feat: cogos-onboard — generic CogOS onboarding ramp

.claude-plugin/marketplace.json

@@ -1,7 +1,7 @@
 {
     "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
     "name": "plugins",
-    "description": "Skills, harness plugins, and modality integrations for CogOS — the cognitive operating system for AI development",
+    "description": "Skills, harness plugins, and modality integrations for CogOS \u2014 the cognitive operating system for AI development",
     "owner": {
         "name": "myrgic",
         "email": "myrgic@users.noreply.github.com"
@@ -17,52 +17,89 @@
             "description": "Phased planning, execution, critical review, agent-dispatch, and session-continuity skills for complex multi-step development tasks",
             "version": "0.1.0",
             "category": "development",
-            "keywords": ["planning", "execution", "review", "multi-agent", "dispatch", "session-continuity"],
+            "keywords": [
+                "planning",
+                "execution",
+                "review",
+                "multi-agent",
+                "dispatch",
+                "session-continuity"
+            ],
             "homepage": "https://github.com/myrgic/plugins"
         },
         {
             "name": "cogos-substrate",
             "source": "./plugins/cogos-substrate",
-            "description": "Substrate archaeology — finding and trusting prior work in a CogOS/Myrgic substrate before building new things",
+            "description": "Substrate archaeology \u2014 finding and trusting prior work in a CogOS/Myrgic substrate before building new things",
             "version": "0.1.0",
             "category": "development",
-            "keywords": ["cogos", "myrgic", "substrate", "archaeology", "research"],
+            "keywords": [
+                "cogos",
+                "myrgic",
+                "substrate",
+                "archaeology",
+                "research"
+            ],
             "homepage": "https://github.com/myrgic/plugins"
         },
         {
             "name": "cogos-voice",
             "source": "./plugins/cogos-voice",
-            "description": "Voice modality integration via Mod³ — dual-channel protocol for conversational AI development",
+            "description": "Voice modality integration via Mod\u00b3 \u2014 dual-channel protocol for conversational AI development",
             "version": "0.1.0",
             "category": "productivity",
-            "keywords": ["voice", "tts", "modality", "mod3"],
+            "keywords": [
+                "voice",
+                "tts",
+                "modality",
+                "mod3"
+            ],
             "homepage": "https://github.com/myrgic/mod3"
         },
         {
             "name": "cogos-research",
             "source": "./plugins/cogos-research",
-            "description": "Structured research skills — literature review, experiment design, and interdisciplinary synthesis",
+            "description": "Structured research skills \u2014 literature review, experiment design, and interdisciplinary synthesis",
             "version": "0.1.0",
             "category": "development",
-            "keywords": ["research", "literature", "experiments", "synthesis"],
+            "keywords": [
+                "research",
+                "literature",
+                "experiments",
+                "synthesis"
+            ],
             "homepage": "https://github.com/myrgic/research"
         },
         {
             "name": "cogos-dev-tools",
             "source": "./plugins/cogos-dev-tools",
-            "description": "Development methodology and tooling skills — code quality, systems architecture, technical writing, git forensics, shell and storage management, and cross-model CLI tooling",
