feat(ci): prune release assets outside the newest two tags (#52) (#70)

Closes #52.

Adds `scripts/prune-release-assets.sh` and a best-effort step at the end of `build-binaries.yml`.

- Lists every release page, sorts by creation date, skips the newest two published releases, deletes the rest's attachments via `DELETE /releases/{id}/assets/{asset_id}`.
- Drafts never consume a keep slot, so a draft cannot strip the newest real release.
- Releases, tags, source archives and release notes are never touched.
- Nothing is excluded: SHA256SUMS and the wrappers go with the binaries. Checksums are only meaningful next to the files they cover, and the consumer drift-check reads the wrappers from the latest release, which stays complete.
- Failures warn and the step has `continue-on-error: true` — a stale asset is cheaper than a blocked publish.

`tests/prune-release-assets.sh` drives the script with a stub `curl`: keep window, out-of-order responses, drafts, no non-asset DELETE, short-page pagination stop, and both failure paths. Wired into `make check`.

v0.2.0 still needs its 9 assets cleared by hand in the web UI, as the issue notes.

🤖 Generated with [Claude Code](https://claude.com/claude-code)Reviewed-on: #70

Co-authored-by: vickydotbat <vickydotbat@tutamail.com>

.gitea/workflows/build-binaries.yml

@@ -65,3 +65,17 @@ jobs:
                 "${api}/releases/${id}/assets?name=$(basename "$f")"
             done
           '
+
+      # Gitea has no asset retention (#52): without this every tag keeps its
+      # ~57 MB binary set forever. Best-effort — a stale asset is cheaper than
+      # a blocked publish, so a failure here never fails the release.
+      - name: Prune assets of older releases
+        continue-on-error: true
+        env:
+          TOKEN: ${{ secrets.GITEA_TOKEN }}
+          REPO: ${{ github.repository }}
+          SERVER: ${{ github.server_url }}
+        run: |
+          nix develop --command bash -c '
+            API="${SERVER}/api/v1/repos/${REPO}" scripts/prune-release-assets.sh
+          '

AGENTS.md

@@ -87,3 +87,21 @@ make smoke                  # assert fail-closed contract
 ## Tests
 
 Tests must survive harmless changes to constants, defaults, wording, ordering, fixture data, and internal implementation details. A test that fails merely because a basic value changed is usually a bad test. Only assert exact values when the value is part of a documented public contract, external protocol, compatibility requirement, security rule, migration, or business rule.
+
+## Agent skills
+
+### Issue tracker
+
+Issues live in Gitea at git.westgate.pw (`ShadowsOverWestgate/sow-tools`), managed with the `tea` CLI. Issues follow ownership — file work in the repo that owns it, not the one you happen to be standing in. See `docs/agents/issue-tracker.md`.
+
+### Triage labels
+
+Default label vocabulary (`needs-triage`, `needs-info`, `ready-for-agent`, `ready-for-human`, `wontfix`). See `docs/agents/triage-labels.md`.
+
+### Domain docs
+
+Single-context: `CONTEXT.md` at the repo root plus `docs/adr/`. See `docs/agents/domain.md`.
+
+### Where work lives
+
+Markdown here is **reference, law, or an ADR — nothing else** (`sow-codebase` ADR-0001). Live work -> wayfinder maps + Gitea issues (closeable, assignable, queryable). Settled decisions -> ADR files in `docs/adr/` (immutable, never closed, only superseded). Standing law -> `DOCTRINE.md` / `AGENTS.md` / `CONTEXT.md`. Current-state reference -> docs describing what the code does now. Everything else — plans, specs, concepts, handoffs, trackers — is process: it belongs in a Gitea issue, not a file. Harvest unfinished intent to an issue before deleting a process doc. `sow-docs` is deprecated and read-only.

Makefile

@@ -7,6 +7,7 @@ check: vet test
 	shellcheck scripts/*.sh
 	yamllint .gitea
 	bash tests/workflow-contract.sh
+	bash tests/prune-release-assets.sh
 
 vet:
 	go vet ./...

README.md

@@ -85,7 +85,9 @@ https://git.westgate.pw/ShadowsOverWestgate/sow-docs).
 - `ci.yml` — vet, test, shellcheck, yamllint, binary smoke, and cross-build all
   targets once per pull request.
 - `build-binaries.yml` — on a `v*` tag, cross-build and upload the binaries,
-  `SHA256SUMS`, and the wrappers to the Gitea release.
+  `SHA256SUMS`, and the wrappers to the Gitea release, then delete the assets
+  of every release except the newest two — Gitea keeps them forever otherwise,
+  and every binary is reproducible from its tag.
 - `sync-wrappers.yml` — on a `main` push that touches `wrappers/`, auto-PR the
   canonical wrappers to the consumer repos in `wrappers/consumers.txt`.
   Consumer drift checks run after those PRs merge to `main`, not on the PRs

docs/adr/0001-markdown-is-reference-or-law-only.md

@@ -0,0 +1,55 @@
+# ADR-0001: Markdown in this repo is reference, law, or an ADR — nothing else
+
+- **Status:** Accepted
+- **Date:** 2026-07-24
+- **Origin:** Adopted workspace-wide from `sow-codebase` ADR-0001, which records
+  the full context (a `docs/` tree that had grown to 434 files, ~27MB, most of
+  it dead process artifacts). This repo adopts the same law so the rule is
+  local, not a cross-repo reference.
+
+## Context
+
+Markdown that *claims to describe current state or a plan* rots, because
+reality diverges and nobody edits the file. Markdown that claims neither — a
+dated, immutable decision record, or a standing rule — does not rot.
+
+Two different things get conflated:
+
+- **Rationale** — why a system is shaped the way it is. Worth keeping.
+- **Process artifacts** — plans, specs, concepts, handoffs, trackers. A record
+  of how work happened, not what is true now.
+
+Left unchecked, every completed effort leaves its planning behind and the repo
+accumulates a "historical" pile that agents and humans must route around to
+find the few live docs.
+
+## Decision
+
+Markdown may exist in this repo only as one of three genres:
+
+1. **Current-state reference** — describes what the code does now, kept honest
+   by code review touching it.
+2. **Standing law** — `DOCTRINE.md`, root and folder-scoped `AGENTS.md`,
+   `CONTEXT.md`. States rules and vocabulary, not plans.
+3. **Immutable decision record** — ADRs under `docs/adr/`. Append-only; never
+   edited, only superseded by a later ADR that points back.
+
+Anything else — implementation plans, design specs, concepts, handoffs,
+progress trackers, scratch — is **process** and does not live in the repo. It
+lives in Gitea issues and wayfinder maps, where it can be assigned, closed, and
+superseded. Small tasks need no written plan at all.
+
+Deleting a process document is not destroying history: `git log -- <path>`
+recovers it. The exception is *unfinished intent* (a design never built, an
+open question still wanted) — that is live, not history, and must be harvested
+to a Gitea ticket before its file is deleted. A citation from a living doc or
+from code must be resolved before the cited file is deleted.
+
+## Consequences
+
+- The documentation map lists only current truth; there is no "historical" pile
+  to route around.
+- No agent, under any plugin or skill, writes process-markdown into the repo.
+  The rule is plugin-agnostic on purpose — it binds the agent, not a named tool.
+- History of deleted process docs is in git; unfinished intent is in the issue
+  tracker; rationale is in ADRs and law. Each thing has exactly one home.

docs/adr/0002-sow-tools-becomes-the-crucible-suite.md

@@ -0,0 +1,12 @@
+# ADR-0002: `sow-tools` becomes the Crucible suite
+
+- **Status:** Accepted
+- **Date:** 2026-06-11
+- **Migrated from:** `sow-docs` `07-decisions-log.md` entry **D11**, on the
+  retirement of `sow-docs`. Content is the original decision, unchanged.
+
+- **Decision:** Rename the future toolchain surface to Crucible: one Go module,
+  multiple `cmd/` binaries, plus a dispatcher so wrapper commands can stay
+  stable.
+- **Rationale:** Depot, HAK, module, topdata, and wiki tooling share internals
+  but have different command surfaces. One module avoids premature repo splits.

docs/adr/0003-crucible-binary-contract-standalone-shims-fail-closed-scaffold.md

@@ -0,0 +1,29 @@
+# ADR-0003: Crucible binary contract: standalone shims, fail-closed scaffold, no committed binaries
+
+- **Status:** Accepted
+- **Date:** 2026-06-11
+- **Migrated from:** `sow-docs` `07-decisions-log.md` entry **D19**, on the
+  retirement of `sow-docs`. Content is the original decision, unchanged.
+
+- **Decision:** Phase 5 builds `migration/sow-tools` (Crucible, D11) as a
+  multi-binary Go scaffold:
+  - One `crucible` dispatcher plus standalone `crucible-{depot,hak,module,topdata,wiki}`
+    binaries sharing one registry (`internal/dispatch`). The dispatcher is for
+    humans/CI; the **standalone shims are canonical for consumers** so wrapper
+    scripts resolve a single-token command and `"$builder" args` quoting stays
+    correct.
+  - Consumer resolution order: explicit env override
+    (`$SOW_MODULE_BUILD`/`$SOW_TOPDATA_BUILD`/`$CRUCIBLE`) → `crucible-<name>` →
+    legacy `sow-<name>-build`. CI runs inside the pinned `crucible:<sha>` image.
+  - Every builder is **unwired** in the scaffold and fails closed (exit 70);
+    it never fakes an artifact. The `internal/` logic is migrated from
+    `gitea/sow-tools` by the operator at cutover (hard rule: no source
+    transplant by tooling).
+  - **Binaries are never committed** — they are CI artifacts / image layers.
+    Retires the committed `nwn-tool` / `tools/sow-toolkit`.
+  - Builders take `NWN_ROOT` only via explicit env/flag; no `$HOME` defaulting.
+- **Rationale:** Locks the command surface, container, and CI shape so migrated
+  logic drops into a stable frame; aligns the three artifact repos onto one
+  Crucible contract while keeping back-compat with the pre-D11 skeletons.
+- **Image name:** `registry.westgate.pw/deployment/crucible:<git-sha>` (the bare
+  `crucible:<sha>` is the local/dev tag).

