@zerwiz/ymir 0.1.27 → 0.1.28

package/docs/fixes/skills/unversioned-2026-09-17-midgard-holds-public-assets-only-the-company-wiki-is-gon.md

@@ -0,0 +1,23 @@
+## skills · unversioned · 2026-09-17 — midgard holds public assets only; the company wiki is gone
+
+### Why
+- **`midgard/company_wiki/` did not belong in this repo.** It was an empty
+  placeholder (one `.gitkeep`), but a company wiki is *tenant data* — policies,
+  vision, specs, client names — and this repo is public. Removed, with every
+  reference to it corrected in `midgard/README.md`, `STRUCTURE.md`, `docs/lore.md`
+  and `assets/Ymir.md`.
+- **`midgard/` is now described precisely.** It was called "shared company
+  assets" in one place and "cross-tenant" in another; a wiki is neither. The
+  contract says **public assets**, and `midgard/README.md` names what does *not*
+  belong: company knowledge, which lives at `$YMIR_HOME/hodd/identity/companies/`.
+- **`bin/private-guard.sh` gains a tenant-document rule** so the shape cannot
+  return: a `company_wiki/`, `wiki/`, `policies/`, `handbook/` or `vision`/
+  `strategy`/`roadmap` document staged in the public tree is refused, with a
+  message naming where it belongs. Verified by planting `midgard/company_wiki/policies.md`.
+- **Nothing had leaked.** A full sweep of every tracked midgard file for personal
+  handles, home paths, key shapes, tunnel UUIDs and LAN/tailnet addresses found
+  no hits — the ingress configs use `${VARS}` and `example.org` throughout, and
+  the DB schema is generic with row-level security on every tenant table.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/skills/unversioned-2026-09-17-the-forge-speaks-each-skill-s-own-text-in-the-index.md

@@ -0,0 +1,16 @@
+## skills · unversioned · 2026-09-17 — the Forge speaks: each skill's own text in the index
+
+### Why
+- **The Forge listed skills as bare names.** `Forge.tsx` rendered only
+  `{s.name}` + the aett rune, while the live `/api/skills` index carries every
+  skill's `description` (240-char truncation) — the data was there, the
+  surface never showed it. The skill list items are now two-line cards: name +
+  aett on the head row, the description beneath (`forge-item-desc`, dimmed
+  until hover), scoped to the skill variant so the Eindri rows keep their
+  side-by-side name+status shape.
+
+galdr-reread: `.agents/skills/galdr-ymirsystem/assets/hlidskjalf-ui.md` — the
+Forge gate bullet (skill list shows name + aett + description).
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/skills/unversioned-2026-09-17-the-local-git-round-joins-the-loom-the-stack-s-doors-are.md

@@ -0,0 +1,19 @@
+## skills · unversioned · 2026-09-17 — the local git round joins the loom; the stack's doors are in the registry
+
+### Why
+- **`bin/nornir-job-forgejo-git.sh` (02:45)** — reads the local forge's open
+  issues each night (`$YMIR_HOME/config/forge.env` names the door; token
+  optional), writes a daily digest to the hoard, carves Rune
+  `forgejo / git.issues` — or `git.door-down` (exit 1) when the tunnel is
+  closed, so a dead door is seen at sunrise. Unarmed without a URL, honestly.
+- **The marketing stack's doors are now in the registry** —
+  `marketing_doors[6]` (Mautic · Postiz · Activepieces · Forgejo · SearXNG ·
+  Grafana) with reach notes, mirrored to the Galdr asset. Agents (Bragi ·
+  Sindri) provision the stack for any user via
+  `bin/ymir-marketing-stack.sh`, or reach the Allfather's live server stack.
+
+galdr-reread: `.agents/skills/galdr-ymirsystem/assets/nornir-jobs.md` — §3.7
+(Forgejo round) and §3.8 (marketing stack provisioner); registry mirrored.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/skills/unversioned-2026-09-17-the-registry-says-where-a-repo-lives-and-what-it-is-for.md

