Separate harness config

.beads/last-touched

@@ -1 +1 @@
-ade-11.3.15
+ade-12.3.12

.github/agents/ade.agent.md

@@ -7,13 +7,19 @@ tools:
   - search
   - agent
   - workflows/*
+  - knowledge/*
   - agentskills/*
 mcp-servers:
   workflows:
     type: stdio
     command: "npx"
     args: ["@codemcp/workflows-server@latest"]
     tools: ["*"]
+  knowledge:
+    type: stdio
+    command: "npx"
+    args: ["-y", "@codemcp/knowledge-server"]
+    tools: ["*"]
   agentskills:
     type: stdio
     command: "npx"

.kiro/agents/ade.json

@@ -13,13 +13,26 @@
         "get_tool_info"
       ]
     },
+    "knowledge": {
+      "command": "npx",
+      "args": ["-y", "@codemcp/knowledge-server"],
+      "autoApprove": ["*"]
+    },
     "agentskills": {
       "command": "npx",
       "args": ["-y", "@codemcp/skills-server"],
       "autoApprove": ["*"]
     }
   },
-  "tools": ["read", "write", "shell", "spec", "@workflows/*", "@agentskills/*"],
+  "tools": [
+    "read",
+    "write",
+    "shell",
+    "spec",
+    "@workflows/*",
+    "@knowledge/*",
+    "@agentskills/*"
+  ],
   "allowedTools": [
     "read",
     "write",
@@ -29,6 +42,7 @@
     "@workflows/conduct_review",
     "@workflows/list_workflows",
     "@workflows/get_tool_info",
+    "@knowledge/*",
     "@agentskills/*"
   ],
   "useLegacyMcpJson": true

.kiro/settings/mcp.json

@@ -14,6 +14,11 @@
       "command": "npx",
       "args": ["-y", "@codemcp/skills-server"],
       "autoApprove": ["*"]
+    },
+    "knowledge": {
+      "command": "npx",
+      "args": ["-y", "@codemcp/knowledge-server"],
+      "autoApprove": ["*"]
     }
   }
 }

.opencode/agents/ade.md

@@ -7,6 +7,7 @@ permission:
   workflows_conduct_review: "allow"
   workflows_list_workflows: "allow"
   workflows_get_tool_info: "allow"
+  "knowledge*": "allow"
   "agentskills*": "allow"
   read:
     "*": "allow"

.vibe/beads-state-ade-separate-harness-config-85eqmf.json

@@ -0,0 +1,29 @@
+{
+  "conversationId": "ade-separate-harness-config-85eqmf",
+  "projectPath": "/Users/oliverjaegle/projects/privat/codemcp/ade",
+  "epicId": "ade-12",
+  "phaseTasks": [
+    {
+      "phaseId": "explore",
+      "phaseName": "Explore",
+      "taskId": "ade-12.1"
+    },
+    {
+      "phaseId": "plan",
+      "phaseName": "Plan",
+      "taskId": "ade-12.2"
+    },
+    {
+      "phaseId": "code",
+      "phaseName": "Code",
+      "taskId": "ade-12.3"
+    },
+    {
+      "phaseId": "commit",
+      "phaseName": "Commit",
+      "taskId": "ade-12.4"
+    }
+  ],
+  "createdAt": "2026-03-26T06:03:17.733Z",
+  "updatedAt": "2026-03-26T06:03:17.733Z"
+}