docs/adr/0004-nix-built-oci-images-start-server-side-with.md

@@ -0,0 +1,24 @@
+# ADR-0004: Nix-built OCI images start server-side with Crucible; local Anvil images deferred
+
+- **Status:** Accepted
+- **Date:** 2026-06-12
+- **Migrated from:** `sow-docs` `07-decisions-log.md` entry **D29**, on the
+  retirement of `sow-docs`. Content is the original decision, unchanged.
+
+- **Decision:** Introduce Nix-built OCI images as an artifact-production option,
+  starting with `sow-tools`/Crucible in server-side CI. `sow-tools` will grow a
+  Nix-built `crucible` package plus `crucible-image`; PR CI builds/smokes it and
+  `main` publishes the normal pinned OCI image
+  `registry.westgate.pw/deployment/crucible:<sha>`. `docker/Dockerfile` stays during
+  parity and as the client-compatible fallback.
+- **Local rule:** Nix users may get optional wrappers that provide tools and
+  call the existing Dockerfiles for local Anvil/NWServer testing. A true
+  `nix build .#nwserver-test-image` is explicitly deferred until real
+  `sow-codebase/src` exists and the Dockerfile image is proven.
+- **Rationale:** Server-side image build shape is harder to change once deploy
+  promotion and registry gates are active, so prove Nix image builds before
+  cutover. Crucible is the lowest-risk pilot because it is a Go toolchain image;
+  NodeBB and Anvil/NWServer depend more heavily on upstream container filesystem
+  semantics.
+- **Non-goals:** no source builds in `sow-platform`; no `nix-sidecar` or
+  Kubernetes-style runtime cache layer; no removal of client Dockerfiles.

docs/agents/domain.md

@@ -0,0 +1,51 @@
+# Domain Docs
+
+How the engineering skills should consume this repo's domain documentation when exploring the codebase.
+
+## Before exploring, read these
+
+- **`CONTEXT.md`** at the repo root, or
+- **`CONTEXT-MAP.md`** at the repo root if it exists — it points at one `CONTEXT.md` per context. Read each one relevant to the topic.
+- **`docs/adr/`** — read ADRs that touch the area you're about to work in. In multi-context repos, also check `src/<context>/docs/adr/` for context-scoped decisions.
+
+If any of these files don't exist, **proceed silently**. Don't flag their absence; don't suggest creating them upfront. The `/domain-modeling` skill (reached via `/grill-with-docs` and `/improve-codebase-architecture`) creates them lazily when terms or decisions actually get resolved.
+
+## File structure
+
+Single-context repo (most repos):
+
+```
+/
+├── CONTEXT.md
+├── docs/adr/
+│   ├── 0001-event-sourced-orders.md
+│   └── 0002-postgres-for-write-model.md
+└── src/
+```
+
+Multi-context repo (presence of `CONTEXT-MAP.md` at the root):
+
+```
+/
+├── CONTEXT-MAP.md
+├── docs/adr/                          ← system-wide decisions
+└── src/
+    ├── ordering/
+    │   ├── CONTEXT.md
+    │   └── docs/adr/                  ← context-specific decisions
+    └── billing/
+        ├── CONTEXT.md
+        └── docs/adr/
+```
+
+## Use the glossary's vocabulary
+
+When your output names a domain concept (in an issue title, a refactor proposal, a hypothesis, a test name), use the term as defined in `CONTEXT.md`. Don't drift to synonyms the glossary explicitly avoids.
+
+If the concept you need isn't in the glossary yet, that's a signal — either you're inventing language the project doesn't use (reconsider) or there's a real gap (note it for `/domain-modeling`).
+
+## Flag ADR conflicts
+
+If your output contradicts an existing ADR, surface it explicitly rather than silently overriding:
+
+> _Contradicts ADR-0007 (event-sourced orders) — but worth reopening because…_

docs/agents/issue-tracker.md

@@ -0,0 +1,88 @@
+# Issue tracker: Gitea (via tea)
+
+Issues for this repo live in Gitea at `git.westgate.pw`, repo
+`ShadowsOverWestgate/sow-tools`. Use the `tea` CLI for all operations —
+`gh` does not work here. For anything `tea` lacks a subcommand for, use
+`tea api <endpoint>` (Gitea's API mirrors GitHub's closely).
+
+Authenticate with your own `tea` login (`tea login add`); never commit tokens
+or tea config into this repo. Note Gitea blocks self-review, so approving a PR
+needs a different account than the one that opened it.
+
+## Where work lives
+
+Markdown in this repo is **reference, law, or an ADR — nothing else**
+(`sow-codebase` ADR-0001). Four homes, no overlap:
+
+- **Live work** → wayfinder maps + Gitea issues. Closeable, assignable,
+  queryable. Never a markdown file.
+- **Settled decisions** → ADR files in `docs/adr/`. Immutable, findable, never
+  closed, never edited — only superseded by a later ADR pointing back. When an
+  issue ends in a durable decision, write the ADR, then close the issue
+  pointing at it. Decisions spanning repos go to `sow-platform/docs/adr/`.
+- **Standing law** → `DOCTRINE.md`, `AGENTS.md`, `CONTEXT.md`. Rules that are
+  always true and vocabulary everyone shares — not the record of one decision.
+- **Current-state reference** → docs describing what the code does now, kept
+  honest by review touching them.
+
+Anything else — implementation plans, design specs, concepts, handoffs,
+progress trackers, scratch — is **process**. It does not live in this repo. It
+lives in a Gitea issue or a wayfinder map, where it can be assigned, closed,
+and superseded. Small tasks need no written plan at all.
+
+Deleting a process doc is not destroying history — `git log -- <path>` recovers
+it. But *unfinished intent* (a design never built, an open question still
+wanted) is live, not history: harvest it to a Gitea issue before deleting.
+
+`sow-docs` is deprecated and read-only. Never add to it, never send work there.
+
+## Which repo gets the issue
+
+Issues follow ownership. File the issue in the repo that **owns the work** —
+see the Repo/Owns/Produces table in the workspace root `AGENTS.md`. Standing
+in one repo is not a reason to file there.
+
+If work spans repos, file it in the repo that owns the *outcome* and reference
+the others from it. A wrong-repo issue is a routing bug, not a filing
+preference — move it.
+
+## Conventions
+
+- **Create an issue**: `tea issues create --title "..." --description "..."`
+- **Read an issue**: `tea issues <number>` and
+  `tea api repos/ShadowsOverWestgate/sow-tools/issues/<number>/comments` for comments.
+- **List issues**: `tea issues list --state open` (add `--labels ...` to filter).
+- **Comment**: `tea comment <number> "..." </dev/null`
+  Always redirect stdin. `tea` reads stdin to EOF and appends it to the body,
+  so any non-interactive shell (every agent) hangs forever without
+  `</dev/null`. Same trap on `tea issues create --description` and
+  `tea pr create`.
+- **Apply / remove labels**: `tea api --method PATCH` on the issue, or
+  `tea api repos/ShadowsOverWestgate/sow-tools/issues/<number>/labels` endpoints.
+- **Close**: `tea issues close <number>`
+
+`tea` infers the repo from the git remote when run inside the clone.
+Gitea shares one number space across issues and PRs.
+
+## Pull requests as a triage surface
+
+**PRs as a request surface: no.**
+
+## When a skill says "publish to the issue tracker"
+
+Create a Gitea issue with `tea issues create`.
+
+## When a skill says "fetch the relevant ticket"
+
+Run `tea issues <number>` plus the comments API call above.
+
+## Wayfinding operations
+
+Used by `/wayfinder`. The **map** is a single issue with **child** issues as tickets.
+
+- **Map**: a single issue labelled `wayfinder:map`, holding the Notes / Decisions-so-far / Fog body. `tea issues create --title "..." --description "..." --labels wayfinder:map`.
+- **Child ticket**: this Gitea instance (v1.27) has no native sub-issue hierarchy, so a child issue carries `Part of #<map>` at the top of its description, and is also added to a task list in the map body. Labels: `wayfinder:<type>` (`research`/`prototype`/`grilling`/`task`). Once claimed, the ticket is assigned to the driving dev.
+- **Blocking**: Gitea's **native dependencies API** — the canonical, UI-visible representation (shows as "Depends on" / "Blocks" on the issue page). Add an edge with `tea api -X POST repos/ShadowsOverWestgate/sow-tools/issues/<child>/dependencies -f owner=ShadowsOverWestgate -f repo=sow-tools -F index=<blocker>`, where `<blocker>` is the blocker's issue **index** (its `#number` — Gitea's dependency API takes the index directly, unlike GitHub's numeric database id). Check status with `tea api repos/ShadowsOverWestgate/sow-tools/issues/<child>/dependencies` (GET) — a ticket is unblocked when every returned issue's `state` is `closed`.
+- **Frontier query**: list the map's open children (`tea issues list --state open`, keep the ones whose description contains `Part of #<map>`), drop any with an open dependency (per the GET above) or an assignee; first in map order wins.
+- **Claim**: `tea issues edit <n> --add-assignees <username>` — the session's first write.
+- **Resolve**: `tea comment <n> "<answer>" </dev/null`, then `tea issues close <n>`, then append a context pointer (gist + link) to the map's Decisions-so-far.

