Changelog

What's new,
release by release.

Every etyb-skills release, the why behind it, and where to find the full technical history. Versions follow SemVer; dates are release dates.

Full technical changelog → GitHub releases →
  1. v5.0.0 Latest

    Plugin-native. Sub-agents do the work. One entry point.

    The Claude Code plugin ships for real: five sub-agents that carry out the heavy work in their own contexts, model-tiered so light work runs cheap and never above your own model, plus five advisory hooks that warn without ever blocking. /etyb stays the only trigger surface — the three role skills prototyped pre-release were withdrawn before tagging in favor of running that same work through the sub-agents. The anthropic-claude Stack is refreshed to the Claude 5 generation, and the vendor manifest now covers 538 pages across all 13 Stacks.

    • Five sub-agents (explorer, planner, reviewer, stack-researcher, cartographer), model-tiered — light work (search-and-cite, fetch-and-distill) pins to the smallest model tier; extensive-reasoning work inherits your own session model, which is always the ceiling. Dispatched into their own contexts so your session stays clean.
    • Five advisory hooks ship via the Claude Code plugin (hooks/hooks.json) — no test file before an edit, no review evidence before a commit, merging without a green test run. They warn; they never block. Verified firing against a real local plugin install before release.
    • Nothing to download or clone. /plugin marketplace add + /plugin install pulls the whole thing — skill, agents, and hooks — straight from GitHub. npx skills add remains the universal path for every other harness.
    • One entry point, enforced. /etyb is the only skill call, explicit or auto-triggered — no peer commands. The pre-release role skills were removed before tagging so this is structural, not just documented.
    • Vendor knowledge grows to 538 currency-stamped pages across 13 Stacks; anthropic-claude refreshed to the Claude 5 generation (Sonnet 5, Opus 4.8, Haiku 4.5, the Fable/Mythos tier) and spot-checked against live vendor docs at release time. Per-page currency CI now hard-fails on drift instead of warning.
    Release notes on GitHub → #
  2. v4.0.0

    One skill. One repo. 12 more vendor Stacks.

    The architecture collapses. v4 ships as a single coordinated skill — /etyb — with 14 specialists, 9 always-on protocols, 6 vertical architects, and 13 vendor Stacks all inside, read from disk at runtime. The v3 sibling-skill model and per-tier bundles are gone. Vendor knowledge moves into the same repo (no docs site to drift against). Everything that was three things on the homepage is now one install.

    • Single trigger surface: /etyb is the only slash command. The 14 specialists, 9 protocols, and 6 verticals become internal references inside skills/etyb/references/ — no peer skills competing at trigger time.
    • 12 new vendor Stacks ship: AWS, GCP, Azure, Cloudflare, Vercel, Supabase, Firebase, Stripe, Anthropic Claude, OpenAI, Expo, Observability — joining Salesforce. Each Stack carries a slim SKILL.md trigger pointer plus sibling index.md, per-product canonical pages, and per-role composed views, all currency-stamped per page.
    • One repo, no docs site. Vendor knowledge moves directly into stacks/<vendor>/ in etyb-skills — read by /etyb from disk, fetched by third-party agents from raw.githubusercontent.com. The brief detour through a separate docs site is undone.
    • Tiers removed. v3.x bundles (full, process-protocols, core-team, verticals) and the --bundle / --skills install flags are gone. Install always copies the full skill set; install.sh detects v3 sibling skills and pre-v4 hook paths and offers to back them up.
    • Senior CTO voice. /etyb never names internal protocols, specialists, or routing lanes in user-facing prose — routing is disclosed only in the signature footer (ETYB · backend-architect). Acknowledge → clarify → offer → execute pattern is the new first-response shape.
    • Knowledge currency contract. Every Stack page carries last_verified_on + drift_risk + authoritative_url frontmatter. Soft path discloses currency to the user; strict path defers to the vendor MCP/skill or fetches the authoritative URL for high-stakes claims.
    Release notes on GitHub → #
  3. v3.0.0

    Stack Packs are here. Salesforce is live.