@@ -0,0 +1,20 @@
+## skills · unversioned · 2026-09-17 — the registry says where a repo lives and what it is for
+
+### Why
+- **A project entry can now name its machine.** `machine:` records the host
+  where a checkout lives; `data/machines.md` holds the fleet (omarchy-1,
+  zerwiz, zerwizserver live on the tailnet). A repo is no longer assumed to be
+  on the box you happen to be standing on.
+- **A project entry can now say what it is for.** `about:` is one plain
+  sentence per repo. `bin/project-git.sh list --field about` prints the whole
+  inventory without opening the YAML.
+- **`project-git.sh` resolves more of the block.** `--field` now accepts
+  `machine`, `company`, `workspace` and `about` alongside the git fields, and
+  non-git keys read from the project block rather than the `git:` line.
+- **One registry, not two.** The live registry is `hodd/identity/projects.yaml`
+  (what the runtime reads). A stale duplicate at `Documents/Ymir/identity/` had
+  been drifting apart from it — different GitHub owners, different project sets
+  — and is retired, with a backup kept in `hodd/state/`.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-16-the-pane-bridge-keeps-its-quoting-phase-3-verified.md

@@ -0,0 +1,22 @@
+## smidja · unversioned · 2026-09-16 — the pane bridge keeps its quoting; Phase 3 verified
+
+### Why
+- **The Apodex plan's last open verification is closed.** Run in a herdr pane —
+  `bin/herdr-run.sh run huginn-research -- bin/huginn-research-worker.sh --brief
+  "…" --output-dir …` — the worker recalled, dispatched to the Apodex seat, wrote
+  its verdict, and observed it into Mimirsbrunn (`status: completed`, verdict
+  *"The capital of Norway is Oslo."*). Phase 3 needed a spawn, which the
+  read-only session lock had forbidden; it was run once the lock was this
+  session's.
+- **Defect found and mended — `bin/herdr-run.sh run` lost its quoting.** A pane
+  runs its command through a *shell*, but the bridge handed herdr the raw argv.
+  An argument containing spaces reached the pane as separate words, so
+  `--brief "In one sentence: …"` arrived as `--brief In one sentence: …` and the
+  worker died with `error: unknown arg: one`. Every real brief is multi-word, so
+  the documented invocation could never have worked. `run` now quotes each
+  argument on the way in (`printf '%q'`), verified by re-running it.
+- Asset updated in the same change: `.agents/skills/ymir-host/assets/thjazi.md`
+  carries the quoting rule.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-16-the-roster-can-finally-run-an-agent-on-pi.md

@@ -0,0 +1,23 @@
+## smidja · unversioned · 2026-09-16 — the roster can finally run an agent on Pi
+
+### Why
+`bin/agents-config.sh apply` wrote **every** agent's model into `opencode.json`'s
+agent block, whatever harness that agent used. So an agent set to run on **Pi**
+(native local models) would have had its Pi model id —
+`llamacpp/qwen3.5-9b` — written into OpenCode's config, which cannot resolve it.
+That is why the roster's two local agents were pinned to `harness: opencode` with
+opencode-style ids: there was no working way to put an agent on Pi.
+
+- **`apply` is now harness-aware.** Only `opencode`-harness agents are written
+  into `opencode.json`; a `pi` (or `hermes`) agent is left out, its model id going
+  to the resolve cache that `bin/agent-run.sh` reads. The providers block is
+  unchanged — a provider's endpoint is a real fact OpenCode may still want.
+- The way this is meant to be used: declare a local agent's model as its **Pi**
+  id (`llamacpp/qwen3.5-9b`, `llamacpp-coder/qwen3-coder-30b` — the ids `pi
+  --list-models` reports) with `harness: pi`; the roster is then Pi-driven with no
+  per-machine hand-editing.
+- `galdr-reread`: `assets/harness-integration/README.md` — the two writers of
+  `opencode.json`, and the rule that only OpenCode agents belong in it.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-16-the-roster-s-effect-lands-and-the-backend-pin-stops-bein.md

