warp-os 1.1.0

package/CHANGELOG.md

@@ -0,0 +1,327 @@
+# Changelog
+
+All notable changes to Warp are documented in this file.
+
+Format: [Semantic Versioning](https://semver.org/). Sections: Added, Changed, Removed, Fixed.
+
+---
+
+## [1.1.0] - 2026-04-05
+
+Persistent cross-session memory via [claude-mem](https://github.com/thedotmack/claude-mem). Automatic observation capture replaces manual session handoffs.
+
+### Added
+- **claude-mem integration.** Automatic session memory — captures every tool use, AI-compresses observations, injects progressive disclosure context index on startup, cross-session search via MCP tools (search, timeline, get_observations).
+- **consistency-check.sh.** New Stop hook validates CLAUDE.md staleness and TODOS.md accuracy when work is done. Extracted from warp-save's consistency checks.
+
+### Changed
+- **Hook architecture.** Warp hooks (identity + consistency + PowerShell blocker) coexist independently with claude-mem's plugin hooks. No coordination needed — both fire on shared events.
+- **install.sh.** Adds `npx claude-mem install` step. Cleans up lifecycle-session-start.sh from previous installs.
+- **project-hooks.json.** SessionStart reduced to 2 hooks (identity-foundation + identity-briefing). Stop event added (consistency-check). claude-mem manages its own hooks via plugin config.
+
+### Removed
+- **warp-save** — session handoff skill replaced by claude-mem automatic capture. 17 → 16 skills.
+- **lifecycle-session-start.sh** — session loading hook replaced by claude-mem context injection.
+- **.warp/sessions/ scaffolding** — warp-setup no longer creates session directories. claude-mem uses SQLite at ~/.claude-mem/.
+
+---
+
+## [1.0.0] - 2026-04-04
+
+First public release. 17 skills, 3 adversarial agents, 4 hooks. Clean history.
+
+### Changed
+- **qa-optimize absorbs qa-polish.** Three modes: `/optimize` (layers 1-3, plan-only), `/delight` (layer 4, implements visual polish + CEO delight pass), `/polish` (all 4 layers). Design token audit, copy polish, micro-interactions, and 15-minute adjacent improvements all live in Layer 4.
+- **release-update absorbs release-retro.** Two modes: ship (diff review, docs update, version bump, push, PR, deploy, canary) and retro (commit analysis, work patterns, code quality, wins, action items). Detects mode intelligently from context. Also triggers: `/ship`, `/retro`.
+- **Push gate.** All pushes to remote go through `/warp-release-update`. Commits happen anytime. The update skill verifies CHANGELOG, README, and CLAUDE.md are current before pushing.
+
+### Removed
+- **warp-integrate** — integration checks handled naturally by the orchestrator.
+- **warp-synthesize** — roadmap synthesis handled by the orchestrator directly.
+- **warp-resume** — session bootstrap fully handled by hooks (lifecycle-session-start + identity-briefing).
+- **warp-qa-polish** — absorbed into warp-qa-optimize Layer 4 (delight mode).
+- **warp-release-retro** — absorbed into warp-release-update (retro mode).
+- **warp-qa-polish-adversarial** — removed with polish skill. 4 → 3 adversarial agents.
+
+---
+
+## [0.5.1] - 2026-04-04
+
+Tier consolidation. Three tiers → two. Simpler, smaller, no lost coverage.
+
+### Changed
+- **T1+T3 merged into single T1.** UX patterns (AskUserQuestion format, completeness scoring, scale detection, artifact I/O, completion banner) folded into Tier 1. One file instead of two. 374 → 176 lines.
+- **T2 → adversarial-only.** Tier 2 no longer prepended to build/QA skills. Only prepended to 4 adversarial agent definitions. 212 → 48 lines. Contains: bias model, verification hierarchy, severity tags, findings format.
+- **build.sh simplified.** Two-tier model: T1 for all skills and agents, T2 only for adversarial agents. Added adversarial agent build loop.
+- **verify.sh updated.** Replaced tier3 checks with T1 line count (≤200) and adversarial agent parity. 11/11 checks.
+- **install.sh updated.** Removed T3 copy, added cleanup of old tier files on upgrade.
+- **identity-foundation.sh updated.** Removed T3 injection (merged into T1).
+- **README.md updated.** Architecture tree, tier model description, hook descriptions — all reflect 2-tier model.
+- **Stale references fixed.** 5 references to removed tier2-verification.md tables in warp-setup, warp-plan-onboarding, warp-plan-architect source files.
