Skip to content

Changelog

All notable changes to this project will be documented in this file.

[1.1.0] — 2026-11-11

✅ Executable routing tests

  • New deterministic router (mega_mind/router.py) over the shipped data — no model call, no network. It implements the precedence the orchestrator only documented: explicit persona invocation → named test type over generic write tests → debugging-first for unknown-cause failures → longest routing phrase → unambiguous trigger fallback → chain selection.
  • New mmo route "<request>" command (--json for machine output, exit 1 when unrouted) — the executable counterpart of the /mega-mind route the docs already referenced.
  • New tests/test_router.py (42 tests) proves 78/78 curated phrases reach their documented target and layer, 53/53 skills and 11/11 agents are reachable, the eight documented agent-vs-skill collisions resolve to the right layer, the six docs/routing-tests.md fixtures behave as recorded, and an installed tree is routable. Mutation-checked: removing the test-type table, the failure signals or the persona table each flips its covering assertion.

🐞 Defects the tests found and closed

  • Whole-chain requests returned unrouted: "build a CSV export feature end-to-end" matched no phrase. It now selects the chain and starts at its head (search-first).
  • All three of mega-mind's published examples matched nothing (OAuth feature, "Users are randomly getting logged out", "Make the app faster"). They now route as documented via feature verb+noun, session-failure and ambiguous-performance rules.
  • CLI crashed on non-UTF-8 consoles: mmo route printed → and mmo init prints emoji, so both raised UnicodeEncodeError on a cp1251 console after doing their work. Streams are reconfigured to UTF-8 with replacement at startup, route output is ASCII, and a subprocess test pins it at PYTHONIOENCODING=cp1251.
  • Validator hardening: routes are kind-aware (a kind: skill route can no longer target an agent), every skill-routed phrase is checked against its target's frontmatter triggers, and a skill with an unparseable changelog table no longer passes silently.

🔗 Routing layer (same audit)

  • mega-mind 2.1.5: the precedence rules now have an executable implementation documented beside the routing matrix.
  • mega-mind 2.1.4: CLASSIFY covers the five agent routes that were matrix-only and points whole-chain selection at the Workflow Chain Selection Guide (four chain-start personas were otherwise unreachable); k8s deploy ownership split from generic CI/CD; search-first scoped to non-trivial work, resolving its contradiction with the over-orchestration anti-pattern; Example 1 aligned with the canonical Z-Pattern chain.

[1.0.5] — 2026-10-14

🧪 Agent Hardening (all 11 personas)

  • Every agent prompt-tested against its own documented rules, then hardened where the audit found gaps: adversarial-tester gained written approval, environment isolation, abort/rollback and rate ceilings (and lost an unsound "a harness OOM proves a vulnerability" rule); architect gained cost/ops axes, an override rule and Deciders plus Migration/Rollout ADR fields; tech-lead gained delegation and alignment rules and lost three stale api-designer links; release-manager gained a 7-row go/no-go gate; security-reviewer gained rotate-then-purge and a ban on self-approving its own fixes; data-privacy-officer gained DSAR and lawful-basis sections; incident-commander gained a communications cadence and 6-hour handover; planner gained unknowns/buffers and critical-path/milestone rules; qa-engineer gained a scope boundary and exit criteria; code-reviewer gained verdict/ownership discipline; accessibility-auditor was tested and passed as-found.
  • New validator check agent_refs: hyphenated cross-references inside agent personas must resolve to a real skill, agent or workflow — the class of rot that let a stale api-designer link survive several releases. The validator now runs 11 checks.
  • sync-assets.py ordering fixed: it built the manifest before rendering, so the renderer rewrote mega-mind after its hash was recorded and the copied manifest was stale (this broke mmo check and two installer tests). Now: build → render → rebuild → copy.

🔗 Routing Layer Consistency

  • All 37 matrix phrases are now frontmatter triggers (31 skills bumped), so host-level skill selection reaches the same target the orchestrator matrix does.
  • Three skills gained matrix routes (brainstorming, writing-plans, finishing-a-development-branch) — routing.json now holds 57 routes — and mega-mind lists all nine named workflows beside execute <workflow>.

