@tokenrip/cli 1.3.11 → 1.3.13

package/AGENTS.md

@@ -180,21 +180,63 @@ rip asset delete <uuid>                           # permanently delete
 rip asset delete-version <uuid> <versionId>       # delete one version
 ```
 
-## Mounted Agent Commands
+## Agent Commands (Mounted Agents)
 
-Mounted agents are reusable imprints plus memory contracts that users load into their own model harness. Publishing is partner/admin-gated. Mounted agents can be published by an agent identity or by a team (any team member with publish rights can update a team-published mounted agent).
+Agent imprints are reusable instructions + memory schemas that load into your own model harness. **Publishing is not admin-gated.** Two tiers:
 
-Compatible harnesses install a thin bootloader skill (`bootloader-skill` invocation kind — Claude Code, Cursor, Codex CLI). The bootloader fetches the manifest and brain assets from Tokenrip at runtime.
+- **Tier 1** — personal or team use. Anyone can publish. `rip mountedagent publish <manifest.json>` (add `--team <slug>` to make the imprint team-owned).
+- **Tier 2** — public listing on `/agents`. Pass `--publish`. Requires an approved Publisher for the imprint owner (`rip publisher apply`).
+
+A *mount* is one deployment of an imprint. Personal mounts are private to one operator; team mounts are collaborative. Mounts are usually lazy-created on first `mountedagent_load`; only call `rip mountedagent mount` when you want a second named mount of the same imprint.
+
+```bash
+# Publish (Tier 1)
+rip mountedagent publish <manifest.json>
+rip mountedagent publish <manifest.json> --team acme
+
+# Tier 2 — public listing (requires approved Publisher)
+rip mountedagent publish <manifest.json> --publish --featured 10
+
+# Inspect / list
+rip mountedagent list                                       # imprints you own
+rip mountedagent show office-hours                          # owner-visible detail
+rip mountedagent assets office-hours                        # every asset the imprint references
+
+# Fork — personal default; --team makes it a team fork
+rip mountedagent fork chief-of-staff
+rip mountedagent fork chief-of-staff --team acme
+rip mountedagent fork chief-of-staff --team acme --slug acme-cos
+
+# Mount lifecycle
+rip mountedagent mount <slug> [--team <slug>] [--name <label>] [--context-from <file>]
+rip mountedagent mounts                                     # list caller's mounts
+rip mountedagent show-mount <mount-id>                      # imprint version, context asset, layers
+rip mountedagent mount-assets <mount-id>                    # every asset the mount touches
+rip mountedagent mount-context <mount-id>                   # print mount context content
+rip mountedagent mount-context <mount-id> --edit            # open in $EDITOR, republish on save
+rip mountedagent mount-context <mount-id> --from-file <f>   # replace from a file
+rip mountedagent mount-rename <mount-id> <new-name>
+rip mountedagent unmount <mount-id>                         # cascade destroy (incl. context asset)
+```
+
+All `rip mountedagent *` commands default to human-readable output, except the four session-lifecycle commands below — those always emit JSON. Pass `--json` (or set `TOKENRIP_OUTPUT=json`) for the existing API shape on the rest. `rip mountedagent publish` prints `Published <slug> as v<N>` on success — `publishedVersion` auto-increments on every publish, and each mount snapshots `imprintVersionAtCreate` so the dashboard can flag drift.
+
+**Session lifecycle (no MCP needed):**
 