+
+### Added
+- **`src/adversarial/` directory.** 4 adversarial agent source files (qa-test, qa-debug, qa-optimize, qa-polish). Build.sh processes them separately with T2 prepended.
+
+### Removed
+- **`shared/tier3-ux-patterns.md`.** Content merged into T1. File deleted.
+
+---
+
+## [0.5.0] - 2026-03-30
+
+Direct-first execution model. The user stays in the loop for every build and QA decision.
+
+### Added
+- **Dual-mode QA.** All 4 QA skills run direct + adversarial dispatch simultaneously. Auto-diff produces categorized comparison: blind spots (adversarial-only), confirmed (both), context-dependent (direct-only). 4 new adversarial agent definitions.
+- **Figma MCP integration.** Design skill creates in Figma, build reads from Figma (`get_design_context`), QA compares against Figma (`get_screenshot`), round-trip captures live app back into Figma (`generate_figma_design`). Figma rules auto-generated during setup via `create_design_system_rules`.
+- **Chrome extension as build canvas.** Build-code starts dev server + Chrome for live UI collaboration. User and Claude share the same browser window during implementation.
+- **AskUserQuestion triggers.** Tier 3 gains mandatory trigger list (8 situations that MUST produce a question) + general "bias toward asking" principle.
+- **Completeness scores in labels.** Every AskUserQuestion option label MUST include score + emoji in the label text.
+- **Preview length rule.** AskUserQuestion previews stay under 8 lines. Full mockups in conversation text.
+- **Reports directory.** `.warp/reports/` organized by phase: planning/, building/, qatesting/, releasing/, roadmap/. Replaces flat `.warp/pipeline/`.
+
+### Changed
+- **Build-code → direct.** No longer dispatched. User collaborates in real-time on TDD cycles. Chrome for visual verification, Figma for design specs.
+- **L1 gate → inline.** Moved from TaskCompleted hook into build-code Phase 3.5 (pre-QA). User sees gate results and fixes inline.
+- **API docs → context injection.** No longer a blocking gate hook. Injected as context when QA skills are invoked.
+- **warp-release-launch → warp-release-update.** Honest name for what the skill does: version bump, changelog, commit, push, PR.
+- **Orchestrator → direct-first routing.** All skills default to direct. QA additionally dispatches adversarial agents. User can request dispatch for any skill.
+
+### Removed
+- **warp-build-sprint** — replaced by `/loop` or manual re-invocation.
+- **warp-build-teleport** — replaced by `/loop`. Autonomous decisions unnecessary with user present.
+- **11 hooks** — guard-orch, guard-block-md, verify-qa-gate, verify-qa-boundary, verify-api-docs, automation-phase-boundary, automation-test-router, lifecycle-session-end, lifecycle-pre-compact, lifecycle-post-compact, lifecycle-subagent-inject, lifecycle-subagent-quicksave. Plus 2 agent hooks (TaskCompleted, SubagentStop). 15 → 4 hooks.
+- **Hook-driven QA automation.** QA is now user-invoked, not auto-dispatched by hooks.
+
+---
+
+## [0.4.2] - 2026-03-29
+
+API doc registry, build visibility, and quality-of-life improvements.
+
+### Added
+- **API doc registry + L1 gate.** New `api_docs` section in `warp-tools.json` maps every dependency to a doc source (Context7 ID, local file, or skipped). New `verify-api-docs.sh` hook blocks cycles with unregistered or unresolved dependencies. Integrated into setup, onboarding, architect, and build-code.
+- **Doc query receipt.** Build-code writes `doc-queries.json` in Phase 1C. The gate verifies imported libraries were actually queried — proves the agent read the docs, not just that docs exist. Three-layer defense: L3 advisory, L1 registry, L1 receipt.
+- **Build visibility.** Build agents report real-time progress via TaskCreate/TaskUpdate. Users see each phase tick off (Red, Green, Refactor, Gate) during dispatched builds. Problem tasks surface blockers immediately.
+- **Pre-dispatch briefing.** Orchestrator enters plan mode before dispatching build agents, presenting granular cycle steps for user approval. Plan file name controls the HUD text.
+- **15th hook.** `verify-api-docs.sh` added to the verify concern group.
+
+### Changed
+- **warp-orchestrate → warp-orchestrator.** Renamed for clarity (noun, not verb). Cascading rename across all references.
+- **warp-resume trimmed.** 326 → 89 lines. Same pattern as warp-save trim: cut preamble, MUST/MUST NOT, calibration examples.
