sh (precedence 2 beats agent frontmatter precedence 6). Interactive mode unaffected - fix only applies to Ralph autonomous sessions. | Low | 1/27/2026 |
| flow-next-v0.18.24 | ### Fixed - **Epic dependency race condition** โ Move `maybe_close_epics()` before selector in Ralph loop. Previously, dependent epics remained blocked when parent epic completed because closing happened after selector returned `NO_WORK`. Now epics close at iteration start, unblocking dependents immediately. Thanks [@tiagoefreitas](https://github.com/tiagoefreitas)! (#79) | Low | 1/26/2026 |
| flow-next-v0.18.23 | ### Added - **Plan Review Gate documentation** โ Comprehensive docs for Ralph's plan review gate: how it works, configuration matrix, review cycle, checkpoint recovery, status inspection, and comparison with impl review. Added troubleshooting for common issues: plan review never starts, blocked forever, dependent epics not starting. | Low | 1/26/2026 |
| flow-next-v0.18.22 | ### Fixed - **Ralph plan prompt aligned with skill** โ Added checkpoint save before plan review, task spec sync mention, and checkpoint restore on context compaction. Ensures Ralph plan gate has same recovery capabilities as interactive `/flow-next:plan-review`. | Low | 1/26/2026 |
| flow-next-v0.18.21 | ### Added - **Backend spec fields for tasks and epics** โ New optional `impl`, `review`, `sync` fields on tasks and `default_impl`, `default_review`, `default_sync` on epics. These fields store preferred AI backend + model specs (e.g., `codex:gpt-5.2-high`, `claude:opus`). Pure storage - flowctl doesn't interpret them; orchestration products like flow-swarm use them to route different tasks to different backends. - **`flowctl task set-backend`** โ Set backend specs on a task: `flowctl task se | Low | 1/26/2026 |
| flow-next-v0.18.20 | ### Changed - **Task sizing: M is the sweet spot** โ Updated plan skill to prefer M-sized tasks over many S tasks. Sequential S tasks should be combined into M tasks. Added "7+ tasks = look for tasks to combine" heuristic. - **OAuth example: 4 tasks โ 2 tasks** โ Task breakdown example now shows combining sequential backend work into one M task + separate frontend S task. Added "over-split" anti-pattern example. - **Plan review checks for over-splitting** โ Added "Task sizing" as review crit | Low | 1/26/2026 |
| flow-next-v0.18.19 | ### Changed - **Memory and Plan-Sync enabled by default** โ New projects now have `memory.enabled: true` and `planSync.enabled: true` out of the box. Cross-epic sync remains disabled by default to avoid long Ralph loops. Disable with `flowctl config set memory.enabled false` or `flowctl config set planSync.enabled false`. | Low | 1/26/2026 |
| flow-next-v0.18.18 | ### Fixed - **Preserve GH-73 COMPLETE handling fix** โ PR #74 inadvertently reverted the fix for premature completion in Ralph. Workers should NEVER output `<promise>COMPLETE</promise>` (prompts forbid it); completion is detected via selector returning `status=none`. Restored the ignore-and-log behavior. ### Documentation - **Improved `--files` guidance in plan-review skills** โ Added explanation of how to identify which files to pass (read epic spec, find affected paths) instead of just a h | Low | 1/25/2026 |
| flow-next-v0.18.17 | ### Fixed - **Filter artifact files using is_task_id() validation** โ Replaced weak `"." not in task_id` check with proper `is_task_id()` regex validation. Fixes `KeyError: 'title'` crash when `.flow/tasks/` contains artifact files like `fn-1.2-review.json`. Works with both legacy (`fn-3.1`) and new (`fn-3-sds.1`) ID formats. Thanks to @kirillzh for the contribution! | Low | 1/25/2026 |
Release History