docs/agents/triage-labels.md

@@ -0,0 +1,15 @@
+# Triage Labels
+
+The skills speak in terms of five canonical triage roles. This file maps those roles to the actual label strings used in this repo's issue tracker (Gitea — see `issue-tracker.md` for how to apply labels with `tea`).
+
+| Label in mattpocock/skills | Label in our tracker | Meaning                                  |
+| -------------------------- | -------------------- | ---------------------------------------- |
+| `needs-triage`             | `needs-triage`       | Maintainer needs to evaluate this issue  |
+| `needs-info`               | `needs-info`         | Waiting on reporter for more information |
+| `ready-for-agent`          | `ready-for-agent`    | Fully specified, ready for an AFK agent  |
+| `ready-for-human`          | `ready-for-human`    | Requires human implementation            |
+| `wontfix`                  | `wontfix`            | Will not be actioned                     |
+
+When a skill mentions a role (e.g. "apply the AFK-ready triage label"), use the corresponding label string from this table.
+
+Edit the right-hand column to match whatever vocabulary you actually use.

docs/command-surface.md

@@ -29,9 +29,15 @@ aliases.
 | `assets` | `fix-mdl` | Lowercase `.mdl` names and rewrite model identity to match. |
 | `assets` | `check-dupes` | Report runtime-name (basename) collisions across dirs. |
 | `assets` | `clean-dupes` | Delete clean-tree files whose basename collides with primary. |
+| `depot` | `status` | Report referenced-vs-present drift against a backend. |
+| `depot` | `push` | Upload referenced-but-absent blobs from a local depot. |
+| `depot` | `verify` | Existence sweep plus sampled download re-hash. |
+| `depot` | `get` | Fetch one blob with sha re-verify. |
+| `depot` | `pull` | Incremental verified pull of every referenced blob. |
 
-`crucible-depot` remains registered but unwired. It fails closed with exit `70`
+`depot status` and `depot get` pick their backend either with `--out DIR`, a
-and never emits placeholder artifacts.
+depot tree on disk, or with `--target bunny|cdn`, a remote backend. The two
+flags are mutually exclusive.
 
 ## Hidden compatibility aliases
 
@@ -49,6 +55,7 @@ but omitted from routine help and the interactive menu:
 | `topdata` | `build-top-package` | `build-top-package` |
 | `topdata` | `compare-topdata` | `compare-topdata` |
 | `topdata` | `convert-topdata` | `convert-topdata` |
+| `depot` | `--target local` | `--out $DEPOT_DIR` on `status`/`get` |
 | `wiki` | `build-wiki` | `build-wiki` |
 | `wiki` | `deploy-wiki` | `deploy-wiki` |
 

internal/depot/run.go

@@ -11,6 +11,7 @@ import (
 	"os"
 	"path/filepath"
 	"sort"
+	"strings"
 	"sync"
 	"time"
 )
@@ -51,13 +52,15 @@ func Run(args []string, stdout, stderr io.Writer, getenv func(string) string) in
 
 func printRunUsage(w io.Writer) {
 	fmt.Fprint(w, `usage:
-  depot status [--manifests DIR] [--source LOCAL_DEPOT] --target bunny|cdn|local
+  depot status [--manifests DIR] --target bunny|cdn | --out DIR
-  depot push   [--manifests DIR] --source LOCAL_DEPOT   --target bunny
+  depot push   [--manifests DIR] --source LOCAL_DEPOT --target bunny
-  depot verify [--manifests DIR] --target bunny|cdn     [--sample N]
+  depot verify [--manifests DIR] --target bunny|cdn   [--sample N]
-  depot get    <sha> <dest>       --target cdn|bunny|local
+  depot get    <sha> <dest>      --target cdn|bunny   | --out DIR
-  depot pull   [--manifests DIR] --dest DIR             --target cdn|bunny
+  depot pull   [--manifests DIR] --dest DIR           --target cdn|bunny
 
---target local uses the DEPOT_DIR environment variable as the local root.
+--out DIR reads and writes a depot tree on disk at DIR instead of a remote
+backend. --target names a remote backend only; the two are mutually exclusive.
+--target local is a deprecated alias for --out $DEPOT_DIR.
 `)
 }
 
@@ -65,16 +68,27 @@ func printRunUsage(w io.Writer) {
 // than internal/backend failures (exit 70).
 var errUsage = errors.New("usage error")
 
-// resolveBackend builds the named backend, resolving "local" against DEPOT_DIR.
+// resolveBackend picks the backend for a command that can work either against a
-func resolveBackend(target string, getenv func(string) string, cfg Config) (Backend, error) {
+// depot tree on disk (--out DIR) or a remote backend (--target bunny|cdn).
-	root := ""
+// "--target local" stays as a deprecated alias for --out $DEPOT_DIR so older
-	if target == "local" {
+// callers keep working.
-		root = getenv("DEPOT_DIR")
+func resolveBackend(out, target string, getenv func(string) string, cfg Config) (Backend, error) {
+	switch {
+	case out != "" && target != "":
+		return nil, fmt.Errorf("--out and --target are mutually exclusive: %w", errUsage)
+	case out != "":
+		return NewBackend("local", out, cfg)
+	case target == "local":
+		root := getenv("DEPOT_DIR")
 		if root == "" {
-			return nil, fmt.Errorf("--target local requires DEPOT_DIR to be set: %w", errUsage)
+			return nil, fmt.Errorf("--target local requires DEPOT_DIR to be set (prefer --out DIR): %w", errUsage)
 		}
+		return NewBackend("local", root, cfg)
+	case target == "":
+		return nil, fmt.Errorf("--out DIR or --target bunny|cdn is required: %w", errUsage)
+	default:
+		return NewBackend(target, "", cfg)
 	}
-	return NewBackend(target, root, cfg)
 }
 
 // backendErrExit maps a resolveBackend error onto the exit contract.
@@ -119,18 +133,18 @@ func runStatus(args []string, stdout, stderr io.Writer, getenv func(string) stri
 	fs := flag.NewFlagSet("status", flag.ContinueOnError)
 	fs.SetOutput(stderr)
 	manifests := fs.String("manifests", "assets", "directory of *.yml manifests")
-	_ = fs.String("source", "", "local depot root (unused for status target=local; see DEPOT_DIR)")
+	source := fs.String("source", "", "deprecated and ignored (use --out DIR for a depot tree on disk)")
-	target := fs.String("target", "", "bunny|cdn|local")
+	out := fs.String("out", "", "depot tree on disk to read instead of a remote backend")
+	target := fs.String("target", "", "bunny|cdn")
 	if err := fs.Parse(args); err != nil {
 		return exitUsage
 	}
-	if *target == "" {
+	if *source != "" {
-		fmt.Fprintln(stderr, "depot status: --target is required")
+		fmt.Fprintln(stderr, "depot status: --source is ignored; use --out DIR")
-		return exitUsage
 	}
 
 	cfg := LoadConfig(getenv)
-	backend, err := resolveBackend(*target, getenv, cfg)
+	backend, err := resolveBackend(*out, *target, getenv, cfg)
 	if err != nil {
 		fmt.Fprintln(stderr, "depot status:", err)
 		return backendErrExit(err)
@@ -309,13 +323,21 @@ func runVerify(args []string, stdout, stderr io.Writer, getenv func(string) stri
 func runGet(args []string, stdout, stderr io.Writer, getenv func(string) string) int {
 	fs := flag.NewFlagSet("get", flag.ContinueOnError)
 	fs.SetOutput(stderr)
-	target := fs.String("target", "", "cdn|bunny|local")
+	out := fs.String("out", "", "depot tree on disk to read instead of a remote backend")
-	if err := fs.Parse(args); err != nil {
+	target := fs.String("target", "", "cdn|bunny")
+	// Positionals come first in the documented usage, and Go's flag package
+	// stops parsing at the first one, so split them off by hand.
+	split := 0
+	for split < len(args) && !strings.HasPrefix(args[split], "-") {
+		split++
+	}
+	positional := args[:split]
+	if err := fs.Parse(args[split:]); err != nil {
 		return exitUsage
 	}
-	positional := fs.Args()
+	positional = append(positional, fs.Args()...)
 	if len(positional) != 2 {
-		fmt.Fprintln(stderr, "depot get: usage: depot get <sha> <dest> --target cdn|bunny|local")
+		fmt.Fprintln(stderr, "depot get: usage: depot get <sha> <dest> --target cdn|bunny | --out DIR")
 		return exitUsage
 	}
 	sha, dest := positional[0], positional[1]
@@ -323,13 +345,9 @@ func runGet(args []string, stdout, stderr io.Writer, getenv func(string) string)
 		fmt.Fprintf(stderr, "depot get: invalid sha %q\n", sha)
 		return exitUsage
 	}
-	if *target == "" {
-		fmt.Fprintln(stderr, "depot get: --target is required")
-		return exitUsage
-	}
 
