Changelog

AcePilot gains a full fleet layer for managing multiple projects simultaneously. New fleet mode (13th), cross-pilot shared knowledge, and a redesigned fleet API.

• New fleet mode — fleet health score, per-project IWPS/MRR/state, AceControl status, Skunkworks pipeline, and shared knowledge count.
• New ABSORB step 12 — Fleet knowledge: shared patterns from ~/.claude/fleet/shared-knowledge.json matching this project's stack are applied automatically.
• Improved Fleet API redesigned around AceControl/AceVault hierarchy — projects report metrics, Control dispatches objectives, operators view via dashboard.
• Improved Dashboard fleet view updated to new data shape (projects/control/skunkworks/health/budget).

Brain compressed from 949 → 924 lines. Tighter token budget, same functionality.

• Improved Copy rules section collapsed from 7-line bullet list to 1 compact inline line (−5 lines).
• Improved Analytics + SEO defaults merged from two separate 7-line bullet sections into one compact 4-line block (−10 lines).
• Improved Human Handoff guide principles condensed from 12-line bullet list to 2-line dense format (−10 lines). All rules and functionality preserved — same semantics, tighter token budget.

God mode can now run indefinitely in the background. Brain compressed to 949 lines for faster startup.

• New god --loop (Never Stops Mode): infinite scan→execute→rescan cycle until you say "stop", "pause", or "halt". Cycles every 60s when queue is clean; compacts context every 5 cycles to prevent overflow. Aliases: god never stops, god watch, never stops.
• Improved Brain compressed 981→949 lines: dashboard sync curl blocks converted from verbose if/fi format to compact one-liners; STATS and GUIDE output templates tightened; auto-resolve section condensed. All functionality preserved.

Dashboard connects instantly when AcePilot starts. Multiple projects can be live simultaneously. Per-project metrics (users, CVR, MRR, ad revenue) with benchmark deltas.

• New ABSORB ping: dashboard connects immediately on session start (all modes) — no longer waits until session end.
• New Multi-project live sessions: each project gets its own active slot. Dashboard Live tab shows all running sessions simultaneously with stale detection (>20 min idle = flagged).
• New Dashboard "All Tasks" + expandable Projects sections. Click any project card to expand task list. All Tasks aggregates across all active/recent sessions.
• New Per-project metrics row: visitors/24h, CVR, MRR, ad revenue — pulled from GROWTH.md and pushed at session end. Benchmark delta (↑/↓ % vs previous session) computed server-side.
• Improved Session push now includes full metrics snapshot for trend tracking across sessions.

Tighter execution engine. Self-evolution process redesigned to eliminate stalls.

• Improved Brain compression: 1059 → 970 lines (8% reduction). All functionality preserved — verbose examples, redundant code blocks, and long reference lists replaced with compact equivalents.
• Improved Evolution anti-stall: structured gap check (version sync, line count, agent count, dashboard TODOs) before every session. Strict stop condition — no specific target found = clean stop after 1 iteration, not 3+ retry cycles.

Unified CLI ↔ Dashboard experience. Real-time task progress. Self-learning execution engine.

• New Mistake Loop — tasks that fail due to execution errors are classified, and a prevention rule is written to PATTERNS.md. Loaded at session start. Never repeats the same mistake twice.
• New Live Dashboard tab — real-time task progress bar, active session view, stats, and elapsed time. CLI pushes updates during execution. 5-second polling when tab is visible.
• New Smart Reprioritization — completed tasks boost unblocked dependents, blocked tasks cascade, human-completed tasks jump queue, revenue tasks auto-promote to P0. Score decay for stale tasks.
• New Session Chains — sessions track source (cli/desktop/web/scheduled/remote), chain_id, machine hostname, brain version, parent session. Dashboard shows source icons and chain badges.
• Improved Dashboard: drag-and-drop task reordering, project card click-through, search + sort for 200+ projects, archive objectives, re-run command copy on sessions.
• Improved Auto-resolve deploy — GitHub PR merges and Vercel secret setup no longer require human action. AcePilot uses GitHub API directly.
• Improved DNS switch from Vercel (static-only) to Hostinger (PHP + static). Dashboard login and all APIs now functional.

Five rapid iterations improving the platform-blocked experience end-to-end.