@@ -0,0 +1,19 @@
+## smidja · unversioned · 2026-09-16 — the roster's effect lands, and the backend pin stops being committable
+
+### Why
+Applying the private roster (`bin/agents-config.sh apply`) writes the resolved
+model into each agent's canonical profile. Flipping the two local agents to Pi
+made that visible — and surfaced a small ignore hole.
+
+- **Profiles follow the roster.** `sindri-developer.md` and `kvasir-scout.md`
+  now carry their **Pi** ids (`llamacpp-coder/qwen3-coder-30b`,
+  `llamacpp/qwen3.5-9b`), and `huginn-researcher.md` its declared model — the
+  output of `apply` against the operator's roster. These profiles are what the
+  harnesses load, so the local agents now run on Pi rather than through OpenCode.
+- **`.agents/config/backend` is ignored.** The pinned terminal backend is
+  machine-local, but `config` is a **symlink** to `.agents/config`, so the
+  `config/*` rule never matched it and the pin was committable. The file is named
+  in `.gitignore` instead.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-16-the-validator-stops-calling-a-dead-visualizer-healthy.md

@@ -0,0 +1,20 @@
+## smidja · unversioned · 2026-09-16 — the validator stops calling a dead visualizer healthy
+
+### Why
+`bin/ymir-validate.sh` reported `visualizer PASS "UI built and served on :8437"`
+by looking at `./dist` alone. When the API had crashed on a bad `CMD_DB` and
+nothing was listening on `:8437`, the validator still said PASS — the false-pass
+class this night kept surfacing, in the very tool meant to catch it.
+
+- **The check now probes the port as well as the build.** `dist` missing → FAIL;
+  built but nothing on `:8437` → FAIL with the fix (`run scripts/start.sh`);
+  built and listening → PASS. A PASS now means the thing is actually up.
+- **The smidja-db check resolves the same DB pair** `scripts/start.sh` and
+  `bin/smidja-bootstrap.sh` do (an existing `$YMIR_HOME/smidja/smidja.db` first,
+  then an in-repo copy, `SMIDJA_DB` override) — previously it named the home path
+  only, so an in-repo DB read as missing.
+- `galdr-reread`: `assets/installation.md` — a `visualizer` PASS means built *and*
+  listening.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-17-the-fleet-was-a-rack-of-terminals-now-it-is-the-roster.md

@@ -0,0 +1,18 @@
+## smidja · unversioned · 2026-09-17 — the Fleet was a rack of terminals; now it is the roster
+
+### Why
+- **We were tracking the tools.** `.agents/agents/` holds **21 agent definitions** —
+  Brokk, Sindri, Bragi, Forseti, Mímir, Kvasir, Snotra, Huginn, Hnoss, Sága, Muninn,
+  Frigg, Gróa, Jörð, Sýn, Týr, Galdr… — and the Fleet served **13 panes all named
+  "OpenCode"**, role `opencode`. The board showed seats and never the smiths who might
+  be standing in them.
+- **`bin/hlidskjalf-agents.sh` now reads the roster** (`.agents/agents/*.md`, canonical
+  per RULES/02) and joins each agent to its standing pane: figure, craft, model from
+  the definition's frontmatter, and `live{}` only when a pane answers to it. An
+  **unseated agent is still an agent** — hiding it is what made the board a rack.
+- Harness seats that answer to no one on the roster are kept visible and named
+  honestly, never dressed as agents.
+- Served: **34 entries — 21 agents (roster first) + 13 harness seats, 14 joined**.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-17-the-plan-learns-the-smithy-s-package-name-and-the-instal.md