 	cfg := LoadConfig(getenv)
-	backend, err := resolveBackend(*target, getenv, cfg)
+	backend, err := resolveBackend(*out, *target, getenv, cfg)
 	if err != nil {
 		fmt.Fprintln(stderr, "depot get:", err)
 		return backendErrExit(err)

internal/depot/run_test.go

@@ -54,6 +54,91 @@ func TestRunUsage(t *testing.T) {
 	})
 }
 
+func TestOutFlag(t *testing.T) {
+	t.Run("status --out reads the given tree", func(t *testing.T) {
+		sha := shaOf("out-blob")
+		manifests := t.TempDir()
+		writeManifest(t, manifests, sha, 8)
+		depotDir := t.TempDir()
+		writeBlob(t, depotDir, sha, "out-blob")
+
+		var out, errb bytes.Buffer
+		code := Run([]string{"status", "--manifests", manifests, "--out", depotDir}, &out, &errb, testGetenv(nil))
+		if code != 0 {
+			t.Fatalf("expected 0, got %d (stdout=%s stderr=%s)", code, out.String(), errb.String())
+		}
+		if !bytesContains(out.String(), "present=1") {
+			t.Fatalf("expected present=1, got %s", out.String())
+		}
+	})
+
+	t.Run("get --out fetches from the given tree", func(t *testing.T) {
+		sha := shaOf("get-blob")
+		depotDir := t.TempDir()
+		writeBlob(t, depotDir, sha, "get-blob")
+		dest := filepath.Join(t.TempDir(), "fetched")
+
+		var out, errb bytes.Buffer
+		code := Run([]string{"get", sha, dest, "--out", depotDir}, &out, &errb, testGetenv(nil))
+		if code != 0 {
+			t.Fatalf("expected 0, got %d (stderr=%s)", code, errb.String())
+		}
+		got, err := os.ReadFile(dest)
+		if err != nil {
+			t.Fatal(err)
+		}
+		if string(got) != "get-blob" {
+			t.Fatalf("got %q", got)
+		}
+	})
+
+	t.Run("--out with any --target is rejected", func(t *testing.T) {
+		for _, target := range []string{"bunny", "cdn", "local"} {
+			var out, errb bytes.Buffer
+			code := Run([]string{"status", "--manifests", t.TempDir(), "--out", t.TempDir(), "--target", target}, &out, &errb, testGetenv(nil))
+			if code != 64 {
+				t.Fatalf("--target %s: expected 64, got %d (stderr=%s)", target, code, errb.String())
+			}
+		}
+	})
+
+	t.Run("get accepts flags before the positionals", func(t *testing.T) {
+		sha := shaOf("flags-first-blob")
+		depotDir := t.TempDir()
+		writeBlob(t, depotDir, sha, "flags-first-blob")
+		dest := filepath.Join(t.TempDir(), "fetched")
+
+		var out, errb bytes.Buffer
+		code := Run([]string{"get", "--out", depotDir, sha, dest}, &out, &errb, testGetenv(nil))
+		if code != 0 {
+			t.Fatalf("expected 0, got %d (stderr=%s)", code, errb.String())
+		}
+	})
+
+	t.Run("neither --out nor --target is rejected", func(t *testing.T) {
+		var out, errb bytes.Buffer
+		code := Run([]string{"status", "--manifests", t.TempDir()}, &out, &errb, testGetenv(nil))
+		if code != 64 {
+			t.Fatalf("expected 64, got %d (stderr=%s)", code, errb.String())
+		}
+	})
+
+	t.Run("--target local still works as a hidden alias", func(t *testing.T) {
+		sha := shaOf("alias-blob")
+		manifests := t.TempDir()
+		writeManifest(t, manifests, sha, 11)
+		depotDir := t.TempDir()
+		writeBlob(t, depotDir, sha, "alias-blob")
+
+		var out, errb bytes.Buffer
+		code := Run([]string{"status", "--manifests", manifests, "--target", "local"}, &out, &errb,
+			testGetenv(map[string]string{"DEPOT_DIR": depotDir}))
+		if code != 0 {
+			t.Fatalf("expected 0, got %d (stdout=%s stderr=%s)", code, out.String(), errb.String())
+		}
+	})
+}
+
 func TestGetInvalidSHA(t *testing.T) {
 	var out, errb bytes.Buffer
 	dir := t.TempDir()
@@ -172,6 +257,17 @@ func bytesContains(s, substr string) bool {
 	return bytes.Contains([]byte(s), []byte(substr))
 }
 
+func writeBlob(t *testing.T, root, sha, content string) {
+	t.Helper()
+	path := filepath.Join(root, BlobKey(sha))
+	if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(content), 0644); err != nil {
+		t.Fatal(err)
+	}
+}
+
 func writeManifest(t *testing.T, dir, sha string, size int64) {
 	t.Helper()
 	content := fmt.Sprintf("assets:\n  - path: foo\n    sha256: %s\n    size: %d\n", sha, size)

internal/dispatch/dispatch.go

@@ -94,12 +94,12 @@ var Registry = []Builder{
 	{
 		Name:    "depot",
 		Bin:     "crucible-depot",
-		Summary: "content-addressed asset depot (local/cdn/bunny)",
+		Summary: "content-addressed asset depot (disk/cdn/bunny)",
 		Commands: []Command{
-			{Name: "status", Summary: "report referenced-vs-present drift against a backend", Usage: "crucible depot status [--manifests DIR] [--source DIR] --target bunny|cdn|local"},
+			{Name: "status", Summary: "report referenced-vs-present drift against a backend", Usage: "crucible depot status [--manifests DIR] --target bunny|cdn | --out DIR"},
 			{Name: "push", Summary: "upload referenced-but-absent blobs from a local depot", Usage: "crucible depot push [--manifests DIR] --source DIR --target bunny"},
 			{Name: "verify", Summary: "existence sweep plus sampled download re-hash", Usage: "crucible depot verify [--manifests DIR] --target bunny|cdn [--sample N]"},
-			{Name: "get", Summary: "fetch one blob with sha re-verify", Usage: "crucible depot get <sha> <dest> --target cdn|bunny|local"},
+			{Name: "get", Summary: "fetch one blob with sha re-verify", Usage: "crucible depot get <sha> <dest> --target cdn|bunny | --out DIR"},
 			{Name: "pull", Summary: "incremental verified pull of every referenced blob", Usage: "crucible depot pull [--manifests DIR] --dest DIR --target cdn|bunny"},
 		},
 		Wired: true,

internal/erf/erf.go

@@ -67,6 +67,19 @@ type resourceEntry struct {
 	Size   uint32
 }
 
+// extensionTypes and typeExtensions must stay exact inverses of each other;
+// init() panics if they drift apart.
+//
+// Numbers below 0x0BB8 follow upstream neverwinter.nim
+// (neverwinter/restype.nim). The 0x0BB8 and up entries are lyt/vis/mdx: real
+// Aurora archive types that NWN1 ships in its own data/*.bif but upstream
+// happens not to register. xoreos corroborates those three numbers
+// (src/aurora/types.h).
+//
+// This table is NWN:EE only. Do not add NWN2 formats (mdb, gr2, wlk, xml):
+// NWN:EE either gives that number to something else (2070 is xbc here and mdb
+// in NWN2) or has no number for it at all, so packing one into a HAK writes a
+// resource the game misreads. See the reserved list in erf_test.go.
 var extensionTypes = map[string]uint16{
 	"res": 0x0000,
 	"bmp": 0x0001,
@@ -96,12 +109,13 @@ var extensionTypes = map[string]uint16{
 	"utt": 0x07F0,
 	"dds": 0x07F1,
 	"uts": 0x07F3,
+	"ltr": 0x07F4,
+	"gff": 0x07F5,
 	"fac": 0x07F6,
-	"gff": 0x07F7,
 	"ute": 0x07F8,
 	"utd": 0x07FA,
 	"utp": 0x07FC,
-	"dfa": 0x07FD,
+	"dft": 0x07FD,
 	"gic": 0x07FE,
 	"gui": 0x07FF,
 	"utm": 0x0803,
@@ -117,20 +131,15 @@ var extensionTypes = map[string]uint16{
 	"ndb": 0x0810,
 	"ptm": 0x0811,
 	"ptt": 0x0812,
-	"ltr": 0x0813,
 	"shd": 0x0815,
-	"mdb": 0x0816,
 	"mtr": 0x0818,
-	"jpg": 0x081C,
 	"lod": 0x081E,
 	"gif": 0x081F,
 	"png": 0x0820,
+	"jpg": 0x0821,
 	"lyt": 0x0BB8,
 	"vis": 0x0BB9,
 	"mdx": 0x0BC0,
-	"wlk": 0x0BCC,
-	"xml": 0x0BCD,
-	"gr2": 0x0FA3,
 }
 