🔧 Routing Fallback Precision

  • CLASSIFY precedence rules (mega-mind 2.1.2): unknown-root-cause failure → debugging before platform skills; named test type beats generic write tests; DevOps split into provisioning / container build / cluster-workload failure. Matching note fields added to the affected routes in routing.json.
  • New routing fixture: docs/routing-tests.md — six prompt-level cases with expected routes and a per-release results table.

[1.0.4] — 2026-10-14

🔧 Skill Routing & EOL Fixes

  • Trigger collisions eliminated: 11 colliding trigger phrases removed across 10 skills (each resolved in favor of the routing matrix's owner; tech-lead gained design system); 13 skills version-bumped (2.0.0 → 2.0.1; mega-mind → 2.1.1) with dated changelog rows. Validator check trigger_collision promoted from WARNING to ERROR.
  • Instinct chain install-safe: mega-mind, skill-generator, and continuous-learning-v2 no longer reference instinct/eval directories that don't ship — reads are conditional, first use bootstraps with mkdir -p.
  • LF everywhere, enforced: new .gitattributes (* text=auto eol=lf); 116 files converted to LF; new validator check eol fails on any tracked CRLF; all repo writers patched with newline="\n" so regeneration can't reintroduce CRLF; .gitignore covers the synced assets/evals/ mirror; validate.yml watches .gitattributes.
  • Version bump: package 1.0.3 → 1.0.4; README/USAGE/AGENTS/docs-site updated to v1.0.4 (docs titles had been stale since v1.0.2).

[1.0.3] — 2026-08-11

🔧 Fix: PyPI Metadata Summary Over Limit

  • The pyproject.toml description (561 chars) exceeded PyPI's 512-character summary limit, rejecting every upload since v1.0.0 with 400 'summary' field must be 512 characters or less. Shortened to a compliant summary; full details live in the README.

[1.0.2] — 2026-08-11

🔧 Hotfix: Platform-Independent Content Hashes

  • Content hashes in skills-manifest.json are now computed over line-ending-normalized bytes (CRLF ≡ LF), so mmo check, mmo doctor, and the validate.yml CI gate produce identical results on Windows and Linux checkouts.

[1.0.1] — 2026-08-11

🔧 Routing & Validation Hardening

  • Single source of truth: scripts/build-manifest.py generates .agent/shared/skills-manifest.json (inventory + SHA-256 hashes + 12-section compliance). Validator, renderer, installer, and docs all consume it — counts and membership can no longer drift.
  • Machine-checkable routing: .agent/shared/routing.json (54 routes, 11 chains) is rendered into the routing matrix and /mega-mind skills listing by scripts/render-skills.py inside <!-- GENERATED --> fences; drift fails CI.
  • Manifest-driven validator: Rewrote scripts/validate-skill-system.py — the old one flagged 51/53 skills falsely; the new one checks manifest freshness, frontmatter, 12/12 sections, routing coverage, chains, trigger collisions, and version attestation.
  • Hash-verified sync: scripts/sync-assets.py regenerates before copying and verifies .agent/ ↔ assets/ byte-for-byte.
  • mmo doctor / mmo check: environment + install-tree verification, and a repository-wide gate ending in a sandboxed install of all 6 platform layouts.
  • Verified installs: full skill-directory copies, byte-for-byte copy verification, structural agent checks.
  • Tests + CI: tests/ (12 tests) and validate.yml on ubuntu + windows.
  • Version & spec alignment: 1.0.1 with metadata-derived __version__; Gold Standard spec references unified to v2.0; skill frontmatter versions corrected to 2.0.0 (matching their own changelogs — validator-enforced); category counts corrected to 12/29/8/4; stale routing names fixed.
  • Generated docs: README catalog, reference.md tables, and index.md category table are generated regions.

[1.0.0] — 2026-07-09

🏆 Gold Standard v1.0 — All 53 Skills Upgraded

Every skill upgraded to the Gold Standard SKILL.md v1.0 specification — 12 required sections: Blocking Violations, Verification with quality gates, Performance & Cost models, structured Examples, Anti-Patterns tables, and full References/Changelog. This is the largest quality upgrade in the project's history.