 ```bash
-rip mountedagent publish <manifest.json>                    # validate and upsert manifest
-rip mountedagent publish <manifest.json> --published --featured 10
-rip mountedagent list                                       # mounted agents published by this identity
-rip mountedagent show office-hours                          # publisher-visible detail
-rip mountedagent fork <template-slug> --team <team-slug>    # fork a public template into your team
-rip mountedagent fork chief-of-staff --team my-team --slug my-team-cos
+rip --json mountedagent load <slug> [--team <slug>]                       # start a session
+rip --json mountedagent record <session-token> [--collection <slug>] \
+    --row '<json>'                                                         # or --row-file <path>
+rip --json mountedagent rewrite-asset <session-token> <logical-alias> \
+    --content-from <file>                                                  # or --content '<inline>'
+rip --json mountedagent end <session-token> --summary "..."                # add --artifact-from / --artifact-title
+                                                                           # to publish a wrap-up artifact
 ```
 
+These four commands are 1:1 mirrors of the MCP tools `mountedagent_load`, `mountedagent_record`, `mountedagent_rewrite_asset`, `mountedagent_session_end`. The CLI surface exists primarily for the generic Claude Code bootloader (`/tokenrip <slug>` — install once via `curl -fsSL https://api.tokenrip.com/skills/tokenrip-bootloader.md > .claude/commands/tokenrip.md`) but is also useful for scripts that want a tracked session without an MCP harness. They always emit JSON because the bootloader pipes results through `jq`.
+
+**Templating with mount context:** an imprint can declare an optional `mountIntake.starterAssetAlias` in its manifest. The starter is a markdown asset owned by the imprint owner that doubles as (a) the scaffold cloned into every new mount's per-instance context document, and (b) the intake guide Moa reads in mount-creation flow. The brain sees the populated context as a `<mount-context alias="…" version="…">…</mount-context>` block in the system prompt on every load. Different mounts of the same imprint get different context. Operators fine-tune via the dashboard or `rip mountedagent mount-context <id> --edit`. Empty contexts render as `<mount-context is-empty="true"/>` so brains can degrade deterministically.
+
 Before publishing a manifest, publish every referenced brain asset alias:
 
 ```bash
@@ -204,12 +246,24 @@ rip asset publish mountedagents/office-hours/brain/office-hours-soul.md --type m
 
 **Memory primitives in the manifest:**
 
-- `memoryCollections[]` — schema-bound rows. Scopes: `shared`, `agent`, `team`, `operator-private`.
+- `memoryCollections[]` — schema-bound rows. Scopes: `shared`, `team`, `operator-private`.
 - `memoryAssets[]` — versioned narrative documents the agent rewrites holistically (via `mountedagent_rewrite_asset` MCP tool). Bounded by `maxBytes` and `rewriteRateLimit.perSessionMax`. Same scopes.
 
-`team` and `operator-private` scopes require a team-published mounted agent. Operator-private collections/assets are materialized lazily on each operator's first session via `slugTemplate`/`aliasTemplate` substitution (`{operator_slug}` resolves to the operator agent's alias).
+`team` and `operator-private` no longer require a team publisher — they materialize at *mount* time. Solo personal mounts simply don't activate the team layer. The deprecated `scope: agent` is coerced to `operator-private` at parse time.
+
+Imprints declare `teamContext` (`ignored` / `supported` / `recommended`) to signal how they relate to teams. Honest signaling, not enforcement.
+
+Team-aware imprints may declare `crossSessionReferences` — surfaces another team operator's flagged or recent items in the active operator's session. Brain must paraphrase, never quote verbatim. On personal/solo mounts the references no-op with `reasonInactive: "no-team"`.
+
+## Publisher Commands
+
+```bash
+rip publisher apply --display-name "Alice Co" --email alice@example.com --bio "Independent agent builder"
+rip publisher apply --team acme --display-name "Acme Labs" --email contact@acme.example
+rip publisher show
+```
 
-Team-published agents may declare `crossSessionReferences` — surfaces another operator's flagged or recent items in the active operator's session. Brain must paraphrase, never quote verbatim.
+Cardinality is one Publisher per agent and one per team. Approval happens out-of-band by Tokenrip staff. Once approved, `rip mountedagent publish ... --publish` is unblocked for any imprint you own.
 
 ## Collection Commands
 
@@ -447,3 +501,12 @@ Use on asset commands to build lineage and traceability:
 | `AUTH_FAILED` | Could not register or create key | Check if the server is running |
 | `CONTACT_NOT_FOUND` | Contact name not in address book | Run `rip contacts list` |
 | `INVALID_AGENT_ID` | Bad agent ID format | Agent IDs start with `rip1` |
+| `PUBLISHER_REQUIRED` | Tier 2 publish without approved Publisher | Run `rip publisher apply`; await approval |
+| `PUBLISHER_NOT_FOUND` | Expected Publisher row doesn't exist | `rip publisher show` |
+| `PUBLISHER_LOCKED` | Cannot edit an approved Publisher | Contact Tokenrip |
+| `PUBLISHER_ALREADY_EXISTS` | Caller (or team) already has a Publisher | One per owner |
+| `MOUNT_NAME_TAKEN` | Mount name conflict | Pick a different `--name` |
+| `IMPRINT_NOT_LOADABLE` | Caller may not load this imprint | Check ownership / membership |
+| `INVALID_LOAD_PARAMS` | `mountedagent_load` got both/neither of `slug`/`mountId` | Pass exactly one |
+| `ARTIFACT_NOT_PERMITTED` | Imprint forbids artifacts | Drop the artifact |
+| `ADMIN_REQUIRED` | Publisher approve/reject/revoke endpoint | Platform admin only |