 var typeExtensions = map[uint16]string{
@@ -162,12 +171,13 @@ var typeExtensions = map[uint16]string{
 	0x07F0: "utt",
 	0x07F1: "dds",
 	0x07F3: "uts",
+	0x07F4: "ltr",
+	0x07F5: "gff",
 	0x07F6: "fac",
-	0x07F7: "gff",
 	0x07F8: "ute",
 	0x07FA: "utd",
 	0x07FC: "utp",
-	0x07FD: "dfa",
+	0x07FD: "dft",
 	0x07FE: "gic",
 	0x07FF: "gui",
 	0x0803: "utm",
@@ -183,20 +193,15 @@ var typeExtensions = map[uint16]string{
 	0x0810: "ndb",
 	0x0811: "ptm",
 	0x0812: "ptt",
-	0x0813: "ltr",
 	0x0815: "shd",
-	0x0816: "mdb",
 	0x0818: "mtr",
-	0x081C: "jpg",
 	0x081E: "lod",
 	0x081F: "gif",
 	0x0820: "png",
+	0x0821: "jpg",
 	0x0BB8: "lyt",
 	0x0BB9: "vis",
 	0x0BC0: "mdx",
-	0x0BCC: "wlk",
-	0x0BCD: "xml",
-	0x0FA3: "gr2",
 }
 
 func init() {
@@ -205,8 +210,13 @@ func init() {
 		if !ok {
 			panic(fmt.Sprintf("missing canonical extension for resource type 0x%04X", resourceType))
 		}
-		if canonicalExt == ext {
+		if canonicalExt != ext {
-			continue
+			panic(fmt.Sprintf("resource type 0x%04X is %q in extensionTypes but %q in typeExtensions", resourceType, ext, canonicalExt))
+		}
+	}
+	for resourceType, ext := range typeExtensions {
+		if registered, ok := extensionTypes[ext]; !ok || registered != resourceType {
+			panic(fmt.Sprintf("extension %q is 0x%04X in typeExtensions but 0x%04X in extensionTypes (registered=%v)", ext, resourceType, registered, ok))
 		}
 	}
 }

internal/erf/erf_test.go

@@ -134,14 +134,10 @@ func TestExtensionMappingsSupportModernAssetTypes(t *testing.T) {
 		"mtr": 0x0818,
 		"shd": 0x0815,
 		"txi": 0x07E6,
-		"jpg": 0x081C,
+		"jpg": 0x0821,
-		"mdb": 0x0816,
 		"lyt": 0x0BB8,
 		"vis": 0x0BB9,
 		"mdx": 0x0BC0,
-		"xml": 0x0BCD,
-		"wlk": 0x0BCC,
-		"gr2": 0x0FA3,
 	}
 	for ext, wantType := range cases {
 		gotType, ok := ResourceTypeForExtension(ext)
@@ -191,3 +187,71 @@ func TestExtensionMappingsSupportAllUTBlueprintTypes(t *testing.T) {
 		}
 	}
 }
+
+// TestRestypeTableMatchesUpstream pins the restype numbers Crucible shares with
+// neverwinter.nim's restype.nim. The values below are upstream's; a mismatch
+// means a HAK we write is mislabelled for the game.
+func TestRestypeTableMatchesUpstream(t *testing.T) {
+	upstream := map[string]uint16{
+		"res": 0, "bmp": 1, "mve": 2, "tga": 3, "wav": 4, "plt": 6, "ini": 7,
+		"bmu": 8, "txt": 10, "mdl": 2002, "nss": 2009, "ncs": 2010, "are": 2012,
+		"set": 2013, "ifo": 2014, "bic": 2015, "wok": 2016, "2da": 2017,
+		"tlk": 2018, "txi": 2022, "git": 2023, "uti": 2025, "utc": 2027,
+		"dlg": 2029, "itp": 2030, "utt": 2032, "dds": 2033, "uts": 2035,
+		"ltr": 2036, "gff": 2037, "fac": 2038, "ute": 2040, "utd": 2042,
+		"utp": 2044, "dft": 2045, "gic": 2046, "gui": 2047, "utm": 2051,
+		"dwk": 2052, "pwk": 2053, "utg": 2055, "jrl": 2056, "utw": 2058,
+		"ssf": 2060, "hak": 2061, "nwm": 2062, "bik": 2063, "ndb": 2064,
+		"ptm": 2065, "ptt": 2066, "shd": 2069, "mtr": 2072, "lod": 2078,
+		"gif": 2079, "png": 2080, "jpg": 2081,
+	}
+	for ext, want := range upstream {
+		got, ok := ResourceTypeForExtension(ext)
+		if !ok {
+			t.Errorf("%s: not registered, upstream has %d", ext, want)
+			continue
+		}
+		if got != want {
+			t.Errorf("%s: registered as %d, upstream has %d", ext, got, want)
+		}
+	}
+	// Extensions upstream registers that Crucible must not reuse for anything else.
+	reserved := map[uint16]string{2039: "bte", 2067: "bak", 2070: "xbc", 2076: "tml"}
+	for number, upstreamExt := range reserved {
+		if ext, ok := ExtensionForResourceType(number); ok {
+			t.Errorf("%d: registered as %s, upstream reserves it for %s", number, ext, upstreamExt)
+		}
+	}
+}
+
+// TestNWN2FormatsAreNotRegistered keeps NWN2 file formats out of an NWN:EE
+// table. mdb and gr2 have Aurora numbers that mean something else (or nothing)
+// in NWN:EE; wlk and xml have no archive number in any Aurora game, so the
+// values Crucible used for them were invented. Packing any of these into a HAK
+// writes a resource the game misreads.
+func TestNWN2FormatsAreNotRegistered(t *testing.T) {
+	for _, ext := range []string{"mdb", "gr2", "wlk", "xml"} {
+		if number, ok := ResourceTypeForExtension(ext); ok {
+			t.Errorf("%s is an NWN2 format but is registered as 0x%04X", ext, number)
+		}
+	}
+}
+
+// TestRestypeTablesAreMutualInverses is the check init() is meant to enforce.
+func TestRestypeTablesAreMutualInverses(t *testing.T) {
+	for ext, number := range extensionTypes {
+		canonical, ok := typeExtensions[number]
+		if !ok {
+			t.Errorf("%s: number 0x%04X missing from typeExtensions", ext, number)
+			continue
+		}
+		if canonical != ext {
+			t.Errorf("%s: maps to 0x%04X, which maps back to %s", ext, number, canonical)
+		}
+	}
+	for number, ext := range typeExtensions {
+		if got, ok := extensionTypes[ext]; !ok || got != number {
+			t.Errorf("0x%04X: maps to %s, which maps back to 0x%04X (ok=%v)", number, ext, got, ok)
+		}
+	}
+}

internal/pipeline/build.go

@@ -593,9 +593,14 @@ func envBool(name string) bool {
 func collectModuleResources(p *project.Project, moduleHakOrder []string) ([]erf.Resource, error) {
 	var moduleResources []erf.Resource
 
+	palettes, err := collectPaletteDescriptors(p)
+	if err != nil {
+		return nil, err
+	}
+
 	for _, rel := range p.Inventory.SourceFiles {
 		abs := filepath.Join(p.SourceDir(), filepath.FromSlash(rel))
-		resource, err := resourceFromJSON(abs, moduleHakOrder)
+		resource, err := resourceFromJSON(abs, moduleHakOrder, palettes)
 		if err != nil {
 			return nil, err
 		}
@@ -1030,7 +1035,7 @@ func compareResourceKeys(a, b erf.Resource) int {
 	return 0
 }
 
