AIWG — June 2026 in review

AIWG — June 2026 in review

Covers releases 2026.6.0 → 2026.6.3 (2026-06-12 → 2026-06-20)

Starting this month, AIWG moves to a monthly announcement cadence. Point releases still ship through June as they're ready (and each has its own CHANGELOG entry), but instead of a separate write-up per patch, this single rollup covers everything that landed across the June cuts.

It was a big month. June added provider #11, shipped the **declarative Flow

  • cross-stack Mission** layer, turned the research framework into a real

corpus-intelligence toolkit, and — the through-line across three of the four cuts — stood up the AIWG Cockpit from empty substrate to a control plane you can actually drive against a live agentic-sandbox. Alongside it: a fleet-wide agent debloat that put every definition under the dispatch ceiling, and a documentation-accuracy pass that reconciled every skill and agent count against source.

Install or upgrade:

npm install -g aiwg          # fresh install (stable)
aiwg refresh                 # update an existing install + redeploy

The headline: AIWG Cockpit — groundwork → usable operator surface

The AIWG Cockpit is a UX-first control plane that sits on top of the AIWG CLI and the per-provider agentic stacks. It lets you manage an install, deployed agents, and live multi-stack sessions from one surface — without ever replacing or weakening the CLI underneath. June took it from nothing to a usable surface in three steps:

Groundwork (2026.6.1). The substrate landed behind an opt-in `@aiwg/cockpit` package (epic #1588):

  • A data-driven core bound to the extension / `aiwg discover` / artifact-index

registry — capabilities are derived from the registry, not hardcoded.

  • Instance control normalized on the agentic-sandbox A2A interface, with a

live session view and a running-agents board.

  • Two shells over one shared core — a Tauri (Rust) desktop app and a VS Code

extension (#1594).

  • A declarative UI contribution model — extensions contribute screens,

actions, workflows, and event-hooks (#1591).

  • Strong local control-surface auth — `127.0.0.1` + Origin + CSRF + per-launch

token + OS-keychain handshake (#1595).

  • A base-footprint CI guard (#1593) so the opt-in package never bloats the base

`npm i -g aiwg` install.

Live proof (2026.6.2). The release gate got stricter: a provider workload now has to emit both `AIWG_COCKPIT_LIVE_OK` and a real AIWG discovery result from inside a managed agentic-sandbox session — proving real discovery, not just shell plumbing. The 2026-06-18 host proof passed with Codex (`codex exec -s read-only`) in a managed `tmux` session.

Usable surface (2026.6.3). The operator UX caught up to the substrate:

  • The running board derives from real A2A tasks (#1639), and **Home stays

usable against a real v2 executor** — degrading the running/approvals panes instead of collapsing to "No stack connected" (#1638).

  • Starting a session is one explicit picker (#1640, #1641): instance ·

runtime · loadout · backend · posture → confirm → attach. It surfaces the full loadout catalog via a new Bridge `/api/loadouts` passthrough, guards against silently replacing an attached session, and shows failures inline — retiring the param-less start that read as a no-op.

  • The attach view is a real terminal — the PTY renders through xterm.js, so

ANSI/VT/tmux redraws are interpreted, not dumped as raw escape bytes.

  • One command brings up the whole stack — a canonical dev bring-up harness

replaces the ad-hoc `/tmp` rigs, with documented test stages and a contract guard pinning the mock's admin surface to the real v2 divergence.

  • Smaller correctness fixes: capability injection uses the plain skill name

(discover-first resolves it), not a `/`-prefixed pseudo-command (#1642); and `/api/show` resolves by the discovered path, so same-named artifacts (e.g. two `aiwg-steward`) return a 4xx instead of a 502 (#1643).

There's a chat-style agent view on deck, too: an ADR (#1645) proposes a per-session Terminal↔Chat toggle with a normalized output-parsing contract, structured-stream-preferred and PTY-parser-fallback, Claude Code first.

Status: Cockpit is opt-in beta. It is not installed by `npm i -g aiwg`;

the base CLI footprint is unchanged. Install `@aiwg/cockpit` only if you want

to drive the foundation as it firms up.

npm install -g @aiwg/cockpit   # opt-in; not part of the base CLI

OpenHuman is provider #11

`aiwg use <framework> --provider openhuman` deploys AIWG to tinyhumansai/openhuman, an OSS personal-AI runtime (epic #1552). Kernel skills install globally to `~/.openhuman/skills/` — the ungated user-scope root OpenHuman's own installer writes to — personas land in the workspace `.agents/agents/` for the coding hosts OpenHuman drives, and commands/rules aggregate into an `AGENTS.md` bridge. Verified live: OpenHuman's agent harness scans the deployed `SKILL.md` bundles and event-bus-wires the triggered ones.

That brings AIWG to 11 deployment providers from one source of truth. See the OpenHuman quickstart.