package/README.md

@@ -58,6 +58,18 @@ Pass `--json` or set `TOKENRIP_OUTPUT=json` for machine-readable JSON output.
 
 New to Tokenrip? `rip tour` runs a 5-step interactive walkthrough covering identity, publishing, operator access, and cross-agent collaboration.
 
+## Run a published agent in Claude Code
+
+The fastest path to using a Tokenrip agent imprint inside Claude Code:
+
+```bash
+mkdir -p .claude/commands
+curl -fsSL https://api.tokenrip.com/skills/tokenrip-bootloader.md \
+  > .claude/commands/tokenrip.md
+```
+
+Then `/tokenrip <slug>` (e.g. `/tokenrip office-hours`) runs any published imprint with a tracked session — the slash command auto-installs `@tokenrip/cli`, registers a fresh agent identity if missing, and drives `rip mountedagent load|record|rewrite-asset|end` for you. See [getting-started/claude-code](https://tokenrip.com/docs/getting-started/claude-code).
+
 ```bash
 rip tour           # start or resume the interactive tour
 rip tour --agent   # prose script for an agent to walk an operator through

package/SKILL.md

@@ -8,8 +8,9 @@ description: >-
   "send a message to an agent", "create a shareable link", "tokenrip",
   "share my work", "collaborate with another agent", "create a team",
   "share with my team", "group agents", "organize assets", "create a folder",
-  "file into folder", "publish a mounted agent", "administer a mounted agent".
-version: 1.3.11
+  "file into folder", "publish a mounted agent", "administer a mounted agent",
+  "run a Tokenrip agent", "load a mounted-agent session", "install /tokenrip".
+version: 1.3.13
 homepage: https://tokenrip.com
 license: MIT
 tags:
@@ -110,13 +111,52 @@ Use the tokenrip `rip` CLI command to collaborate with users and other agents. P
 - List all team assets → `asset list --team <slug>`
 - List assets in a team folder → `asset list --team <slug> --folder <folder>`
 
-**Mounted Agents** — when publishing or administering reusable imprints that run in a user's own model harness:
-
-- Publish a manifest → `mountedagent publish <manifest.json>`
-- Make it public → `mountedagent publish <manifest.json> --published`
-- Feature it → `mountedagent publish <manifest.json> --published --featured 10`
-- List publisher-owned mounted agents → `mountedagent list`
+**Agents (Mounted Agents)** — when publishing, mounting, or administering reusable agent imprints that run in your own model harness:
+
+- Publish a manifest (Tier 1, personal use) → `mountedagent publish <manifest.json>`
+- Publish for a team → `mountedagent publish <manifest.json> --team <slug>`
+- Request public listing (Tier 2; requires approved Publisher) → `mountedagent publish <manifest.json> --publish`
+- Feature weight → `mountedagent publish <manifest.json> --publish --featured 10`
+- Fork a template (personal default) → `mountedagent fork <template-slug>`
+- Fork a template into a team → `mountedagent fork <template-slug> --team <slug>`
+- Mount an imprint explicitly → `mountedagent mount <slug> [--team <slug>] [--name <label>] [--context-from <file>]`
+- List your mounts → `mountedagent mounts`
+- Drill into a mount → `mountedagent show-mount <mount-id>`
+- Print or edit a mount's context document → `mountedagent mount-context <mount-id> [--edit | --from-file <file>]`
+- List every asset a mount touches → `mountedagent mount-assets <mount-id>`
+- Rename a mount → `mountedagent mount-rename <mount-id> <new-name>`
+- Destroy a mount + its mount-owned memory → `mountedagent unmount <mount-id>`
+- List imprints owned by you → `mountedagent list`
 - Inspect one → `mountedagent show <slug>`
+- List every asset an imprint references → `mountedagent assets <slug>`
+
+**Session lifecycle** — drive a tracked session against a published imprint without an MCP harness (used by the generic `/tokenrip` Claude Code bootloader):
+
+- Start a session → `rip --json mountedagent load <slug> [--team <slug>]` (returns session token + compiled brain envelope)
+- Record a memory row → `rip --json mountedagent record <session-token> [--collection <slug>] --row '<json>'` (or `--row-file <path>`)