+- **Build-code Phase 1C rewritten.** Registry-driven doc lookup replaces advisory Context7 instructions. Resolved deps go straight to `query-docs` with stored ID.
+- **Context7 Integration section rewritten.** Now centered on the API doc registry rather than ad-hoc queries.
+
+### Fixed
+- **Forced model fields removed.** Skills inherit the user's selected model instead of forcing a specific one. Likely root cause of CC5 rate limit issue.
+- **Tier duplication in orchestrator context eliminated.**
+
+---
+
+## [0.4.0] - 2026-03-29
+
+Major architecture redesign: governance-by-hooks, not governance-by-prose.
+
+### Changed
+- **Tier 3 decomposed.** 565 lines → 99 lines. Mechanical content moved to hooks, cognitive UX patterns retained. Renamed `tier3-ux-infrastructure.md` → `tier3-ux-patterns.md`.
+- **Tier 2 optimized.** 272 → 196 lines. Removed duplicates (bias classification, command allowlist, gate protocol). Absorbed cycle/QA protocol from Tier 3.
+- **14 hooks renamed to taxonomy.** 6 concern groups: `lifecycle-*`, `identity-*`, `guard-*`, `verify-*`, `automation-*`. Old names cleaned up on install/upgrade.
+- **Orchestrator thinned.** 5,409 → 3,524 tokens (-35%). State assessment and briefing moved to hooks. Orchestrator is now a pure router: read state → ask user → route → display.
+- **Plan skills default to direct.** Orchestrator defaults plan/meta/util skills to direct execution (collaborative), build/QA/release to dispatch (fresh context).
+- **All 24 skills normalized.** Zero mechanical boilerplate, zero stale references, all under 15k token budget.
+
+### Added
+- **Governance receipt.** `verify-qa-gate.sh` outputs formatted banner after every L1 gate: per-tool lines with ✓/✗, name, duration, test count, summary.
+- **3 new verify.sh checks** (9 → 12): tier3 line count ≤100, skill token budget ≤15k, mechanical content detector.
+- **`_warp_html.sh` utility.** Markdown → HTML with warp streak CSS. XSS-safe (HTML tag sanitization). Cross-platform browser opening.
+- **`HOOKS.md`.** Hook lifecycle flowchart documenting all 14 hooks, events, and execution order.
+- **First-run detection.** `lifecycle-session-start.sh` warns if project not set up.
+- **First-run orientation.** `identity-foundation.sh` shows welcome on first use, marks `.intro-seen`.
+- **Welcome banner.** `identity-briefing.sh` generates formatted 48-char banner with pipeline state, branch, P1 priorities, model warning.
+- **Description fields.** Every hook entry in `project-hooks.json` has a `description` field.
+
+### Removed
+- Old `tier3-ux-infrastructure.md` (565 lines of mixed mechanical+cognitive content).
+- Bias classification table from Tier 2 (duplicate of Tier 1).
+- Command allowlist from Tier 2 (lives in `verify-qa-gate.sh`).
+- Deterministic gate protocol from Tier 2 (IS `verify-qa-gate.sh`).
+- State assessment phase from orchestrator (hooks handle this).
+- Briefing kit from orchestrator (hooks inject context).
+
+---
+
+## [0.3.6] - 2026-03-28
+
+Install and upgrade polish.
+
+### Added
+- **Stale skill/agent cleanup.** install.sh removes warp-* symlinks and agent definitions that no longer exist in the repo. Fixes ghost entries like warp-learn surviving after removal.
+- **Smart restart detection.** warp-upgrade.sh hashes hooks config and agent definitions before/after. Only recommends restart when hooks or agents actually changed, with specific reasons. Content-only updates say "no restart needed."
+- **Git remote in upgrade output.** Shows `Repo: github.com/...` alongside local path so you know what you're upgrading.
+
+---
+
+## [0.3.5] - 2026-03-28
+
+QA findings tracker — persistent checklist that gates phase progression.
+
+### Added
+- **Unified findings tracker** (`.warp/pipeline/qa/findings.md`). All QA skills append findings. Format: `- [ ]` OPEN, `- [x]` FIXED, `- [~]` DEFERRED. Persists across skills and sessions.
+- **Findings gate in phase-boundary-detect.sh.** Open findings block phase/subphase progression. Writes "blocked" signal instead of boundary signal when findings remain.
+- **FINDINGS TRACKER section** in qa-test, qa-optimize, qa-polish, and qa-debug skills. Each skill knows its role: test/optimize/polish append findings, debug marks them FIXED, orchestrator marks DEFERRED (user approval required).