-func resourceFromJSON(path string, moduleHakOrder []string) (erf.Resource, error) {
+func resourceFromJSON(path string, moduleHakOrder []string, palettes paletteProjection) (erf.Resource, error) {
 	name, extension, err := splitSourceName(path)
 	if err != nil {
 		return erf.Resource{}, err
@@ -1059,6 +1064,9 @@ func resourceFromJSON(path string, moduleHakOrder []string) (erf.Resource, error
 	if extension == ".ifo" && name == "module" && len(moduleHakOrder) > 0 {
 		setModuleHAKList(&document, moduleHakOrder)
 	}
+	if extension == ".itp" && isPaletteProjectionResref(name) {
+		projectPaletteDocument(&document, palettes[strings.ToLower(name)])
+	}
 
 	var buf bytes.Buffer
 	if err := gff.Write(&buf, document); err != nil {

internal/pipeline/compare.go

@@ -117,6 +117,10 @@ func expectedResources(p *project.Project) (map[string]resourceExpectation, erro
 	if err != nil {
 		return nil, err
 	}
+	palettes, err := collectPaletteDescriptors(p)
+	if err != nil {
+		return nil, err
+	}
 
 	for _, rel := range p.Inventory.SourceFiles {
 		abs := filepath.Join(p.SourceDir(), filepath.FromSlash(rel))
@@ -136,6 +140,9 @@ func expectedResources(p *project.Project) (map[string]resourceExpectation, erro
 		if extension == ".ifo" && strings.EqualFold(name, "module") && len(moduleHakOrder) > 0 {
 			setModuleHAKList(&document, moduleHakOrder)
 		}
+		if extension == ".itp" && isPaletteProjectionResref(name) {
+			projectPaletteDocument(&document, palettes[strings.ToLower(name)])
+		}
 
 		canonical, err := json.Marshal(document)
 		if err != nil {

internal/pipeline/extract.go

@@ -124,6 +124,13 @@ func extractArchiveResources(p *project.Project, archive erf.Archive, desired ma
 			skippedCount++
 			continue
 		}
+		// *palcus.itp are Toolset-generated palette projections; the module
+		// build regenerates them from source blueprints, so extraction never
+		// writes them back into source.
+		if ext == "itp" && isPaletteProjectionResref(resource.Name) {
+			skippedCount++
+			continue
+		}
 
 		target, data, err := extractedFile(p, resource, ext)
 		if err != nil {

internal/pipeline/palette.go

@@ -0,0 +1,298 @@
+package pipeline
+
+// Custom palette projections (*palcus.itp) are generated artifacts: the Toolset
+// derives them from the category frameworks plus the blueprints present in the
+// module. Crucible reproduces that projection deterministically at build time so
+// the module source only carries descriptor-free category skeletons and the
+// blueprint files themselves. Extract never writes palcus files back.
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+
+	"git.westgate.pw/ShadowsOverWestgate/sow-tools/internal/gff"
+	"git.westgate.pw/ShadowsOverWestgate/sow-tools/internal/project"
+)
+
+const noStrref = 0xFFFFFFFF
+
+// hiddenPaletteID suppresses a blueprint from the Custom palette (engine
+// convention; see docs in sow-module's palette research notes).
+const hiddenPaletteID = 255
+
+var paletteFamilyByExtension = map[string]string{
+	".utc": "creaturepalcus",
+	".utd": "doorpalcus",
+	".ute": "encounterpalcus",
+	".uti": "itempalcus",
+	".utm": "storepalcus",
+	".utp": "placeablepalcus",
+	".uts": "soundpalcus",
+	".utt": "triggerpalcus",
+	".utw": "waypointpalcus",
+}
+
+type paletteDescriptor struct {
+	name     string
+	strref   uint32
+	resref   string
+	creature bool
+	cr       float32
+	faction  string
+}
+
+func (d paletteDescriptor) sortKey() string {
+	if d.strref != noStrref || d.name == "" {
+		return strings.ToLower(d.resref)
+	}
+	return strings.ToLower(d.name)
+}
+
+// palette resref (e.g. "itempalcus") -> terminal category ID -> descriptors.
+type paletteProjection map[string]map[uint8][]paletteDescriptor
+
+func isPaletteProjectionResref(name string) bool {
+	return strings.HasSuffix(strings.ToLower(name), "palcus")
+}
+
+func collectPaletteDescriptors(p *project.Project) (paletteProjection, error) {
+	projection := paletteProjection{}
+	factions, err := loadFactionNames(p)
+	if err != nil {
+		return nil, err
+	}
+
+	for _, rel := range p.Inventory.SourceFiles {
+		abs := filepath.Join(p.SourceDir(), filepath.FromSlash(rel))
+		name, extension, err := splitSourceName(abs)
+		if err != nil {
+			return nil, err
+		}
+		family, ok := paletteFamilyByExtension[extension]
+		if !ok {
+			continue
+		}
+
+		raw, err := os.ReadFile(abs)
+		if err != nil {
+			return nil, fmt.Errorf("read %s: %w", abs, err)
+		}
+		var document gff.Document
+		if err := json.Unmarshal(raw, &document); err != nil {
+			return nil, fmt.Errorf("parse gff json %s: %w", abs, err)
+		}
+
+		descriptor, paletteID, ok := blueprintDescriptor(document.Root, name, extension, factions)
+		if !ok {
+			continue
+		}
+		if projection[family] == nil {
+			projection[family] = map[uint8][]paletteDescriptor{}
+		}
+		projection[family][paletteID] = append(projection[family][paletteID], descriptor)
+	}
+
+	for _, byID := range projection {
+		for _, descriptors := range byID {
+			sort.SliceStable(descriptors, func(i, j int) bool {
+				a, b := descriptors[i], descriptors[j]
+				if a.sortKey() != b.sortKey() {
+					return a.sortKey() < b.sortKey()
+				}
+				return a.resref < b.resref
+			})
+		}
+	}
+	return projection, nil
+}
+
+func blueprintDescriptor(root gff.Struct, fileName, extension string, factions []string) (paletteDescriptor, uint8, bool) {
+	descriptor := paletteDescriptor{
+		strref:   noStrref,
+		resref:   strings.ToLower(fileName),
+		creature: extension == ".utc",
+	}
+	paletteID := -1
+
+	for _, field := range root.Fields {
+		switch field.Label {
+		case "PaletteID":
+			if v, ok := field.Value.(gff.ByteValue); ok {
+				paletteID = int(v)
+			}
+		case "TemplateResRef":
+			if v, ok := field.Value.(gff.ResRefValue); ok && v != "" {
+				descriptor.resref = strings.ToLower(string(v))
+			}
+		case "LocalizedName", "LocName", "FirstName":
+			if v, ok := field.Value.(gff.LocString); ok {
+				name, strref := locStringLabel(v)
+				if field.Label == "FirstName" {
+					descriptor.name = strings.TrimSpace(descriptor.name + " " + name)
+					if descriptor.strref == noStrref {
+						descriptor.strref = strref
+					}
+				} else {
+					descriptor.name = name
+					descriptor.strref = strref
+				}
+			}
+		case "LastName":
+			if v, ok := field.Value.(gff.LocString); ok {
+				name, _ := locStringLabel(v)
+				descriptor.name = strings.TrimSpace(descriptor.name + " " + name)
+			}
+		case "ChallengeRating":
+			if v, ok := field.Value.(gff.FloatValue); ok {
+				descriptor.cr = float32(v)
+			}
+		case "FactionID":
+			if v, ok := field.Value.(gff.WordValue); ok && int(v) < len(factions) {
+				descriptor.faction = factions[v]
+			}
+		}
+	}
+
+	if paletteID < 0 || paletteID == hiddenPaletteID {
+		return paletteDescriptor{}, 0, false
+	}
+	return descriptor, uint8(paletteID), true
+}
+
+func locStringLabel(value gff.LocString) (string, uint32) {
+	if value.StringRef != noStrref {
+		return "", value.StringRef
+	}
+	for _, entry := range value.Entries {
+		if entry.Value != "" {
+			return entry.Value, noStrref
+		}
+	}
+	return "", noStrref
+}
+
+func loadFactionNames(p *project.Project) ([]string, error) {
+	for _, rel := range p.Inventory.SourceFiles {
+		if !strings.HasSuffix(strings.ToLower(rel), "repute.fac.json") {
+			continue
+		}
+		abs := filepath.Join(p.SourceDir(), filepath.FromSlash(rel))
+		raw, err := os.ReadFile(abs)
+		if err != nil {
+			return nil, fmt.Errorf("read %s: %w", abs, err)
+		}
+		var document gff.Document
+		if err := json.Unmarshal(raw, &document); err != nil {
+			return nil, fmt.Errorf("parse gff json %s: %w", abs, err)
+		}
+		var names []string
+		for _, field := range document.Root.Fields {
+			if field.Label != "FactionList" {
+				continue
+			}
+			list, ok := field.Value.(gff.ListValue)
+			if !ok {
+				continue
+			}
+			for _, faction := range list {
+				name := ""
+				for _, f := range faction.Fields {
+					if f.Label == "FactionName" {
+						if v, ok := f.Value.(gff.StringValue); ok {
+							name = string(v)
+						}
+					}
+				}
+				names = append(names, name)
+			}
+		}
+		return names, nil
+	}
+	return nil, nil
+}
+
+// projectPaletteDocument strips every blueprint descriptor from the palette
+// tree and re-inserts the descriptors derived from module source. Category
+// structure (branches, terminal IDs, labels) passes through untouched.
+func projectPaletteDocument(document *gff.Document, byID map[uint8][]paletteDescriptor) {
+	for i, field := range document.Root.Fields {
+		if field.Label != "MAIN" {
+			continue
+		}
+		if list, ok := field.Value.(gff.ListValue); ok {
+			document.Root.Fields[i] = gff.NewField("MAIN", projectPaletteNodes(list, byID))
+		}
+	}
+}
+
+func projectPaletteNodes(nodes gff.ListValue, byID map[uint8][]paletteDescriptor) gff.ListValue {
+	out := make(gff.ListValue, 0, len(nodes))
+	for _, node := range nodes {
+		if structHasField(node, "RESREF") {
+			continue // blueprint descriptor — regenerated below
+		}
+
+		terminalID := -1
+		fields := make([]gff.Field, 0, len(node.Fields))
+		for _, field := range node.Fields {
+			if field.Label == "LIST" {
+				continue // rebuilt for terminals, recursed for branches
+			}
+			if field.Label == "ID" {
+				if v, ok := field.Value.(gff.ByteValue); ok {
+					terminalID = int(v)
+				}
+			}
+			fields = append(fields, field)
+		}
+
+		switch {
+		case terminalID >= 0:
+			if descriptors := byID[uint8(terminalID)]; len(descriptors) > 0 {
+				fields = append(fields, gff.NewField("LIST", descriptorList(descriptors)))
+			}
+		default:
+			for _, field := range node.Fields {
+				if field.Label == "LIST" {
+					if list, ok := field.Value.(gff.ListValue); ok {
+						fields = append(fields, gff.NewField("LIST", projectPaletteNodes(list, byID)))
+					}
+				}
+			}
+		}
+		out = append(out, gff.Struct{Type: node.Type, Fields: fields})
+	}
+	return out
+}
+
+func descriptorList(descriptors []paletteDescriptor) gff.ListValue {
+	list := make(gff.ListValue, 0, len(descriptors))
+	for _, d := range descriptors {
+		fields := make([]gff.Field, 0, 4)
+		if d.strref != noStrref {
+			fields = append(fields, gff.NewField("STRREF", gff.DWordValue(d.strref)))
+		} else {
+			fields = append(fields, gff.NewField("NAME", gff.StringValue(d.name)))
+		}
+		fields = append(fields, gff.NewField("RESREF", gff.ResRefValue(d.resref)))
+		if d.creature {
+			fields = append(fields, gff.NewField("CR", gff.FloatValue(d.cr)))
+			fields = append(fields, gff.NewField("FACTION", gff.StringValue(d.faction)))
+		}
+		list = append(list, gff.Struct{Fields: fields})
+	}
+	return list
+}
+
+func structHasField(s gff.Struct, label string) bool {
+	for _, field := range s.Fields {
+		if field.Label == label {
+			return true
+		}
+	}
+	return false
+}