@@ -0,0 +1,25 @@
+## smidja · unversioned · 2026-09-17 — the plan learns the smithy's package name, and the install forwards its flags
+
+### Why
+Two defects the first real end-to-end npm install surfaced — which is what an
+end-to-end test is for.
+
+- **A surface's name is not its package's name.** The plan looked for
+  `apps/smidja` or `@zerwiz/smidja`, but npm serves the smithy as
+  `@zerwiz/smidja-factory` — so the row read `BLOCKED` on a machine where the
+  smithy was installed and standing. `app_dir`/`app_row` now take the surface
+  name and the package name separately, and the row tells the truth: installed,
+  with the visualizer's UI still to build (the tarball ships its source, never a
+  stale build).
+- **`ymir install --plan` forwarded only `--json`.** `--phase` and `--blocked`
+  were dropped on the way through the installer's door, so `--blocked` answered
+  *unknown flag*. Every plan flag is forwarded now.
+
+**The install this came out of:** `npm install -g @zerwiz/ymir` from the registry
+— `@zerwiz/ymir@0.1.7`, 331 packages, and all four surfaces
+(`hlidskjalf` · `odrerir` · `sessrumnir` · `smidja-factory`) arrived inside the
+distro. The tarball carries no memory store — the leak is closed in the artefact
+as well as in the repo.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-17-the-smithy-can-see-the-well.md

@@ -0,0 +1,20 @@
+## smidja · unversioned · 2026-09-17 — the smithy can see the well
+
+### Why
+- **Why the memory looked absent in the smithy:** the visualizer proxies the bridge's
+  `/inspect`, and `/inspect` reported three fields — store, episodes, agents — while
+  the well holds six layers. A panel told only the episode count cannot show that
+  facts, entities and reflections exist. `/inspect` now reports them all:
+  episodes, facts (active/superseded), entities, edges, reflections (with the last
+  run's time) and the vector index.
+- **Through the smithy right now:**
+  `store $YMIR_HOME/memory/kaia.engram · episodes 8 · facts 36 (35 active) ·
+  entities 60 · edges 480 · reflections 1 (2026-09-16T21:01) · vec index 8`.
+- **A trap found while doing it:** restarting the bridge RAW (`python3 bin/mimir-bridge.py`)
+  loses the env that carries the well's path, and the bridge silently re-points at the
+  old store in the repo — 364 stale episodes and no facts. The well is the **blessed
+  starter's** to raise: `bin/mimir-bridge.sh --start`. A raw restart is a different
+  well wearing the same port.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/docs/fixes/smidja/unversioned-2026-09-17-the-smithy-s-dependency-names-the-package-npm-will-actua.md

@@ -0,0 +1,20 @@
+## smidja · unversioned · 2026-09-17 — the smithy's dependency names the package npm will actually serve
+
+### Why
+- **`@zerwiz/smidja` is published and not served.** npm accepted it twice (0.1.0,
+  then 0.1.1): the tarball and the version document existed, the package document
+  never did, and a republish of 0.1.0 was refused as *"previously published"*. The
+  name answers `404`, the website `403`, while `@zerwiz/hlidskjalf`,
+  `@zerwiz/odrerir` and `@zerwiz/sessrumnir` resolve normally and always have.
+- **A second free name failed identically.** `@zerwiz/smidja-factory` was published
+  and shows the same shape — version served, package document withheld. Two names,
+  one behaviour, while packages published yesterday serve fine: the withholding is
+  npm's, not the name's, and the CLI does not report it.
+- **The distro depends on the name that can resolve.** `optionalDependencies` now
+  names `@zerwiz/smidja-factory`, so the moment npm serves that package the smithy
+  arrives with the rest. Being optional, it is skipped silently until then — the
+  plan's phase-5 row is what says so out loud.
+- **`@zerwiz/ymir` 0.1.7** — the version published once this lands.
+
+### Files
+- *(carried from the frozen CHANGELOG.md)*

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zerwiz/ymir",
-  "version": "0.1.27",
+  "version": "0.1.28",
   "description": "Ymir \u2014 the single-tenant agent operating system. Omarchy-first, host-aware, self-healing.",
   "license": "Apache-2.0",
   "homepage": "https://github.com/zerwiz/ymir",