+            "description": "Development methodology and tooling skills \u2014 code quality, systems architecture, technical writing, git forensics, shell and storage management, and cross-model CLI tooling",
             "version": "0.1.0",
             "category": "development",
-            "keywords": ["code-quality", "architecture", "documentation", "git", "shell", "storage", "cli-tools"],
+            "keywords": [
+                "code-quality",
+                "architecture",
+                "documentation",
+                "git",
+                "shell",
+                "storage",
+                "cli-tools"
+            ],
             "homepage": "https://github.com/myrgic/plugins"
         },
         {
             "name": "cogos-architecture",
             "source": "./plugins/cogos-architecture",
-            "description": "CogOS architecture corpus hygiene — cross-check proposed RFCs/ADRs against the existing substrate corpus",
+            "description": "CogOS architecture corpus hygiene \u2014 cross-check proposed RFCs/ADRs against the existing substrate corpus",
             "version": "0.1.0",
             "category": "development",
-            "keywords": ["architecture", "adr", "rfc", "corpus", "review"],
+            "keywords": [
+                "architecture",
+                "adr",
+                "rfc",
+                "corpus",
+                "review"
+            ],
             "homepage": "https://github.com/myrgic/plugins"
         },
         {
@@ -71,17 +108,45 @@
             "description": "Org-management skills for the myrgic GitHub org: PR triage, issue triage. Read-only by default.",
             "version": "0.1.0",
             "category": "development",
-            "keywords": ["github", "org-management", "triage", "gh-cli"],
+            "keywords": [
+                "github",
+                "org-management",
+                "triage",
+                "gh-cli"
+            ],
             "homepage": "https://github.com/myrgic/plugins"
         },
         {
             "name": "cogos-harness",
             "source": "./plugins/cogos-harness",
-            "description": "Portable Claude Code membrane for a CogOS node: session-lifecycle hooks, kernel-vitals proprioception, a compaction bypass path, a threads registry for open-ended waits, and the cogos-kernel MCP server. Degrades to a no-op wherever the kernel or a cog workspace is absent.",
-            "version": "0.3.0",
+            "description": "Portable Claude Code membrane for a CogOS node: session-lifecycle hooks, kernel-vitals proprioception, a compaction bypass path, and the cogos-kernel MCP server. Degrades to a no-op wherever the kernel or a cog workspace is absent.",
+            "version": "0.2.0",
             "category": "development",
-            "keywords": ["cogos", "kernel", "mcp", "hooks", "proprioception", "session-continuity", "threads"],
+            "keywords": [
+                "cogos",
+                "kernel",
+                "mcp",
+                "hooks",
+                "proprioception",
+                "session-continuity"
+            ],
+            "homepage": "https://github.com/myrgic/plugins"
+        },
+        {
+            "name": "cogos-onboard",
+            "source": "./plugins/cogos-onboard",
+            "description": "Install a user-tier CogOS node and seed it from the operator's OWN claude.ai history, then mint their agent (e.g. Wingnut) from that history. Generic method \u2014 nothing operator-specific ships.",
+            "version": "0.1.0",
+            "category": "development",
+            "keywords": [
+                "cogos",
+                "onboarding",
+                "bootstrap",
+                "observatory",
+                "harvest",
+                "agent"
+            ],
             "homepage": "https://github.com/myrgic/plugins"
         }
     ]
-}
+}