internal/pipeline/palette_test.go

@@ -0,0 +1,191 @@
+package pipeline
+
+import (
+	"bytes"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"git.westgate.pw/ShadowsOverWestgate/sow-tools/internal/erf"
+	"git.westgate.pw/ShadowsOverWestgate/sow-tools/internal/gff"
+	"git.westgate.pw/ShadowsOverWestgate/sow-tools/internal/project"
+)
+
+const paletteTestSkeleton = `{
+  "file_type": "ITP ",
+  "file_version": "V3.2",
+  "root": {
+    "struct_type": 4294967295,
+    "fields": [
+      {
+        "label": "MAIN",
+        "type": "List",
+        "value": [
+          {
+            "struct_type": 0,
+            "fields": [
+              {"label": "STRREF", "type": "DWord", "value": 500},
+              {
+                "label": "LIST",
+                "type": "List",
+                "value": [
+                  {
+                    "struct_type": 0,
+                    "fields": [
+                      {"label": "STRREF", "type": "DWord", "value": 6699},
+                      {"label": "ID", "type": "Byte", "value": 23},
+                      {
+                        "label": "LIST",
+                        "type": "List",
+                        "value": [
+                          {
+                            "struct_type": 0,
+                            "fields": [
+                              {"label": "NAME", "type": "CExoString", "value": "Stale Junk"},
+                              {"label": "RESREF", "type": "ResRef", "value": "stalejunk"}
+                            ]
+                          }
+                        ]
+                      }
+                    ]
+                  },
+                  {
+                    "struct_type": 0,
+                    "fields": [
+                      {"label": "STRREF", "type": "DWord", "value": 6753},
+                      {"label": "ID", "type": "Byte", "value": 24}
+                    ]
+                  }
+                ]
+              }
+            ]
+          }
+        ]
+      }
+    ]
+  }
+}
+`
+
+func paletteTestItem(resref, name string, paletteID int) string {
+	return `{
+  "file_type": "UTI ",
+  "file_version": "V3.2",
+  "root": {
+    "struct_type": 4294967295,
+    "fields": [
+      {"label": "TemplateResRef", "type": "ResRef", "value": "` + resref + `"},
+      {
+        "label": "LocalizedName",
+        "type": "CExoLocString",
+        "value": {"string_ref": 4294967295, "entries": [{"id": 0, "value": "` + name + `"}]}
+      },
+      {"label": "PaletteID", "type": "Byte", "value": ` + itoa(paletteID) + `}
+    ]
+  }
+}
+`
+}
+
+func itoa(v int) string {
+	data, _ := json.Marshal(v)
+	return string(data)
+}
+
+func TestBuildProjectsPaletteDescriptorsAndExtractSkipsThem(t *testing.T) {
+	root := t.TempDir()
+	mustMkdir(t, filepath.Join(root, "src", "module"))
+	mustMkdir(t, filepath.Join(root, "src", "palettes"))
+	mustMkdir(t, filepath.Join(root, "src", "blueprints", "items"))
+	mustMkdir(t, filepath.Join(root, "build"))
+
+	mustWriteFile(t, filepath.Join(root, "nwn-tool.json"), `{
+  "module": {"name": "Test Module", "resref": "testmod"},
+  "paths": {"source": "src", "build": "build"}
+}
+`)
+	mustWriteFile(t, filepath.Join(root, "src", "module", "module.ifo.json"), `{
+  "file_type": "IFO ",
+  "file_version": "V3.2",
+  "root": {"struct_type": 4294967295, "fields": [{"label": "Mod_Name", "type": "CExoString", "value": "Test Module"}]}
+}
+`)
+	mustWriteFile(t, filepath.Join(root, "src", "palettes", "itempalcus.itp.json"), paletteTestSkeleton)
+	mustWriteFile(t, filepath.Join(root, "src", "blueprints", "items", "i_b.uti.json"), paletteTestItem("i_b", "Bravo Item", 23))
+	mustWriteFile(t, filepath.Join(root, "src", "blueprints", "items", "i_a.uti.json"), paletteTestItem("i_a", "Alpha Item", 23))
+	mustWriteFile(t, filepath.Join(root, "src", "blueprints", "items", "i_hidden.uti.json"), paletteTestItem("i_hidden", "Hidden Item", 255))
+
+	p, err := project.Load(root)
+	if err != nil {
+		t.Fatalf("load project: %v", err)
+	}
+	if err := p.Scan(); err != nil {
+		t.Fatalf("scan: %v", err)
+	}
+	if _, err := BuildModule(p); err != nil {
+		t.Fatalf("build: %v", err)
+	}
+
+	archiveFile, err := os.Open(p.ModuleArchivePath())
+	if err != nil {
+		t.Fatalf("open module archive: %v", err)
+	}
+	defer archiveFile.Close()
+	archive, err := erf.Read(archiveFile)
+	if err != nil {
+		t.Fatalf("read module archive: %v", err)
+	}
+
+	var palette *gff.Document
+	for _, resource := range archive.Resources {
+		if resource.Name == "itempalcus" {
+			document, err := gff.Read(bytes.NewReader(resource.Data))
+			if err != nil {
+				t.Fatalf("decode itempalcus: %v", err)
+			}
+			palette = &document
+		}
+	}
+	if palette == nil {
+		t.Fatal("itempalcus missing from built module")
+	}
+
+	canonical, err := json.Marshal(palette)
+	if err != nil {
+		t.Fatalf("marshal palette: %v", err)
+	}
+	text := string(canonical)
+	if bytes.Contains(canonical, []byte("stalejunk")) {
+		t.Fatalf("stale descriptor survived projection: %s", text)
+	}
+	for _, resref := range []string{"i_a", "i_b"} {
+		if !bytes.Contains(canonical, []byte(resref)) {
+			t.Fatalf("descriptor %s missing from projection: %s", resref, text)
+		}
+	}
+	if bytes.Contains(canonical, []byte("i_hidden")) {
+		t.Fatalf("PaletteID 255 blueprint leaked into projection: %s", text)
+	}
+	if a, b := bytes.Index(canonical, []byte("i_a")), bytes.Index(canonical, []byte("i_b")); a > b {
+		t.Fatalf("descriptors not name-sorted: %s", text)
+	}
+
+	if _, err := Compare(p); err != nil {
+		t.Fatalf("compare after build: %v", err)
+	}
+
+	// Extraction must never write palcus files back into source.
+	if err := os.Remove(filepath.Join(root, "src", "palettes", "itempalcus.itp.json")); err != nil {
+		t.Fatalf("remove skeleton: %v", err)
+	}
+	if err := p.Scan(); err != nil {
+		t.Fatalf("rescan: %v", err)
+	}
+	if _, err := Extract(p); err != nil {
+		t.Fatalf("extract: %v", err)
+	}
+	if _, err := os.Stat(filepath.Join(root, "src", "palettes", "itempalcus.itp.json")); !os.IsNotExist(err) {
+		t.Fatalf("extract wrote palcus file back (stat err: %v)", err)
+	}
+}