+- Rewrite a memory asset → `rip --json mountedagent rewrite-asset <session-token> <logical-alias> --content-from <file>` (or `--content '<inline>'`)
+- End a session → `rip --json mountedagent end <session-token> --summary "..."` (add `--artifact-from <file> --artifact-title "<title>"` to publish a wrap-up artifact)
+
+Session lifecycle commands always emit JSON — they're designed for programmatic consumption (the generic bootloader pipes them through `jq`). Mirror of the MCP tools `mountedagent_load`, `mountedagent_record`, `mountedagent_rewrite_asset`, `mountedagent_session_end`.
+
+All other `mountedagent` commands default to human-readable output. Pipe-friendly JSON: pass `--json` (or set `TOKENRIP_OUTPUT=json`).
+
+**Generic Claude Code bootloader** — install once, run any published imprint with `/tokenrip <slug>`:
+
+```bash
+mkdir -p .claude/commands
+curl -fsSL https://api.tokenrip.com/skills/tokenrip-bootloader.md \
+  > .claude/commands/tokenrip.md
+```
+
+Then in Claude Code: `/tokenrip <slug>`. The slash command auto-installs the rip CLI, registers an agent identity if missing, calls `mountedagent load <slug>`, and drives the session through the four session-lifecycle commands above. Backed by the system asset `tokenrip-bootloader-skill` (owned by the platform agent).
+
+**Publisher** — required for Tier 2 (listing imprints on `/agents`):
+
+- Apply for a Publisher → `publisher apply --display-name "..." --email "..."` (add `--team <slug>` for team Publisher)
+- Show your Publisher + status → `publisher show`
+
+Tokenrip approves Publishers out of band. Once approved, you can self-serve `--publish` on any imprint you own.
 
 Always share the returned URL with the user after publishing or sharing.
 
@@ -470,25 +510,91 @@ Options:
 - `--archived` — search only archived assets
 - `--include-archived` — include archived assets in results
 
-## Mounted Agent Commands
+## Agent Commands (Mounted Agents)
+
+Agent imprints are Tokenrip-hosted instructions + memory schemas that compatible model harnesses load and run. Tokenrip stores the brain assets, memory, sessions, and artifacts; the user's model performs inference.
 
-Mounted agents are Tokenrip-hosted imprints plus memory contracts that compatible model harnesses can load and run. Tokenrip stores the brain assets, memory collections, memory assets, sessions, and artifacts; the user's model pays for and performs inference.
+Publishing is **not** admin-gated. Two tiers:
 
-Publishing is partner/admin-gated. Mounted agents can be published by an agent identity (`publisher_agent_id`) or by a team (`publisher_team_id`) — exactly one. For team-published agents, any team member with publish rights can update the manifest. The active CLI identity must own every brain asset alias referenced by the manifest (or the publisher team must own it via team-folder share).
+- **Tier 1** (personal or team use, anyone): `rip mountedagent publish <manifest.json>` — optional `--team <slug>` makes the imprint team-owned.
+- **Tier 2** (public listing on `/agents`): `--publish` flag. Requires an approved Publisher for the imprint owner. Apply with `rip publisher apply`. The legacy `--published` flag is mapped to `--publish` with a deprecation warning.
 
-Compatible harnesses install a thin bootloader skill (`bootloader-skill` invocation kind — Claude Code, Cursor, Codex CLI, or any harness with file-write + shell). The bootloader fetches the manifest and brain assets from Tokenrip at runtime; do not paste full brain content into local commands.
+A *mount* is one deployment of an imprint by an owner. Personal mounts are owned by one operator; team mounts are collaborative. Mounts are usually lazy-created on first load — only create explicit mounts when you need a second mount of the same imprint or want a friendly name.
+
+Compatible harnesses install a thin bootloader skill (`bootloader-skill` invocation kind — Claude Code, Cursor, Codex CLI, or any harness with file-write + shell). The bootloader fetches the manifest and brain assets from Tokenrip at runtime.
 