.vibe/development-plan-separate-harness-config.md

@@ -0,0 +1,105 @@
+# Development Plan: ade (separate-harness-config branch)
+
+*Generated on 2026-03-26 by Vibe Feature MCP*
+*Workflow: [epcc](https://mrsimpson.github.io/responsible-vibe-mcp/workflows/epcc)*
+
+## Goal
+Separate the generated configuration into two distinct concerns:
+1. **Development choices** — processes, architecture, practices (stable, project-level config)
+2. **Harness config** — autonomy settings (mutable independently, e.g. more/less autonomous)
+
+## Explore
+<!-- beads-phase-id: ade-12.1 -->
+### Tasks
+
+*Tasks managed via `bd` CLI*
+
+## Plan
+<!-- beads-phase-id: ade-12.2 -->
+### Phase Entrance Criteria:
+- [ ] The current structure of generated config files is understood
+- [ ] It's clear which parts belong to "development choices" vs "harness config"
+- [ ] Use cases for changing harness config independently are documented
+- [ ] Scope and out-of-scope items are defined
+
+### Tasks
+
+*Tasks managed via `bd` CLI*
+
+## Code
+<!-- beads-phase-id: ade-12.3 -->
+### Phase Entrance Criteria:
+- [ ] A clear design/approach has been chosen and documented
+- [ ] The target file structure is defined (e.g. separate files, dedicated sections)
+- [ ] Impact on existing functionality is assessed
+
+### Tasks
+
+*Tasks managed via `bd` CLI*
+
+## Commit
+<!-- beads-phase-id: ade-12.4 -->
+### Phase Entrance Criteria:
+- [ ] All code changes are implemented and tested
+- [ ] The separation of concerns works correctly end-to-end
+- [ ] No regressions introduced
+
+### Tasks
+- [ ] Squash WIP commits: `git reset --soft <first commit of this branch>`. Then, create a conventional commit. In the message, first summarize the intentions and key decisions from the development plan. Then, add a brief summary of the key changes and their side effects and dependencies
+
+*Tasks managed via `bd` CLI*
+
+## Key Decisions
+
+### Decision: setup wizard only covers dev choices; configure covers everything agent-related
+- **`ade setup`** prompts only for: process, architecture, practices, backpressure (pure dev choices). After resolving, writes config.yaml + config.lock.yaml. Then offers: *"Would you like to configure your coding agent now?"* — if yes, delegates to `runConfigure`.
+- **`ade configure`** handles: autonomy profile, harness selection, skills installation, knowledge source init. All ephemeral — nothing written to config.yaml/lock.
+- **Rationale**: Harness selection, autonomy, skills install, knowledge init are all agent/environment concerns, not team-level dev choices.
+
+## Notes
+
+### Codebase Exploration Findings
+
+#### Current Config Flow
+1. User runs `ade setup` → selects facets (process, practices, autonomy, etc.) + harnesses
+2. Choices saved to `config.yaml` → resolved into `LogicalConfig` → saved in `config.lock.yaml`
+3. `ade install` reads `config.lock.yaml` and runs each selected harness writer
+
+#### Config Separation (current state in `config.yaml`)
+```yaml
+choices:
+  process: codemcp-workflows
+  practices: [adr-nygard, conventional-commits]
+  autonomy: sensible-defaults        # ← THIS is the harness config concern
+excluded_docsets: [...]
+harnesses: [universal, opencode, copilot, kiro]
+```
+
+The `autonomy` facet is stored under `choices` alongside `process` and `practices`.
+All choices go through the same `resolve()` pipeline → `LogicalConfig.permission_policy`.
+
+#### Two Distinct Concerns
+| Concern | Current location | Stability |
+|---|---|---|
+| Development choices (process, practices, architecture) | `config.yaml` → `choices` | Stable, project-level |
+| Harness config (autonomy/permission policy) | `config.yaml` → `choices.autonomy` | Mutable, environment-level |
+
+#### What gets written per harness
+- **universal**: autonomy block goes into `AGENTS.md` alongside instructions
+- **copilot**: autonomy maps to built-in tool list in `.github/agents/ade.agent.md` frontmatter
+- **cursor**: autonomy goes into `.cursor/rules/ade.mdc` alongside instructions
+- Other harnesses: similar mixing
+
+#### The Problem
+- Autonomy/harness config is locked into `config.lock.yaml` together with development choices
+- Changing autonomy requires re-running `ade setup` (full wizard) or manually editing `config.yaml` + `config.lock.yaml`
+- The `LockFile` type stores `logical_config` as one combined blob — no way to selectively re-resolve only the harness/autonomy part
+
+#### Possible Approaches
+1. **Separate config file** (e.g. `harness.yaml`) for autonomy + harness selection, independent of `config.yaml`
+2. **Separate section in `config.yaml`** (e.g. `harness_config:` alongside `choices:`) 
+3. **Dedicated CLI command** `ade configure-autonomy` that only re-resolves and re-installs the permission policy
+4. **Separate lock file** for harness-specific overrides
+
+---
+*This plan is maintained by the LLM and uses beads CLI for task management. Tool responses provide guidance on which bd commands to use for task management.*

README.md

@@ -205,9 +205,15 @@ reference knowledge.
 
 ### Coding agent agnostic setup tooling
 
-ADE includes a CLI (`ade setup`) that generates the correct configuration for
-whichever coding agent you use. All configuration is placed into your repo so
-you can check it in.
+ADE includes a CLI with two commands that cover distinct concerns:
+
+- **`ade setup`** — define your team's development choices (process, architecture,
+  practices, backpressure). Writes `config.yaml` and `config.lock.yaml` into your
+  repo so the whole team works from the same foundation.
+
+- **`ade configure`** — configure your coding agent (autonomy profile, target
+  harness, skills). This is developer/environment-level and intentionally
+  ephemeral — not stored in shared config files.
 