internal/project/project.go

@@ -31,7 +31,7 @@ var SourceExtensions = []string{
 }
 
 var AssetExtensions = []string{
-	".2da", ".bik", ".bmp", ".bmu", ".dds", ".dwk", ".gr2", ".itp", ".jpg", ".lod", ".lyt", ".mdb", ".mdl", ".mdx", ".mtr", ".plt", ".png", ".pwk", ".set", ".shd", ".tga", ".txi", ".uti", ".vis", ".wav", ".wlk", ".wok", ".xml",
+	".2da", ".bik", ".bmp", ".bmu", ".dds", ".dwk", ".itp", ".jpg", ".lod", ".lyt", ".mdl", ".mdx", ".mtr", ".plt", ".png", ".pwk", ".set", ".shd", ".tga", ".txi", ".uti", ".vis", ".wav", ".wok",
 }
 
 var BuiltinScriptPrefixes = []string{

scripts/prune-release-assets.sh

@@ -0,0 +1,67 @@
+#!/usr/bin/env bash
+# Delete the binary assets of every release except the newest KEEP ones.
+#
+# Gitea has no asset retention, so 9 assets x ~57 MB per tag pile up forever on
+# the host disk (sow-tools #52). Nothing consumes an old asset: CI takes the
+# Crucible binary from the Nix flake, and every deleted file is reproducible
+# from its tag. Tags, source archives, release notes and the releases
+# themselves are never touched — only attachments.
+#
+# Nothing is excluded, SHA256SUMS and the wrappers included: the checksums are
+# only meaningful next to the binaries they cover, and the drift-check reads
+# the wrappers from the latest release, which always stays complete.
+#
+# Best-effort by design: a stale asset is cheaper than a blocked release, so
+# every API failure is a warning, never a non-zero exit.
+#
+# Env:
+#   API    repo API base, e.g. https://git.example/api/v1/repos/owner/repo  [required]
+#   TOKEN  Gitea token with releases:write                                  [required]
+#   KEEP   how many newest releases stay complete (default 2)
+set -uo pipefail
+
+: "${API:?set API}"
+: "${TOKEN:?set TOKEN}"
+keep="${KEEP:-2}"
+auth="Authorization: token ${TOKEN}"
+per_page=50
+
+warn() { echo "prune-release-assets: $*" >&2; }
+
+# Walk every page so an old backlog is cleared, not only the tag that fell out
+# of the window on this run. Each page already embeds its releases' assets, so
+# no follow-up request per release is needed.
+releases='[]'
+page=1
+while :; do
+  body="$(curl -fsS -H "$auth" "${API}/releases?limit=${per_page}&page=${page}&draft=false")" || {
+    warn "listing releases failed on page ${page}; pruning what was listed so far"
+    break
+  }
+  count="$(jq 'length' <<<"$body")" || { warn "unreadable release page ${page}"; break; }
+  releases="$(jq -s 'add' <(printf '%s' "$releases") <(printf '%s' "$body"))"
+  # A short page is the last one; this also stops a server that ignores `page`.
+  (( count < per_page )) && break
+  page=$(( page + 1 ))
+done
+
+# Sort here rather than trusting the response order, and skip drafts in case
+# the server ignored `draft=false` — a draft must not consume a keep slot and
+# strip the binaries off the newest real release.
+mapfile -t stale < <(jq -r --argjson keep "$keep" '
+  [.[] | select(.draft != true)]
+  | sort_by(.created_at) | reverse | .[$keep:]
+  | .[] | "\(.id) \(.assets // [] | map(.id) | join(","))"
+' <<<"$releases")
+
+(( ${#stale[@]} )) || { echo "prune-release-assets: nothing older than the newest ${keep} releases"; exit 0; }
+
+for line in "${stale[@]}"; do
+  id="${line%% *}"
+  assets="${line#* }"
+  for asset in ${assets//,/ }; do
+    echo "deleting asset ${asset} of release ${id}"
+    curl -fsS -X DELETE -H "$auth" "${API}/releases/${id}/assets/${asset}" >/dev/null \
+      || warn "deleting asset ${asset} of release ${id} failed"
+  done
+done

tests/prune-release-assets.sh

@@ -0,0 +1,78 @@
+#!/usr/bin/env bash
+# Contract for scripts/prune-release-assets.sh, driven by a stub curl on PATH:
+# only assets of releases outside the keep window go, the window is decided by
+# release date rather than response order, drafts never consume a keep slot,
+# releases and tags stay, and an API failure never fails the run.
+set -euo pipefail
+
+repo_root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+script="$repo_root/scripts/prune-release-assets.sh"
+tmp="$(mktemp -d)"
+trap 'rm -rf "$tmp"' EXIT
+
+# Stub curl: one short page of releases, deliberately out of date order and
+# with a draft that is newer than every published release. Each release embeds
+# two assets. Every request is appended to $CALLS for the assertions below.
+mkdir -p "$tmp/bin"
+cat >"$tmp/bin/curl" <<'STUB'
+#!/usr/bin/env bash
+url="${!#}"
+printf '%s\n' "$*" >>"$CALLS"
+rel() { printf '{"id":%s,"created_at":"%s","draft":%s,"assets":[{"id":%s01},{"id":%s02}]}' "$1" "$2" "$3" "$1" "$1"; }
+case "$url" in
+  *releases\?*)
+    [[ "${FAIL_LIST:-0}" == 1 ]] && exit 22
+    printf '[%s,%s,%s,%s,%s]\n' \
+      "$(rel 12 2024-01-02T00:00:00Z false)" \
+      "$(rel 10 2024-01-04T00:00:00Z false)" \
+      "$(rel 9  2024-06-01T00:00:00Z true)" \
+      "$(rel 13 2024-01-01T00:00:00Z false)" \
+      "$(rel 11 2024-01-03T00:00:00Z false)"
+    ;;
+  */assets/*) [[ "${FAIL_DELETE:-0}" == 1 ]] && exit 22 ;;
+esac
+exit 0
+STUB
+chmod +x "$tmp/bin/curl"
+export PATH="$tmp/bin:$PATH" API=https://git.example/api/v1/repos/o/r TOKEN=t
+
+run() { CALLS="$tmp/calls" bash "$script" >"$tmp/out" 2>"$tmp/err"; }
+
+# Default keep=2: the two newest published releases (10, 11) stay whole, the
+# older two (12, 13) lose their assets, and the newer draft is ignored.
+: >"$tmp/calls"; run
+deletes="$(grep -c -- '-X DELETE' "$tmp/calls" || true)"
+[[ "$deletes" == 4 ]] || { echo "expected 4 asset deletes, got $deletes" >&2; exit 1; }
+grep -q 'assets/1201' "$tmp/calls" && grep -q 'assets/1302' "$tmp/calls" || {
+  echo "expected assets of releases 12 and 13 to be deleted" >&2; exit 1; }
+if grep -q 'releases/10/assets/\|releases/11/assets/' "$tmp/calls"; then
+  echo "kept releases lost assets" >&2; exit 1
+fi
+if grep -q 'releases/9/assets/' "$tmp/calls"; then
+  echo "a draft release was pruned" >&2; exit 1
+fi
+# A release or tag must never be deleted, only attachments under /assets/.
+if grep -- '-X DELETE' "$tmp/calls" | grep -qv '/assets/'; then
+  echo "a non-asset DELETE was issued" >&2; exit 1
+fi
+# A short page ends the sweep: no second page, so a server ignoring `page`
+# cannot spin the job until the job timeout.
+[[ "$(grep -c 'releases?' "$tmp/calls")" == 1 ]] || {
+  echo "a short release page did not end the sweep" >&2; exit 1; }
+
+# KEEP=4 covers every published release: nothing to delete.
+: >"$tmp/calls"; KEEP=4 run
+if grep -q -- '-X DELETE' "$tmp/calls"; then echo "KEEP=4 still deleted" >&2; exit 1; fi
+
+# Failures are warnings, never a non-zero exit.
+: >"$tmp/calls"; FAIL_DELETE=1 run || { echo "delete failure failed the run" >&2; exit 1; }
+grep -q 'failed' "$tmp/err" || { echo "delete failure was not warned about" >&2; exit 1; }
+
+: >"$tmp/calls"; FAIL_LIST=1 run || { echo "listing failure failed the run" >&2; exit 1; }
+grep -q 'listing releases failed' "$tmp/err" || {
+  echo "listing failure was not warned about" >&2; exit 1; }
+if grep -q -- '-X DELETE' "$tmp/calls"; then
+  echo "deleted assets despite an unreadable release list" >&2; exit 1
+fi
+
+echo "prune-release-assets: prunes by release date outside the keep window, ignores drafts, never deletes a release, never fails the build"

tests/workflow-contract.sh

@@ -46,6 +46,8 @@ grep -Fqx '  releases: write' "$release"
 grep -Fqx '    timeout-minutes: 30' "$release"
 [[ "$(grep -Fc "$checkout" "$release")" -eq 1 ]]
 grep -Fq 'secrets.GITEA_TOKEN' "$release"
+grep -Fq 'scripts/prune-release-assets.sh' "$release"
+grep -Fq 'continue-on-error: true' "$release"
 if grep -Fq 'secrets.GITHUB_TOKEN' "$release"; then
   echo "workflow-contract: release uses the GitHub token alias" >&2
   exit 1