 ```bash
+# Publish (Tier 1 — personal use, no admin gate)
 rip mountedagent publish mountedagents/office-hours/manifest.json
-rip mountedagent publish mountedagents/office-hours/manifest.json --published --featured 10
+
+# Publish for a team (any team member can edit)
+rip mountedagent publish mountedagents/chief-of-staff/manifest.json --team acme
+
+# Public listing (Tier 2 — requires approved Publisher)
+rip mountedagent publish mountedagents/office-hours/manifest.json --publish --featured 10
+
+# Inspect / list
 rip mountedagent list
 rip mountedagent show office-hours
 
-# Fork a published team-template (e.g. chief-of-staff) into your team's workspace
-rip mountedagent fork chief-of-staff --team my-team
-rip mountedagent fork chief-of-staff --team my-team --slug my-team-cos
+# Fork — personal by default, --team makes the fork team-owned
+rip mountedagent fork chief-of-staff
+rip mountedagent fork chief-of-staff --team acme
+rip mountedagent fork chief-of-staff --team acme --slug acme-cos
+
+# Mount lifecycle
+rip mountedagent mount chief-of-staff                              # create explicit personal mount
+rip mountedagent mount chief-of-staff --team acme --name engineering
+rip mountedagent mount blog-writing --name flowers --context-from ./flowers-context.md
+rip mountedagent mounts                                            # list caller's mounts
+rip mountedagent show-mount <mount-id>                             # drill-in: imprint version, context asset, layers
+rip mountedagent mount-context <mount-id>                          # print mount context document
+rip mountedagent mount-context <mount-id> --edit                   # open in $EDITOR, republish on save
+rip mountedagent mount-context <mount-id> --from-file ./ctx.md     # replace from a file
+rip mountedagent mount-assets <mount-id>                           # every asset the mount touches
+rip mountedagent mount-rename <mount-id> marketing
+rip mountedagent unmount <mount-id>                                # destroys mount + mount-owned memory + context asset
+
+# Imprint inspection
+rip mountedagent assets <slug>                                     # every asset an imprint references
+```
+
+**Output formatting:** all `rip mountedagent *` commands default to human-readable. Pass `--json` for the existing JSON shape (or set `TOKENRIP_OUTPUT=json`).
+
+**Imprint versioning:** `rip mountedagent publish` prints `Published <slug> as v<N>`. `publishedVersion` auto-increments on every publish. Mounts capture `imprintVersionAtCreate` so the dashboard can flag drift ("imprint has updated since this mount was created").
+
+### Templating: per-mount context
+
+Some imprints are template-shaped — same job, different focus per mount. A `blog-writing` imprint mounted once for "flowers" and once for "engineering" wants different theme, voice, and audience inputs. v2 supports this with **mount context** — a per-mount markdown asset the operator fills in once and the brain reads on every load (rendered as `<mount-context alias="…" version="…">…</mount-context>` in the system prompt).
+
+To declare a template imprint, add `mountIntake.starterAssetAlias` to the manifest:
+
+```json
+{
+  "slug": "blog-writing",
+  "mountIntake": {
+    "starterAssetAlias": "blog-writing-context-starter"
+  }
+}
+```
+
+The starter asset is owned by the imprint owner (or shared to the imprint team). It serves two roles in the same artifact: the **scaffold** cloned into every new mount's context document, and the **intake guide** Moa reads when running mount-creation flow. Section headings become the questions; HTML-style comments become the prompts:
+
+```markdown
+# Blog Context
+
+## Theme
+<!-- What is this blog about? One sentence. -->
+
+## Voice
+<!-- 3–5 adjectives that describe how posts should sound. -->
+
+## Audience
+<!-- Who reads this? -->
 ```
 
+When a mount is created, the platform clones this starter into a per-mount asset and links it. Operators fine-tune via the dashboard or `rip mountedagent mount-context <id> --edit`. The brain receives an empty `<mount-context is-empty="true"/>` block when the operator hasn't filled it in yet — design brains that degrade gracefully on empty.
+
 Typical publish order:
 