+- **DEFERRED workflow** in orchestrator. Present open findings, user chooses fix vs defer, orchestrator updates tracker accordingly.
+
+### Fixed
+- **Orchestrator dropped QA findings after fixing high-priority items.** Without a persistent tracker, lower-priority findings existed only in conversation context and were forgotten. Now they persist on disk with a deterministic gate.
+
+---
+
+## [0.3.4] - 2026-03-28
+
+Hook-driven phase boundary QA and qa-debug improvements.
+
+### Added
+- **phase-boundary-detect.sh** — L1 hook on SubagentStop. Scans roadmap for completed phases/subphases when a build subagent returns. Writes `.phase-boundary` signal for agent hook dispatch. Tracks completed boundaries to prevent re-signaling.
+- **Phase boundary agent hook** — SubagentStop agent hook reads `.phase-boundary` and auto-dispatches the full QA suite: qa-test (Exhaustive) → qa-debug (if bugs found) → qa-optimize → qa-polish. Orchestrator is completely out of the loop.
+- **qa-debug step in QA chains** — Both SubagentStop and TaskCompleted agent hooks now dispatch qa-debug between qa-test and qa-optimize when bugs are found.
+
+### Fixed
+- **Orchestrator asked instead of auto-dispatching phase boundary QA.** Phase 6 said "orchestrator never dispatches QA" which was correct for cycle-level (hooks handle it) but wrong for phase/subphase boundaries. Now hook-driven at all levels.
+- **qa-debug checked arbitrary last N commits.** Now uses session-start HEAD for full session history. Falls back to asking the user for commit range if no session HEAD exists — no more magic numbers.
+
+### Changed
+- **Orchestrator Phase 6 rewritten.** Clarifies: cycle QA = hooks inside build subagent, boundary QA = hooks on SubagentStop, orchestrator = display results only.
+
+---
+
+## [0.3.3] - 2026-03-28
+
+Hooks fix: orch-guard was never wired up, upgrade now auto-patches project hooks.
+
+### Added
+- **test-failure-router.sh** — PostToolUse hook on Bash. Detects test failure output and injects routing guidance to dispatch @warp-qa-debug instead of fixing directly.
+- **shared/project-hooks.json** — Single source of truth for per-project hook config. Both warp-setup and warp-upgrade read from it.
+- **Auto-patch on upgrade.** warp-upgrade.sh auto-patches `settings.local.json` with latest hooks after install. Users never need to re-run /warp-setup for hook updates.
+
+### Fixed
+- **orch-guard.sh was never installed.** Defined in orchestrator SKILL.md frontmatter but missing from warp-setup's hook config template. The orchestrator could freely edit code files — the "architect doesn't lay bricks" guard was dead on arrival. Now included in PreToolUse config.
+- **Dead hooks frontmatter.** Removed non-functional `hooks:` block from orchestrator SKILL.md frontmatter (hooks live in settings.local.json, not skill frontmatter).
+
+### Changed
+- **warp-upgrade no longer asks to re-run /warp-setup.** Upgrade patches hooks automatically. Setup is for first-time install only.
+- **install.sh copies project-hooks.json** to `~/.warp/project-hooks.json`.
+- **warp-setup reads hook config from template** instead of inline JSON.
+- **TODOS.md updated** to reflect v0.3.x shipped state (24 skills, correct filenames, completed items checked).
+
+---
+
+## [0.3.0] - 2026-03-27
+
+Hook architecture rewrite. Fixes the v0.2.0 Stop-fires-every-response bug. All hooks moved to correct CC lifecycle events with per-project ownership, shared JSON parsing, and adversarial review fixes.
+
+### Added
+- **Shared JSON parsing library** (`_warp_json.sh`). All 12 hook scripts source one parser — no jq dependency, tested against adversarial inputs (quotes, newlines, backslashes, 10KB+ markdown).
+- **Orchestrator foundation injection** (`orchestrator-foundation.sh`). Injects Tier 1 (engineering constitution) + Tier 3 (UX patterns) into Claude's context via JSON `additionalContext` on every SessionStart. Raw stdout fallback. This is what makes Claude BE the orchestrator.
+- **Orchestrator briefing injection** (`orchestrator-briefing.sh`). Pipeline state, priorities, branch, and last session info — injected via `additionalContext` at session start.
+- **Post-compaction capture** (`post-compact.sh`). Saves the AI-generated `compact_summary` after context compaction.
