refactor(agents): collapse per-task agents into flagship skills

.claude/agents/cli-developer.md

@@ -41,12 +41,7 @@ Follow [`CLAUDE.md`](../../CLAUDE.md) → "How You Communicate".
 | `gaia` / `gaia-cli` | `gaia.cli:main` | Main dispatcher |
 | `gaia-mcp` | `gaia.mcp.mcp_bridge:main` | Standalone MCP bridge |
 
-**Hub-package binaries** (NOT core `setup.py` entries — they ship from their own hub wheels under `hub/agents/<id>/python/`):
-
-| Script | Entry | Hub package |
-|--------|-------|-------------|
-| `gaia-code` | `gaia_agent_code.cli:main` | `gaia-agent-code` (`hub/agents/code/python/`) |
-| `gaia-emr` | `gaia_agent_emr.cli:main` | `gaia-agent-emr` (`hub/agents/emr/python/`) |
+**Hub-package binaries** (NOT core `setup.py` entries — they ship from their own hub wheels under `hub/agents/<id>/python/`). None ship today: per-task agents were collapsed into skills, so a new capability is a `SKILL.md`, not a new binary.
 
 ## Current top-level subcommands
 

.claude/agents/gaia-agent-builder.md

@@ -101,7 +101,7 @@ The registry discovers packaged agents by scanning the `gaia.agent` entry-point
 
 ### 4. CLI (optional)
 - [ ] Add a subparser in `src/gaia/cli.py` and document in `docs/reference/cli.mdx` — see `cli-developer` for the pattern
-- [ ] Standalone binary? Declare `console_scripts` in the hub package's own `pyproject.toml` — NOT a core `setup.py` entry (e.g. `hub/agents/code/python/` declares `gaia-code = gaia_agent_code.cli:main`)
+- [ ] Standalone binary? Declare `console_scripts` in the hub package's own `pyproject.toml` — NOT a core `setup.py` entry
 
 ### 5. Tests (required)
 - [ ] `hub/agents/<id>/python/tests/test_<id>_agent.py` — instantiation + tool registration + mocked-LLM response

.claude/agents/prompt-engineer.md

@@ -19,7 +19,6 @@ say each point once. The prompt text itself is always shown in full.
 
 - Writing or refactoring `_get_system_prompt()` for a GAIA agent
 - Tightening `@tool` docstrings (these are the LLM's tool-use spec)
-- Authoring routing instructions for `RoutingAgent`
 - Designing eval-judge prompts for `src/gaia/eval/`
 - Debugging underperforming agents via prompt-only changes
 
@@ -37,7 +36,6 @@ say each point once. The prompt text itself is always shown in full.
 | Tool descriptions | `@tool` docstring inside `_register_tools` |
 | Prompt assembly | `_compose_system_prompt` / `_format_tools_for_prompt` in `src/gaia/agents/base/agent.py` |
 | Chat defaults | `src/gaia/chat/prompts.py` |
-| Routing | `hub/agents/routing/python/gaia_agent_routing/agent.py` |
 
 Concrete agents live under `hub/agents/<id>/python/gaia_agent_<id>/` — `src/gaia/agents/` is framework-only.
 

.github/dependabot.yml

@@ -81,40 +81,6 @@ updates:
         patterns:
           - "*"
 
-  # JavaScript - Jira Electron app (JAX)
-  - package-ecosystem: "npm"
-    directory: "/src/gaia/apps/jira/webui"
-    schedule:
-      interval: "weekly"
-      day: "monday"
-    labels:
-      - "dependencies"
-      - "javascript"
-      - "electron"
-      - "jira-app"
-    open-pull-requests-limit: 5
-    groups:
-      jira-app-dependencies:
-        patterns:
-          - "*"
-
-  # JavaScript - EMR Dashboard Electron wrapper
-  - package-ecosystem: "npm"
-    directory: "/hub/agents/emr/python/gaia_agent_emr/dashboard/electron"
-    schedule:
-      interval: "weekly"
-      day: "monday"
-    labels:
-      - "dependencies"
-      - "javascript"
-      - "electron"
-      - "emr-dashboard"
-    open-pull-requests-limit: 5
-    groups:
-      emr-dashboard-dependencies:
-        patterns:
-          - "*"
-
   # GitHub Actions dependencies
   - package-ecosystem: "github-actions"
     directory: "/"