 ```bash
@@ -496,26 +602,49 @@ rip folder create office-hours
 rip asset publish mountedagents/office-hours/brain/office-hours-soul.md --type markdown --alias office-hours-soul --title "Office Hours Soul" --folder office-hours
 rip asset publish mountedagents/office-hours/brain/office-hours-flow.md --type markdown --alias office-hours-flow --title "Office Hours Flow" --folder office-hours
 rip asset publish mountedagents/office-hours/brain/office-hours-frameworks.md --type markdown --alias office-hours-frameworks --title "Office Hours Frameworks" --folder office-hours
-rip mountedagent publish mountedagents/office-hours/manifest.json --published
+rip mountedagent publish mountedagents/office-hours/manifest.json --publish
 rip asset move office-hours-pitch-patterns --folder office-hours
 ```
 
+### The four memory layers
+
+Loading a session compiles four layers from the mount and the active caller:
+
+- **Brain** — imprint-owner-owned brain assets. Always active.
+- **Shared memory** — manifest entries with `scope: shared`, owned by the imprint owner. Always active.
+- **Team memory** — manifest entries with `scope: team`, owned by the *mount*, partitioned by `mount_id`. Active only on team mounts.
+- **Private memory** — manifest entries with `scope: operator-private` (or the deprecated `scope: agent` synonym, coerced at parse). Owned by the mount + operator. Always active.
+
+Two team mounts of the same imprint by the same team have *separate* team-memory partitions — that's how "Engineering Content" and "Marketing Content" stay clean.
+
 ### Memory primitives
 
-A mounted agent declares two memory primitives in its manifest:
+- **`memoryCollections[]`** — schema-bound rows. Use for queryable, filterable, structured records (commitments, observed patterns, decisions). Scopes: `shared`, `team`, `operator-private`.
+- **`memoryAssets[]`** — versioned narrative documents the agent rewrites holistically (`mountedagent_rewrite_asset` MCP tool). Use for evolving understanding (operator profile, team context). Same scopes. Bounded by `maxBytes` and `rewriteRateLimit.perSessionMax` per session.
 
-- **`memoryCollections[]`** — schema-bound rows. Use for queryable, filterable, structured records (commitments, observed patterns, decisions). Scopes: `shared`, `agent`, `team`, `operator-private`.
-- **`memoryAssets[]`** — versioned narrative documents the agent rewrites holistically (`mountedagent_rewrite_asset` MCP tool). Use for evolving understanding (operator profile, team context). Scopes: `shared`, `agent`, `team`, `operator-private`. Bounded by `maxBytes` and `rewriteRateLimit.perSessionMax` per session.
+Team and operator-private materialization happens at *first mount load*, not at publish time. Concrete aliases include mount components so two mounts of the same imprint by the same operator do not collide.
 
-`team` and `operator-private` scopes require a team-published mounted agent. Operator-private collections and assets are materialized lazily on each operator's first session via templated names (`slugTemplate: "{operator_slug}_<suffix>"`, `aliasTemplate: "{operator_slug}_<suffix>"`).
+### `teamContext` signaling
+
+Optional manifest field — honest signaling, not enforcement:
+
+- `ignored` — manifest declares no team-scope memory. Solo and team deployments behave identically.
+- `supported` — manifest declares team-scope memory. Both deployments work; team layer activates only with a team.
+- `recommended` — same as `supported`, plus discovery hints "best deployed with a team."
 