+- **Subagent context injection** (`subagent-inject.sh`). Injects cycle specs into build subagents at spawn time via `additionalContext`. Zero orchestrator token cost.
+- **Subagent quicksave** (`subagent-quicksave.sh`). Lightweight state capture on subagent return.
+- **Signal-file QA architecture.** qa-gate.sh writes `.qa-gate-result`, qa-boundary.sh checks it before writing `.qa-signal`, agent hook checks `.qa-signal` existence before dispatching. Each hook independently safe.
+- **`[WARP]` task prefix.** Build tasks use `[WARP] Cycle 1a.1: Title` naming. qa-gate.sh filters on prefix first — instant exit 0 for non-Warp tasks. Prevents false positives from TaskCompleted's no-matcher behavior.
+- **Session-scoped HEAD tracking.** session-start.sh writes `HEAD` to `~/.warp/sessions/.session-start-head-${session_id}`. session-end.sh captures ALL session commits, not just last 5. Concurrent Claude instances don't collide.
+- **Graceful degradation on every hook.** Missing dependencies produce visible `[WARP WARNING]` messages, never silent failures, never crashes.
+- **Hook health verification.** warp-setup runs post-install checks: JSON syntax, script existence/permissions, dry-run.
+- **Glob-pattern .md allowlist.** block-md.sh reads `docs/pipeline/md-allowlist.md` with shell glob patterns. Seeded by warp-setup.
+- **Interactive pipeline flowchart** (`docs/warp-pipeline.html`). Full vertical visualization of the Warp pipeline with hook integration highlighted.
+
+### Changed
+- **session-end.sh: Stop → SessionEnd.** Fires once at session termination, not after every response. The v0.2.0 bug that caused flooding.
+- **qa-gate.sh + qa-boundary.sh: SubagentStop → TaskCompleted.** Exit 2 blocks task completion. L1 QA gate at build cycle boundaries.
+- **All scripts: jq → pure bash.** No external JSON parser dependency. Works on Windows without jq.
+- **All scripts: guard clauses removed.** Per-project hooks (via warp-setup) guarantee Warp context. No "is this a Warp project?" checks.
+- **install.sh: no global hook config.** Hook configuration is per-project in `.claude/settings.local.json`, written by warp-setup. install.sh only distributes scripts.
+- **Single SessionStart group.** Empty matcher catches all sources (startup, resume, clear, compact). No separate compact group — prevents double injection.
+- **Repo renamed** from `warp-skills` to `WarpOS`.
+
+### Removed
+- **Global hooks in `~/.claude/settings.json`.** All Warp hooks moved to per-project `.claude/settings.local.json`.
+- **Haiku prompt hook on Stop.** Tested: SessionEnd output lost. AI summaries run via orchestrator-gated qa-optimize at next session start.
+- **4-hook SubagentStop QA chain.** Replaced by TaskCompleted + single subagent quicksave.
+- **`stop_hook_active` guard.** No longer needed — session-end.sh fires on SessionEnd (once), not Stop (every response).
+
+### Migration
+- **Run `/warp-setup` on all existing projects.** The hook configuration in `.claude/settings.local.json` must be regenerated for the new events.
+- **Restart Claude Code** after setup completes.
+
+---
+
+## [0.2.0] - 2026-03-27
+
+Major architecture restructure. Skills become native Claude Code agents. Orchestrator auto-activates. QA is fully hook-driven. One-command setup for new and existing projects.
+
+### Added
+- **Native agent definitions.** Every skill now has an agent definition in `agents/` (Tier 1 + Tier 2). Agents are dispatched as subagents by the orchestrator with proper context isolation.
+- **Orchestrator auto-activation.** `warp-setup` writes `"agent": "warp-orchestrate"` to project settings. The orchestrator becomes the default session agent — no manual invocation needed.
+- **`initialPrompt` for orchestrator.** Pipeline state shows automatically on session start.
+- **Smart bypass.** Orchestrator detects simple requests and handles them directly without subagent dispatch.
+- **Parallel dispatch.** Design + security run in parallel after architect (the one true independence point in the planning pipeline).
+- **warp-synthesize skill.** Roadmap synthesis extracted from orchestrator into its own skill (opus). Orchestrator stays lean.
+- **Hook-driven QA.** SubagentStop chain: qa-boundary.sh (detect boundary) -> qa-gate.sh (run L1 tools) -> agent hook (dispatch QA subagent) -> quicksave. Orchestrator only displays saved results.