  • Created .agent/shared/GOLD-STANDARD-SKILL.md — universal 12-section template with scoring rubric
  • All 53 skills now at 12/12 gold standard sections (100% compliance, 19,531 total lines)
  • Fixed RTK skill Redux anti-patterns (critical copy-paste bug)
  • Fixed broken rtk bun test (or npm test) syntax across all skills
  • Added Blocking Violations, Performance & Cost, References to every skill
  • Enhanced mega-mind orchestrator to 671 lines with workflow chain selection guide
  • Enhanced security-reviewer with OWASP Top 10 (2025), CWE mappings, STRIDE threat modeling
  • Updated AGENTS.md, README.md, USAGE.md, and pyproject.toml for v1.0

[0.8.0] — 2026-05-23

🏗️ Full Lifecycle Coverage

  • 5 new agent personas: incident-commander (production incidents), release-manager (versioning/rollouts), accessibility-auditor (WCAG compliance), adversarial-tester (chaos/fuzz testing), data-privacy-officer (GDPR/CCPA/SOC2)
  • 2 new workflows: incident-response (full incident lifecycle), release (version→changelog→rollout→monitor)
  • 5 new workflow chains: Incident Response, Release, Accessibility Audit, Adversarial Test, Compliance Review — all registered in the mega-mind routing matrix
  • Agent personas expanded from 6 → 11, workflows from 7 → 9

🔧 Structural Fixes (14 items)

# Issue Resolution
1 task.md single-point-of-failure Backup script + stub + workflow reference
2 Routing matrix drift Validator enforces coverage; all 53 skills + 11 agents covered
3 Skill count inconsistency Reconciled to 53 across AGENTS.md, disk, routing matrix
4 Cross-skill vocabulary drift 18 headings normalized; verification phases unified
5 Z-Pattern rigidity 7 decomposition patterns added (API-First, ML, IaC, etc.)
6 Grep-fragile checklists 115 patterns fixed across 34 skills (-E flag, POSIX-compliant)
7 Agent/skill boundary blurred Routing note + code-reviewer/qa-engineer added to matrix
8 No formal handoff interface Handoff block template + 3 key skills updated
9 Instinct decay mechanism last_validated/review_by fields + decay table + conflict detection
10 search-first after routing Moved to step 2 in Request Analysis
11 RTK hard dependency Fallback protocol with detection + bare commands
12 No skill regression tests Validator script + chain integrity eval
13 Flat verification depth Tiered verification (Tier 1 Surface / Tier 2 Standard / Tier 3 Deep)
14 Claude-specific idioms Runtime-agnostic terms + compatibility section in multi-execute

🌐 Universal Compatibility

  • Unified compatibility: frontmatter across all 53 skills, 11 agents, and 9 workflows to: Any AI coding agent (Antigravity, Claude Code, Copilot, Cursor, OpenCode, Codex, pi, and all tools supporting the Agent Skills open standard)
  • Updated AGENTS.md, README.md, USAGE.md, and pyproject.toml descriptions to reflect universal compatibility

📦 Distribution

  • Synced all .agent/ changes to src/mega_mind/assets/ for correct package distribution
  • Created scripts/validate-skill-system.py — automated validator (skill count, routing, headings, required sections, grep portability)
  • Created scripts/backup-task-state.sh — timestamped task.md backups
  • Created scripts/sync-assets.py — one-command sync from .agent/ to src/mega_mind/assets/
  • Fixed Unicode encoding error in CLI (✅/❌ → [OK]/[ERROR]) for Windows console compatibility

[0.7.0] — 2026-04-30

🔧 Orchestration Fixes

  • Fixed 23 stale skill references across the entire skill system (systematic-debugging → debugging, verification-before-completion → verification-loop, bug-hunter → debugging, api-designer → backend-architect, using-mega-mind → mega-mind)
  • Reconciled skill counts in AGENTS.md — now correctly shows 9 Core + 30 Domain Expert + 12 Meta + 2 Token = 53
  • Fixed phase count inconsistency in execute-plan and high-complexity-dev workflows (6 → 10 phases matching verification-loop)
  • Added missing autoresearch-loop to Meta & Learning skills list
  • Added missing regex-vs-llm-structured-text to Domain Expert skills
  • Recategorized planner and architect as agent personas (not skills) in AGENTS.md routing matrix
  • Completed ship workflow chain with continuous-learning-v2