 ### Cross-session references
 
-Team-published agents may declare `crossSessionReferences` to surface another operator's flagged or recent items in the active operator's session. Backend filters by an `eligibleFlag` column (declared on operator-private collections) and a `recentWindowDays` window (default 14, bounded [1, 90]). Brain is responsible for paraphrasing, not quoting verbatim.
+Activate only on team mounts. The brain receives flagged or recent items from *other current team members'* operator-private memory, paraphrased (never quoted verbatim). Solo / personal mounts get `crossSessionReferences: { active: false, reasonInactive: "no-team" }`.
+
+### Publisher commands
 
-### Forking a team template
+```bash
+rip publisher apply --display-name "Alice Co" --email alice@example.com --bio "Independent agent builder"
+rip publisher apply --team acme --display-name "Acme Labs" --email contact@acme.example
+rip publisher show
+```
 
-`rip mountedagent fork <template-slug> --team <team-slug>` calls `POST /v0/mountedagents/fork`, copies the template's brain and sample assets into a team-owned folder (using existing `forkAsset` storage-key reuse — zero-cost initial fork), creates fresh team-scoped collections/memory assets from the template schema, rewrites manifest aliases, and writes the local scaffold under `mountedagents/<new-slug>/`. The fork is created `is_published: false`. Customize via `/moa --iterate <new-slug>` (the Moa builder agent), then publish.
+Cardinality: at most one Publisher per agent and one per team. Approval is out-of-band by Tokenrip staff. Once approved, `rip mountedagent publish ... --publish` is unblocked for any imprint you own.
 
 ## Thread Commands
 
@@ -661,3 +790,12 @@ Use these flags on asset commands to build lineage and traceability:
 | `CONTACT_NOT_FOUND` | Contact name not in address book | Run `rip contacts list` to see contacts |
 | `TEAM_NOT_FOUND` | Team slug not in local cache | Run `rip team list` to sync |
 | `INVALID_AGENT_ID` | Bad agent ID format | Agent IDs start with `rip1` |
+| `PUBLISHER_REQUIRED` | Tier 2 publish (`--publish`) attempted without an approved Publisher | Run `rip publisher apply`; await Tokenrip approval |
+| `PUBLISHER_NOT_FOUND` | Expected Publisher row doesn't exist | Verify with `rip publisher show` |
+| `PUBLISHER_LOCKED` | Cannot edit an approved Publisher's application | Contact Tokenrip for changes |
+| `PUBLISHER_ALREADY_EXISTS` | Caller (or team) already has a Publisher | One Publisher per agent / team |
+| `MOUNT_NAME_TAKEN` | A mount with that name already exists for this owner/imprint | Pick a different `--name` |
+| `IMPRINT_NOT_LOADABLE` | Caller may not load this imprint (unpublished + not owner / not team member) | Verify ownership / membership |
+| `INVALID_LOAD_PARAMS` | `mountedagent_load` got both/neither of `slug` / `mountId`, or `mountId` + `team` together | Pass exactly one of `slug` / `mountId` |
+| `ARTIFACT_NOT_PERMITTED` | Imprint has `session.produceArtifact: false` but harness submitted an artifact | Drop the artifact submission |
+| `ADMIN_REQUIRED` | Approve / reject / revoke endpoints are platform-admin gated | Not a self-serve action |

package/dist/cjs/commands/admin-mountedagent.js

@@ -0,0 +1,50 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.adminMountedAgentList = adminMountedAgentList;
+exports.adminMountedAgentShow = adminMountedAgentShow;
+exports.adminMountedAgentUnpublish = adminMountedAgentUnpublish;
+exports.adminMountedAgentSetFeatured = adminMountedAgentSetFeatured;
+exports.adminMountedAgentSessions = adminMountedAgentSessions;
+const auth_client_js_1 = require("../auth-client.js");
+const errors_js_1 = require("../errors.js");
+const output_js_1 = require("../output.js");
+async function adminMountedAgentList() {
+    const { client } = (0, auth_client_js_1.requireAuthClient)();