Releases and the CLI run on declarative Flows + cross-stack Missions

The `flow-*` migration to `flow.aiwg.io/v1` completed this month (#1539): `flow-release`, `flow-architecture-evolution`, and the rest are now declarative YAML Flows with an executor fan-out contract — discoverable directly via `aiwg discover`, runnable as orchestration with intra-step multi-agent panels + synthesis. The release gate sequence lives in `.aiwg/release.config`, so the Flow body is portable across CalVer/SemVer projects.

On top of Flows, the cross-stack Mission conductor landed (#1546): one conductor fans heterogeneous workers across stacks — a Claude session dispatching Codex subagents, for instance — over the `runtime:<name>` executor convention. `/aiwg-mission` ships AIWG-owned on Codex with no plugin dependency.

The research framework became a corpus-intelligence toolkit

June turned the research corpus from a document pile into something searchable and auditable:

  • Semantic + lexical search — `aiwg index query --semantic` (embeddings) and

`--fulltext` (BM25), plus `similar` and a `dedup-report` (#1493, #1494).

  • Citation-graph densification — `extract-crossrefs` + `citation-backfill`

(#1505), profile-graph edges + embedding similarity (#1501).

  • Integrity + provenance — corpus integrity / submission-risk scan (#1506),

sidecar lint & metadata repair (#1503), per-type induction-depth audit (#1504).

  • Vision extraction — provider-neutral scanned-page extraction (#1507).
  • Discoverability — an extensible source-type registry + by-source-type and

by-project / project-impact index views (#1495, #1509).

Plus a Fortemi index export — `aiwg index export --format fortemi` emits a browser-consumable, schema-validated project index for React integrations.

Leaner agents, and docs that match the code

  • Every agent definition is under the dispatch ceiling. A repo-wide debloat

(#1587, #1600) moved worked examples and restated rule boilerplate out of agent definitions into the discoverable example catalog, and added a hard 16 KB size ceiling enforced by `aiwg doctor`. Oversized definitions had been silently failing subagent dispatch with `Prompt is too long` at 0 tokens; that class of failure is now gone, with few-shot coverage preserved via the catalog.

  • Docs reconciled against source. A `doc-sync code-to-docs` pass corrected

every skill/agent count across the documentation — kernel skills to 20, addons to 29, per-framework counts fixed (research → 39, security-engineering → 27, sdlc agents → 93) — and resolved three doc-to-doc contradictions.

Cross-provider hardening & discovery

  • Discovery matches how you actually ask (#1581). Full natural-language

questions ("help me choose the right AIWG framework") normalize to keyword phrases and match; single-content-token queries no longer dead-end. `discover` is the exclusive agentic-capability surface and induces discover-first on new directives, not just on decline.

  • Cursor emits rules as native `.mdc` (and cleanup preserves operator `.mdc`).
  • OpenClaw gets the Steward identity in `SOUL.md`, correctly-aggregated

commands, and a discover-first bridge.

  • Warp's `WARP.md` is now a lean discover-first managed bridge (down from a

17.5k-line aggregate).

  • OpenCode loads `AIWG.md` directive classification via `instructions[]` and

wires deployed rules into `opencode.json`.

  • Codex ships AIWG-owned `/aiwg-mission` and loads `AIWG.md` via

`instructions[]`.

  • Public-facing output — the `aiwg use` version stamp, `aiwg diagnose` bug

link, and entrypoint messages now read `github.com/jmagly/aiwg`, sourced from `package.json` rather than the internal build origin.

The June cuts at a glance

ReleaseDateHeadline
2026.6.02026-06-12OpenHuman provider #11; declarative Flows + cross-stack Missions; research-corpus intelligence; cross-provider hardening
2026.6.12026-06-15AIWG Cockpit groundwork (beta); declarative-Flow migration complete; agent debloat under the 16 KB ceiling; doc-accuracy pass
2026.6.22026-06-18Cockpit live-proof hardening — the live UAT gate must prove real AIWG discovery from inside a session
2026.6.32026-06-20Cockpit groundwork → usable operator surface — running board from real A2A tasks, session-start picker, xterm.js terminal, one-command bring-up

Upgrade notes

  • No action required for the base CLI — all June changes are additive or

self-healing on the next `aiwg refresh` / `aiwg use`.

  • OpenHuman operators: `aiwg use sdlc --provider openhuman` installs kernel

skills globally to `~/.openhuman/skills/`; they load in OpenHuman's agent harness immediately. (The beta "Skills Explorer → Installed" tab reads `~/.openhuman/workflows/`, not `skills/`, so they won't render there yet — an upstream OpenHuman limitation tracked in #1560.)

  • Cockpit is the opt-in `@aiwg/cockpit` package; it is not installed by

`npm i -g aiwg`, and the base footprint is unchanged.

npm install -g aiwg