🧩 New Platform: pi-coding-agent

  • Added --pi flag to CLI and installer
  • Installs to .pi/skills/, .pi/prompts/, .pi/agents/, and .agents/skills/ (cross-tool Agent Skills standard)
  • Full documentation in README.md and USAGE.md with platform-specific notes

📦 Distribution

  • Synced all .agent/ fixes to src/mega_mind/assets/ for correct package distribution
  • Updated README.md: fixed skill counts, shared/ directory listing, removed phantom tests/ dir
  • Updated USAGE.md: Step 5 is now general usage guide for all platforms with command reference table

[0.6.0] — 2026-04-17

🔄 Skill Library Consolidation

  • Consolidated and enhanced the agent skill library — merged overlapping skills, removed stale ones
  • Enhanced agent definitions with structured frameworks and protocols
  • Synchronized workflow references with consolidated skill library
  • Updated skill listings post-consolidation across all documentation

📚 New Shared Operational Guides

  • Added DE-SLOPPIFY.md — code quality cleanup checklist
  • Added RTK_GUIDE.md — Rust Token Killer usage guide (60-90% token savings)
  • Added VERIFICATION-GATE.md — structured 6-phase verification checkpoint

📖 Documentation

  • Updated skill documentation and usage guide post-consolidation
  • Added .agent/evals to .gitignore for evaluation artifacts

[0.5.0] — 2026-04-05

🏷️ Rebrand to mmo

  • CLI entry point renamed to mmo (was mega-mind-orchestrator)
  • PyPI package name: mmo
  • Both mmo and mega-mind-orchestrator console scripts registered

🌐 Platform Support

  • Expanded platform flags for targeted installation
  • Full Claude Code, GitHub Copilot, OpenCode, and Codex compatibility

[0.4.0] — 2026-04-03

🪝 Context-Mode Hook Integration

  • Implemented context-mode hook system for all supported environments
  • Hooks for PreToolUse, PostToolUse, PreCompact, and SessionStart events
  • Added .agent/hooks/hooks.json generation

🛡️ Behavior Guardrails

  • Added session rules: no proactive commits, mandatory task tracking, search-first, de-sloppify, security by design
  • Added autoresearch rules: continuous-learning-v2 loop, self-eval before done
  • Enforced quality gates before any task marked complete

[0.3.2] — 2026-03-28

🔧 Fixes

  • Fixed installation commands to use mega-mind-orchestrator console script entry point

[0.3.1] — 2026-03-27

🔧 Fixes

  • Updated GitHub Actions workflow versions

[0.3.0] — 2026-03-27

🚀 Platform Installer

  • Added --claude flag for Claude Code-compatible installation
  • Added --copilot flag for GitHub Copilot-compatible installation
  • Skills copied to platform-specific directories with proper file naming

🧠 Mega-Mind Orchestration

  • Added structured workflow system via AGENTS.md with mega-mind orchestrator
  • Request routing matrix, skill chains, workflow definitions
  • Task tracking via docs/plans/task.md

🏷️ Metadata

  • Added compatibility frontmatter field to all skills for AI coding assistant targeting
  • Added --from option to uv tool install

[0.2.0] — 2026-03-25

🤖 GitHub Copilot Support

  • Added GitHub Copilot-compatible file structure to the installer
  • Targeted brainstorming gate for structured exploration

[0.1.1] — 2026-03-24

🔧 Fixes

  • Added --version option to CLI with proper package name

[0.1.0] — 2026-03-24

🎉 Initial Release

  • Mega-Mind CLI tool for skill system installation
  • 53+ specialized AI coding assistant skills across 4 categories
  • mmo init CLI command for project initialization
  • PyPI package published as mega-mind-orchestrator
  • CI/CD pipeline for automated PyPI publishing
  • RTK token optimization integration