• New Human Ops Dashboard tab — platform-blocked tasks as expandable cards with platform badge, time estimate, interactive step checklist, "Open Platform ↗" button, Mark Done / Dismiss.
• New Dashboard sync from brain — when AcePilot generates a platform-blocked guide and a license key exists, it silently POSTs the full guide to acepilot.app/api/actions.php for follow-along on phone or second screen.
• New Social platform templates — Reddit, HN Show HN, Twitter/X thread — exact step-by-step with copy-paste content. HN karma gate gets a concrete path with ETA.
• Improved Guide voice — guides now sound like a knowledgeable collaborator. "Reddit blocks automated posts — here's how to do it yourself in 2 minutes", not "Blocked by Chrome MCP safety restrictions".
• Improved APS scale fix — brain and researcher now use the same scale: Impact 1-5, Urgency 1-5, Unblock 0-3, Confidence 0.5/0.8/1.0, Effort 1-3, Range 0.3–13.0.
• Improved god/auto no-objective handling — bare /acepilot god with no argument now checks TASKS.md → executes; if empty → scans with @researcher → executes. Never stalls.
• Improved Dashboard nav link added to main site and all 18 sub-pages (compare, use-cases, blog).

AcePilot tells you exactly what to do when it can't do it itself. Stripe blocked? HN submission? DNS registrar? Precise step-by-step guides instead of BLOCKED. Pipeline keeps going.

• New Human Handoff: 5 blockage classes with structured step-by-step guides (exact URLs, buttons, values).
• New Guide mode: /acepilot guide — interactive checklist of pending human actions. Free tier.
• New HUMAN status [👤]: actionable tasks with guides instead of terminal BLOCKED.
• New Auto-verify: test webhooks, DNS lookups, curl checks after human completes action.
• Improved 55 self-improvement iterations: APS scales, graceful stop, deploy troubleshooting, verifiable criteria.

Smarter about what it reads, what it reports, and what you ask it to do.

• New Lazy ABSORB: two-phase state loading — ~40% fewer tokens on startup for modification sessions.
• New Stats mode: /acepilot stats — human-readable performance insights from execution history.
• New Multi-objective: "fix auth AND add dark mode" parsed into sequential task groups.
• New Error taxonomy: 6 error classes with auto-recovery strategies in circuit breaker.

Leaner brain, sharper agents. Same capabilities in 28% fewer tokens.

• Improved Brain compressed 959 to 692 lines. Faster response, lower cost, same capability.
• Improved @designer: responsive breakpoint checks at 375/768/1024px, dark mode awareness.
• Improved @security: client-side secret leakage detection, sensitive data logging scan.
• Improved @architect: cache invalidation, memory leak, missing timeout detection.
• Improved @strategist: clarified cold-start threshold, B2B annual-only pricing support.
• Improved @reviewer: test quality checks — assertion quality, edge cases, over-mocking.
• Fixed install.sh: license key validation (min 8 chars, whitespace handling).

The last version before public launch. Everything about making first contact with real users successful.

• feat Trial mode — 3 free Pro sessions (auto, ship, god) without a license key. Experience the best mode before deciding to pay.
• feat Smart first run — detects first session, skips empty-file reads, fast project scan, instant onboarding in <10 seconds.
• feat Deploy pipeline — GitHub Actions CI/CD awareness, PR-based deployment flow, static site optimization (zero build minutes).
• fix README install paths — all paths were stuck on v10.0, now correctly reference v12.0.
• fix install.sh — post-install message was showing v10.0 features instead of current version.

Every product needs users. Auto-generates acquisition infrastructure alongside the product — SEO, comparison pages, viral loops, blog scaffold, analytics, launch assets. Research-backed.

• feat Growth Engine — auto-generates SEO, comparison pages, blog scaffold, badge system, share mechanics, email capture, and analytics for every build.
• feat Launch Sequence — god --launch generates Show HN, Twitter threads, Reddit posts, pre-launch checklist.
• feat Content Engine — 5 blog post templates (origin story, tutorial, comparison, deep dive, changelog) with content calendar.
• feat GROWTH.md — 14th state file. Channels, funnel, content performance, viral metrics.
• improve Conversion framework — sticky CTAs, exit-intent capture, pricing optimization applied to every generated landing page.

The engine that improves itself. Playbooks capture proven workflows. Self-calibration tunes gates and routing from data. Session chains persist objectives across sessions. Deploy pipeline automates the last mile.

• feat Playbook system — successful workflows captured as replayable recipes. Second time is always faster than first.
• feat Self-calibrating engine — gates, specialist routing, and ceremony tiers auto-adjust from execution data every 10 sessions.
• feat Session chains — objectives persist across sessions with handoff state. Next session resumes with momentum.
• feat Deploy pipeline — auto-detects Vercel, Netlify, Docker, GitHub Pages, FTP. Builds, deploys, verifies in one step.
• feat PLAYBOOKS.md — 13th state file. Proven execution sequences stored and replayed.
• improve @researcher gains PLAYBOOK MODE for workflow capture. @strategist gains session chain and deploy verification lenses.
• improve ANALYTICS.md expanded with Calibration Log and Chain Log sections.