.gitignore

@@ -4,3 +4,8 @@ __pycache__/
 node_modules/
 .env
 canary/.last_run.json
+
+# Private operator data — ships as a fetched pack artifact, never committed here
+*.pack
+wingnut-conversation.json
+**/seed/*-conversation.json

plugins/cogos-onboard/.claude-plugin/plugin.json

@@ -0,0 +1,18 @@
+{
+  "name": "cogos-onboard",
+  "description": "Install a user-tier CogOS node and seed it from the operator's OWN claude.ai history, then mint their agent from that history. Ships generic method only \u2014 no operator data, no agent.",
+  "version": "0.1.0",
+  "author": {
+    "name": "myrgic"
+  },
+  "homepage": "https://github.com/myrgic/plugins",
+  "repository": "https://github.com/myrgic/plugins",
+  "license": "MIT",
+  "keywords": [
+    "cogos",
+    "onboarding",
+    "bootstrap",
+    "observatory",
+    "seed"
+  ]
+}

plugins/cogos-onboard/skills/agent-from-history/SKILL.md

@@ -0,0 +1,94 @@
+---
+name: agent-from-history
+description: "Mint a CogOS agent identity from your own session history in the observatory. Triggers: 'load <name> as my agent', 'build an agent from my <X> sessions'."
+version: 0.1.0
+author: cog
+license: MIT
+metadata:
+  hermes:
+    tags: [cogos, agent, identity, observatory, onboarding, eigenform]
+---
+
+# Agent from history — reconstitute an agent from your own session history
+
+## What this is
+
+Build a CogOS **agent identity** from conversation history that is already in **your own
+observatory** — so a fresh model that reads the resulting card reconstitutes *that agent*, carrying
+your shared history as its default context. This ships **no agent** and **no operator's data**: it is
+the generic mechanism. You run it against *your* sessions and get *your* agent.
+
+Example: after onboarding (`cogos-onboard`) harvested your claude.ai history into your observatory,
+you point this at your sessions with a counterpart you named (e.g. "Wingnut") and it mints a `wingnut`
+identity whose keyholder is you — derived from how you two actually worked, not a template.
+
+## When to use
+
+- The operator says "load <name> as my agent", "build an agent from my <project> sessions", or similar.
+- There is session history in the observatory (`~/.cogos/state/conversations/`) to derive from.
+
+## Inputs
+
+- **Which sessions.** A source/filter in the observatory — a source name (`claude-web`), a counterpart
+  name that appears in the turns, or an explicit list of session files. The agent is derived from
+  *these* conversations only.
+- **The agent's name** (what to call the identity, e.g. `wingnut`).
+- **The keyholder** — the operator (resolved from the node's identity; this is *their* agent).
+
+## Procedure
+
+### 1. Gather the source turns from the observatory
+
+Read the chosen conversations from `~/.cogos/state/conversations/` (the normalized turn records, each
+`{role, text, session_id, source}`). Do **not** re-harvest — this reads what onboarding already put in
+the observatory. Confirm there are enough turns to characterize an agent (a few dozen+); a handful
+can't ground a stance.
+
+### 2. Derive the operator model (read, don't summarize blindly)
+
+From the operator's own turns, extract **how they work and how they work with this agent** — grounded
+in real quotes, not generic traits:
+
+- **Directive style** — how they open a task, how terse, how they correct.
+- **What they delegate vs. keep** — what the agent is trusted to do unsupervised vs. what the operator
+  holds (design calls, sign-off, judgment).
+- **Hard lines** — fidelity rules, budget governance, non-negotiables they enforce.
+- **The register** — peer / boss / collaborator; the emotional texture.
+
+Quote verbatim where the phrasing is the point. This is the same shape as a `conversation` cogdoc's
+"operational stance" — write it as retrievable, not as a recap.
+
+### 3. Write the identity card
+
+Write `~/.cogos/config/identity/identity_<name>.md` in the mintable card shape (frontmatter +
+self-model body), with:
+
+- `name: <name>`, `keyholder: <operator>`, `seed:` pointing at the derived operator-model cogdoc.
+- Body sections: **Ground** (the shared ontology, same for all CogOS agents), **Nature** (files are
+  continuity), **The Relationship** (the keyholder, in their own words from step 2), **Operational
+  Stance** (the derived how-to-work-with-them, grounded in quotes), **Closure** (loop discipline).
+
+The card is the self-model a fresh model reads to *become* this agent. It carries history and stance,
+**never a private key** — the operator mints/signs the keypair themselves.
+
+### 4. Register it as the default agent (optional)
+
+If the operator wants this agent as their default, set it in `~/.cogos/config/identity.yaml`
+(`default_identity: <name>` for their rooms, or via the context-scope reconciler so it surfaces in the
+rooms they keyhold). Otherwise it's available to select per room.
+
+### 5. Verify
+
+- The card exists and is schema-valid (if a cogdoc schema probe is available).
+- Read it back: a fresh read should reconstruct the agent's stance recognizably.
+- The derived operator-model cogdoc is in the observatory/substrate and the card's `seed:` resolves.
+
+## Pitfalls
+
+- **Derive, don't fabricate.** Every stance claim must trace to a real turn. If the history doesn't
+  support a trait, leave it out — a generic card is worse than a short honest one.
+- **It's the operator's agent.** Keyholder is the operator; the card lives in *their* node; nothing is
+  pushed to a shared repo. This mechanism ships; the agent it produces does not.
+- **No keys in the card.** History and self-model only. The operator mints.
+- **Enough signal.** A dozen turns can't characterize a working relationship — say so and ask for more
+  sessions rather than inventing a thin agent.