    A new artifact type lands alongside the specialists and protocols. Stack Packs are platform-specific knowledge overlays that load across the team when work involves a specific stack. The first pack ships full Salesforce coverage current to Spring '26 — capturing the 2025–2026 reset (Agentforce, Data 360, MCP-native dev, ECA migration, MFA mandate) that a 2024-cutoff LLM gets wrong.

    • Salesforce Stack Pack with 11 role overlays — system, backend, frontend, AI/ML, database, devops, security, QA, SaaS, healthcare, fintech — currency-stamped to Spring '26, Dreamforce '25, TrailblazerDX 2026.
    • The 2025–2026 platform reset is captured: Einstein Copilot → Agentforce, Data Cloud → Data 360, Apex Cursors, LWC TypeScript types, Salesforce-Hosted MCP Servers, Headless 360, External Client Apps mandate (May 11 2026), MFA mandate (June–August 2026).
    • New top-level concept: specialists + protocols + Stack Packs. Three orthogonal axes. The team doesn't grow when stacks grow — adding a pack adds a folder, not a roster slot.
    • Roadmap: AWS, GCP, Azure, Vercel, Supabase, Snowflake, Databricks, Stripe, Shopify, SAP, ServiceNow — same pattern, same install.
    • Single-version semantics across the bundle: every skill, every stack, every artifact tracks the bundle version. Validator extended to enforce — no per-skill version drift.
    Release notes on GitHub → #
  4. v2.2.0

    Install anywhere. Pick your team.

    Bundle-aware installs reach every platform — not just Claude Code's native marketplace. One installer, four shapes. Plus a hardened release pipeline that catches defects before they ship.

    • Install just the protocols, just the core team, or just your vertical — the same bundles that ship on the Claude plugin marketplace are now a single command on every platform.
    • One source of truth: bundle definitions live in one manifest; CI refuses to merge if any surface drifts out of sync.
    • ShellCheck with zero exclusions across every shell script — three latent defects surfaced and fixed along the way.
    • A log-injection bug in the plan-execution edit log is fixed and regression-tested.
    • All version fields and skill counts agree across the repo, the marketplace, and the plugin manifest.
    Release notes on GitHub → #
  5. v2.1.0

    Codex becomes a first-class citizen.

    Real runtime support for OpenAI Codex — lifecycle hooks, custom agents, and per-skill metadata. Gates that were model-trusted are now partial runtime-enforced.

    • 4 Codex lifecycle hooks guard prompts, Bash, and stop-time claims so the team cannot reason its way past a gate.
    • 4 Codex custom agents — explorer, planner, reviewer, docs researcher — bring parallel dispatch to the Codex runtime.
    • Every one of the 30 skills ships with Codex-native discovery metadata.
    • A portability linter enforces cross-platform parity before release.
    Release notes on GitHub → #
  6. v2.0.0

    Portable across every platform.

    The cross-platform release. One virtual engineering team, three runtimes: Claude Code, OpenAI Codex, and Google Antigravity. Transparent adapters, no silent divergence.

    • Platform adapters — Claude flagship (hook-enforced), Codex (hooks + agents), Antigravity (model-trusted with optional ADK).
    • The orchestrator is now simply `etyb` — one name across brand, repo, marketplace, and invocation.
    • Verification and debugging graduate into standalone peer skills (28 installable skills → 30).
    • Every specialist works on its own — uninstall etyb and the team still functions as focused experts.
    Release notes on GitHub → #
  7. v1.0.0

    First public release.

    The virtual engineering company goes public — one CTO-level orchestrator, 14 core teams, 6 domain specialists, and 7 always-on process protocols, all MIT-licensed.

    • 28 installable AI agent skills covering the full SDLC.
    • TDD, review, and verification protocols always on — no opt-in, no escape hatch.
    • Open source from day one at github.com/e-t-y-b/etyb-skills.
    Release notes on GitHub → #
Stay current

Ship what's new.

Re-run the same install command to pick up the latest release — everything else stays as it was. Plans, local settings, and backups are preserved.

Install etyb All GitHub releases