Say “website” and ship one. Directive Expansion Engine turns minimal input into full business specifications. Revenue-first decisions across tech, design, pricing, and conversion.

• feat Directive Expansion Engine — one word becomes a full business spec with revenue model, tech stack, and conversion strategy.
• feat IDENTITY.md — 12th state file. Project identity persists across sessions with source annotations.
• feat Business Decision Framework — pricing psychology, conversion defaults, copy rules, analytics, and SEO built in.

Four feedback loops close the learning gap. Every gate decision, specialist call, and cycle time feeds back into the next session.

• feat ANALYTICS.md — 10th state file with five feedback sections.
• feat Data-informed smart dispatch — historical precision overrides heuristic routing.
• feat Gate calibration from historical accuracy. Cycle time as north star metric.

Smart specialist dispatch routes by diff content. Micro ceremony tier for trivial changes. PATTERNS.md learns from execution history.

• feat Smart dispatch — ~60% fewer specialist calls, same quality.
• feat Micro tier — sub-30s execution for trivial changes.
• feat PATTERNS.md — 9th state file. Execution learning across sessions.

Every prompt line audited for ROI. 1039 to 544 lines across 7 files. Same execution quality, 48% fewer tokens burned.

• improve All 6 agent prompts compressed — verbose domain knowledge replaced with concise references. Models already know these frameworks.
• improve acepilot.md modes compressed to table format. Research Foundation docs removed (45 lines of citations never used for execution).
• improve God mode per-task review cost: ~500 to ~200 Sonnet tokens per cycle.

Designer agent and main command gain anti-cliche awareness for public-facing pages.

• feat @designer anti-cliche filter — flags generic SaaS patterns, AI-tool cliches, weak copy, missing conversion signals, and visual sameness.
• feat COMPOUND anti-mediocrity rules for landing page work. Grounded in Stripe, Linear, Cursor, CXL, Baymard, Copyhackers conversion research.

Auto mode now accepts objectives — give it a goal and it strategizes before executing.

• feat Objective-driven auto mode — Working Backwards + strategy before execution
• improve Consistent objective support across plan, auto, and god modes

14 modes consolidated to 8. Clear trust ladder from plan to god.

• feat God mode — all 6 specialists review every task, zero interruptions
• feat Ship mode — handle it and push a PR in one command
• improve 8 modes total (down from 14) — cleaner, clearer escalation
• improve Mid-task re-orient — smarter recovery when blocked

Six domain experts — design, security, architecture, strategy, and more — automatically review everything you ship.

• feat Dream Team mode — parallel specialist agents working together on complex tasks
• feat Overview — see status across all your projects in one view
• feat Web dashboard — track sessions, set objectives, manage subscription
• feat Action items — tasks requiring your input appear in the dashboard automatically
• improve Pro subscription with instant activation — 60 seconds from payment to working

Major quality and speed improvements. AcePilot gets smarter with every session you run.

• feat Learns your codebase and preferences — each session is faster and needs fewer clarifications
• feat Progress estimation — shows remaining tasks and estimated time
• feat Never makes the same mistake twice — remembers error patterns across sessions
• improve Smarter error recovery — tries alternative approaches before giving up

Simple fixes complete in under 60 seconds. Quality checks still run — just faster.

• feat Quick tasks (1-2 files) now complete dramatically faster
• feat Plan mode — see the strategy before committing to execution
• improve Smarter verification — skips unnecessary checks for simple changes

Fundamentally better decision-making. AcePilot now understands your code before changing it, and makes safer autonomous choices.

• feat Deep project understanding before every task — catches wrong direction early
• feat Safer autonomous decisions — risky changes require confirmation, safe changes execute instantly
• feat Cleaner output — eliminates unnecessary code before writing new code
• feat Three-stage quality check — correctness first, then code quality, then performance
• feat Auto-skips tasks that become irrelevant as work progresses
• feat CEO mode — give a business objective, get a PR
• feat Cycle time tracking in every session summary

AcePilot goes public. Landing page, branding, and the first installable release.

• feat acepilot.app — landing page with install instructions and pricing
• feat 30-second install — one command, zero dependencies
• feat 11 execution modes from fully interactive to fully autonomous

50+ iterations building the core engine. From a single prompt to a complete autonomous execution system.

• feat Autonomous scan-to-PR pipeline — finds tasks, prioritizes, executes, commits
• feat Smart priority scoring — tasks ranked by impact, urgency, and confidence
• feat Parallel execution — multiple tasks run simultaneously when safe
• feat Project memory — AcePilot remembers your codebase across sessions
• feat Operator memory — learns how you work and adapts to your style
• feat Safety systems — automatic rollback, error recovery, never commits to main
• fix 30+ correctness fixes across the engine