+- **Session lifecycle hooks.** PreCompact (quicksave), SessionStart (auto-load), Stop (full save + Haiku summary). All installed globally by install.sh.
+- **Project hooks.** PostToolUse formatting/linting/types + PreToolUse .md blocker. Auto-generated by warp-setup from warp-tools.json.
+- **Two-layer tool architecture.** `warp_tools` (Warp-native, e.g., gitleaks) + `project_tools` (ecosystem-specific). Single JSON source of truth.
+- **MCP server support.** `mcp_servers` section in warp-tools.json. warp-setup detects and offers Context7 (live library docs) and mgrep (semantic search).
+- **Opinionated project template.** New projects get optional clean-architecture structure (app/ -> modules/ -> shared/) aligned with STANDARDS S8 layers.
+- **Auto-onboarding.** warp-setup Mode B automatically dispatches warp-plan-onboarding. One command for existing projects.
+- **Layer mapping in onboarding.** Maps existing directories to standard dependency layers for downstream skills.
+- **Worktree support in sprint.** Optional git worktree isolation per cycle.
+- **Parallel hypothesis testing in debug.** /fork guidance for multi-hypothesis investigation.
+- **Version tracking.** VERSION file at repo root, `~/.warp/version` tracks installed version.
+- **warp-upgrade.sh.** One-command upgrade: pull, build, install, show changelog, detect re-setup needs.
+- **verify.sh expanded to 9 checks.** Added model field validation and agent definition parity.
+- **Model field on all 23 skills.** 10 opus, 13 sonnet.
+
+### Changed
+- **build.sh dual output.** Now produces compiled skills (Tier 1+2+3) to dist/ AND agent definitions (Tier 1+2) to agents/.
+- **install.sh expanded.** Installs skills (symlinks), agents (symlinks), hooks (copy), Tier 1 (copy), global hook config (jq merge into settings.json).
+- **Orchestrator rewritten.** From roadmap producer to lean pipeline brain. Phases: State Assessment -> Routing -> Dispatch -> Evaluation -> Hard Gate -> QA Display -> Loop.
+- **build-code Phase 3.5 simplified.** Gate logic moved to hook (bin/hooks/qa-gate.sh). Skill references hook, doesn't inline gate.
+- **build-sprint QA simplified.** "Hooks handle QA" replaces inline trigger loop.
+- **VERIFICATION_PATTERNS.md updated.** warp-tools.json schema (two-layer + MCP), command allowlist, CLAUDE.md mirror format.
+- **SKILL_PATTERNS.md skill count.** 22 -> 23 (warp-synthesize added).
+- **Context7 integration conditional.** Checks warp-tools.json for availability, falls back with explicit uncertainty flag.
+
+### Removed
+- **Inline gate logic in build-code.** ~186 lines replaced by hook reference.
+- **Inline QA triggers in sprint/teleport.** Hooks handle all QA dispatch.
+
+---
+
+## [0.1.0] - 2026-03-20
+
+Initial public release. 23 skills, deterministic verification, tiered foundation.
+
+### Added
+- 23 skills across Plan, Build, QA, Release, and Root groups
+- Tiered shared foundation: SKILL_PATTERNS.md (all skills) + VERIFICATION_PATTERNS.md (build/QA only)
+- Deterministic gate with Level 1 tool enforcement
+- Test design separation (bulletproof writes manifests, build-code implements)
+- Tiered QA: Quick per cycle, Optimize per subphase, Full per phase
+- Engineering constitution (hard_rules/engineering-constitution.md)
+- Dual mode: pipeline (reads prior artifacts) or standalone (asks for context)
+- Quicksave protocol for session continuity
+- Completeness scoring and anti-shortcut protocol
+- build.sh with variable resolution and frontmatter extraction
+- install.sh with per-skill symlinks and Windows support
+- verify.sh with 7 deterministic checks
+- Pseudocode mode (toggleable)
+
+### Changed
+- `warp-plan-boost` renamed to `warp-plan-scope`
+- `warp-plan-orchestrate` renamed to `warp-orchestrate`
+- `master-blueprint.md` renamed to `README.md` in docs/pipeline/
+- "Master Blueprint" renamed to "Roadmap" everywhere
+- `warp-util-startup` renamed to `warp-resume`
+- `warp-util-save-session` + `warp-util-annotate` merged into `warp-save`
+- `warp-debug-investigate` renamed to `warp-qa-debug`
+- All `warp-util-*` prefixes dropped

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 WolfOnWings
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.