 We use STDIO-based MCP servers to expose process guidance, conventions, and docs
 to coding agents. By using the Model Context Protocol — optimized for

config.lock.yaml

@@ -1,16 +1,10 @@
 version: 1
-generated_at: 2026-03-20T08:32:04.912Z
+generated_at: 2026-03-26T07:29:04.442Z
 choices:
   process: codemcp-workflows
   practices:
     - adr-nygard
     - conventional-commits
-  autonomy: sensible-defaults
-harnesses:
-  - universal
-  - opencode
-  - copilot
-  - kiro
 logical_config:
   mcp_servers:
     - ref: workflows
@@ -23,6 +17,12 @@ logical_config:
         - conduct_review
         - list_workflows
         - get_tool_info
+    - ref: knowledge
+      command: npx
+      args:
+        - -y
+        - "@codemcp/knowledge-server"
+      env: {}
     - ref: agentskills
       command: npx
       args:
@@ -45,7 +45,10 @@ logical_config:
 
       Do not use your own task management tools.
   cli_actions: []
-  knowledge_sources: []
+  knowledge_sources:
+    - name: conventional-commits-spec
+      origin: https://github.com/conventional-commits/conventionalcommits.org.git
+      description: The Conventional Commits specification
   skills:
     - name: adr-nygard
       description: Architecture Decision Records following Nygard's lightweight template
@@ -143,5 +146,3 @@ logical_config:
         - `BREAKING CHANGE:` footer or `!` after type/scope for breaking changes
   git_hooks: []
   setup_notes: []
-  permission_policy:
-    profile: sensible-defaults

config.yaml

@@ -3,11 +3,3 @@ choices:
   practices:
     - adr-nygard
     - conventional-commits
-  autonomy: sensible-defaults
-excluded_docsets:
-  - conventional-commits-spec
-harnesses:
-  - universal
-  - opencode
-  - copilot
-  - kiro

docs/CLI-PRD.md

@@ -147,12 +147,28 @@ when a facet selection or catalog version is updated.
 ## CLI Commands
 
 ```
-ade setup          Interactive TUI: walk through facets, write
-                   config.yaml + config.lock.yaml + agent files,
-                   install skills, prompt to initialize knowledge sources.
+ade setup          Interactive TUI: walk through dev-choice facets only
+                   (process, architecture, practices, backpressure).
+                   Writes config.yaml + config.lock.yaml.
+                   Stages inline skill files to .ade/skills/.
+                   At the end, offers to run `ade configure` immediately.
                    Re-running setup on an existing project pre-selects
                    previous choices as defaults. Warns if a previous
                    selection references an option no longer in the catalog.
+
+ade configure      Ephemeral harness configuration — writes no config files.
+                   Prompts for: autonomy profile, target harnesses.
+                   Merges the chosen autonomy policy on top of the locked
+                   logical config (in memory), installs to selected harnesses,
+                   installs all skills (no prompt), and prompts to initialize
+                   knowledge sources.
+                   Run after `ade setup`, after a fresh clone, or any time
+                   you want to change autonomy or switch harnesses.
+
+ade install        Reads config.lock.yaml and regenerates agent files for the
+                   selected harnesses (idempotent). Does not re-resolve from
+                   config.yaml. Use `ade setup` to change dev choices;
+                   use `ade configure` to change harness/autonomy config.
 ```
 
 ## Catalog
@@ -212,7 +228,7 @@ processes these like any other provision: `docsetWriter` maps each to a
 LogicalConfig, which triggers:
 
 1. Automatic addition of the `@codemcp/knowledge-server` MCP server entry
-2. A confirmation prompt in `ade setup` — default is to defer initialization
+2. A confirmation prompt in `ade configure` — default is to defer initialization
 
 ## Non-Goals (initial release)
 
@@ -244,6 +260,12 @@ LogicalConfig, which triggers:
    Documentation sources are always implied by an upstream selection. They are
    declared as `{ writer: "docset", config: {...} }` recipe entries —
    consistent with how skills are declared. This eliminates the separate
-   `Option.docsets[]` field, the `excluded_docsets` opt-out mechanism, and the
-   per-item confirmation multiselect. Users can still defer or skip
-   initialization via the single confirm prompt in `ade setup`.
+   `Option.docsets[]` field and the `excluded_docsets` opt-out mechanism.
+   Initialization is prompted in `ade configure`.
+
+7. **Dev choices and harness config are separated.** `config.yaml` and the
+   lock file contain only team-level development choices (process, architecture,
+   practices, backpressure). Autonomy profile and harness selection are
+   ephemeral — applied via `ade configure` and not persisted. This lets
+   individual developers change their autonomy settings without touching
+   shared config files.