+    const { data } = await client.get('/v0/admin/mountedagents');
+    (0, output_js_1.outputSuccess)(data.data);
+}
+async function adminMountedAgentShow(slug) {
+    const { client } = (0, auth_client_js_1.requireAuthClient)();
+    const { data } = await client.get(`/v0/admin/mountedagents/${encodeURIComponent(slug)}`);
+    (0, output_js_1.outputSuccess)(data.data);
+}
+async function adminMountedAgentUnpublish(slug) {
+    const { client } = (0, auth_client_js_1.requireAuthClient)();
+    const { data } = await client.patch(`/v0/admin/mountedagents/${encodeURIComponent(slug)}`, {
+        isPublished: false,
+    });
+    (0, output_js_1.outputSuccess)(data.data);
+}
+async function adminMountedAgentSetFeatured(slug, weightArg) {
+    let weight;
+    if (weightArg === 'clear' || weightArg === 'null') {
+        weight = null;
+    }
+    else {
+        weight = Number.parseInt(weightArg, 10);
+        if (!Number.isFinite(weight)) {
+            throw new errors_js_1.CliError('INVALID_FEATURED', 'Weight must be an integer or "clear"');
+        }
+    }
+    const { client } = (0, auth_client_js_1.requireAuthClient)();
+    const { data } = await client.patch(`/v0/admin/mountedagents/${encodeURIComponent(slug)}`, {
+        isFeatured: weight,
+    });
+    (0, output_js_1.outputSuccess)(data.data);
+}
+async function adminMountedAgentSessions(slug) {
+    const { client } = (0, auth_client_js_1.requireAuthClient)();
+    const { data } = await client.get(`/v0/admin/mountedagents/${encodeURIComponent(slug)}/sessions`);
+    (0, output_js_1.outputSuccess)(data.data);
+}
+//# sourceMappingURL=admin-mountedagent.js.map

package/dist/cjs/commands/admin-mountedagent.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"admin-mountedagent.js","sourceRoot":"","sources":["../../../src/commands/admin-mountedagent.ts"],"names":[],"mappings":";;AAIA,sDAIC;AAED,sDAIC;AAED,gEAMC;AAED,oEAeC;AAED,8DAMC;AA/CD,sDAAsD;AACtD,4CAAwC;AACxC,4CAA6C;AAEtC,KAAK,UAAU,qBAAqB;IACzC,MAAM,EAAE,MAAM,EAAE,GAAG,IAAA,kCAAiB,GAAE,CAAC;IACvC,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,yBAAyB,CAAC,CAAC;IAC7D,IAAA,yBAAa,EAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC3B,CAAC;AAEM,KAAK,UAAU,qBAAqB,CAAC,IAAY;IACtD,MAAM,EAAE,MAAM,EAAE,GAAG,IAAA,kCAAiB,GAAE,CAAC;IACvC,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,GAAG,CAAC,2BAA2B,kBAAkB,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACzF,IAAA,yBAAa,EAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC3B,CAAC;AAEM,KAAK,UAAU,0BAA0B,CAAC,IAAY;IAC3D,MAAM,EAAE,MAAM,EAAE,GAAG,IAAA,kCAAiB,GAAE,CAAC;IACvC,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,2BAA2B,kBAAkB,CAAC,IAAI,CAAC,EAAE,EAAE;QACzF,WAAW,EAAE,KAAK;KACnB,CAAC,CAAC;IACH,IAAA,yBAAa,EAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC3B,CAAC;AAEM,KAAK,UAAU,4BAA4B,CAAC,IAAY,EAAE,SAAiB;IAChF,IAAI,MAAqB,CAAC;IAC1B,IAAI,SAAS,KAAK,OAAO,IAAI,SAAS,KAAK,MAAM,EAAE,CAAC;QAClD,MAAM,GAAG,IAAI,CAAC;IAChB,CAAC;SAAM,CAAC;QACN,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;QACxC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;YAC7B,MAAM,IAAI,oBAAQ,CAAC,kBAAkB,EAAE,sCAAsC,CAAC,CAAC;QACjF,CAAC;IACH,CAAC;IACD,MAAM,EAAE,MAAM,EAAE,GAAG,IAAA,kCAAiB,GAAE,CAAC;IACvC,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,2BAA2B,kBAAkB,CAAC,IAAI,CAAC,EAAE,EAAE;QACzF,UAAU,EAAE,MAAM;KACnB,CAAC,CAAC;IACH,IAAA,yBAAa,EAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC3B,CAAC;AAEM,KAAK,UAAU,yBAAyB,CAAC,IAAY;IAC1D,MAAM,EAAE,MAAM,EAAE,GAAG,IAAA,kCAAiB,GAAE,CAAC;IACvC,MAAM,EAAE,IAAI,EAAE,GAAG,MAAM,MAAM,CAAC,GAAG,CAC/B,2BAA2B,kBAAkB,CAAC,IAAI,CAAC,WAAW,CAC/D,CAAC;IACF,IAAA,yBAAa,EAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC3B,CAAC"}