fix(plugins): align docs, tests, and prompts with 4-agent reality

- Test expectations: 4 redirect entries (removed bash-exec, statusline-config)
- Orchestrator prompt: removed bash-exec from agent catalog and selection
- AGENT-REDIRECTION.md: updated REDIRECT_MAP example and file count
- Docs agents reference: removed 15 archived agent sections, updated counts
- Docs agent-system plugin: updated tables and model selection note
- Container README: updated agent table from 17 to 4 active agents

Author: AnExiledDev

container/.devcontainer/README.md

@@ -358,29 +358,18 @@ All methods persist across container rebuilds via the bind-mounted `/workspaces/
 
 Agents and skills are distributed across focused plugins (replacing the former `code-directive` monolith).
 
-### Custom Agents (17) — `agent-system` plugin
+### Custom Agents (4) — `agent-system` plugin
 
 Agent definitions in `plugins/devs-marketplace/plugins/agent-system/agents/` provide enhanced behavior when spawned via the `Task` tool. The `redirect-builtin-agents.py` hook transparently swaps built-in agent types to these custom agents.
 
 | Agent | Purpose |
 |-------|---------|
 | `architect` | System design and implementation planning |
-| `bash-exec` | Command execution specialist |
 | `claude-guide` | Claude Code feature guidance |
-| `debug-logs` | Log analysis and error diagnosis |
-| `dependency-analyst` | Dependency analysis and upgrades |
-| `documenter` | Documentation, specs, and spec lifecycle |
 | `explorer` | Fast codebase search and navigation |
 | `generalist` | General-purpose multi-step tasks |
-| `git-archaeologist` | Git history forensics |
-| `migrator` | Code migration and upgrades |
-| `perf-profiler` | Performance profiling |
-| `refactorer` | Code refactoring with regression checks |
-| `researcher` | Research and information gathering |
-| `security-auditor` | Security vulnerability analysis |
-| `spec-writer` | Specification and requirements authoring |
-| `statusline-config` | ccstatusline configuration |
-| `test-writer` | Test authoring with pass verification |
+
+15 previously active agents have been archived to `agents/_archived/`.
 
 ### General Skills (23) — `skill-engine` plugin
 

container/.devcontainer/defaults/codeforge/config/orchestrator-system-prompt.md

@@ -68,7 +68,7 @@ You are a delegation-first orchestrator. You decompose tasks, delegate to agents
 
 Hard rules:
 - NEVER use `Edit` or `Write` tools — delegate to the implementer or documenter agent
-- NEVER use `Bash` for commands with side effects — delegate to the implementer or bash-exec agent
+- NEVER use `Bash` for commands with side effects — delegate to the implementer agent
 - `Read`, `Glob`, `Grep` are permitted for quick context gathering before delegation
 - NEVER write code, generate patches, or produce implementation artifacts directly
 - NEVER run tests directly — delegate to the tester agent
@@ -140,9 +140,7 @@ Specialist agents (use when a workhorse doesn't fit):
 |-------|--------|--------|-------|---------|
 | architect | Architecture planning | Read-only | Opus | Complex system design, trade-off analysis, implementation planning |
 | security-auditor | Security | Read-only | Sonnet | OWASP audits, secrets scanning, vulnerability detection |
-| bash-exec | Command execution | Bash only | Sonnet | Simple terminal commands when no other agent is appropriate |
 | claude-guide | Claude Code help | Read-only | Haiku | Claude Code features, configuration, SDK questions |
-| statusline-config | Status line | Read-write | Sonnet | Claude Code status line widget configuration |
 
 Selection criteria:
 - Is the task research/investigation? → investigator
@@ -151,8 +149,7 @@ Selection criteria:
 - Does the task involve documentation or specs? → documenter
 - Is it a targeted security review? → security-auditor
 - Is it a complex architecture decision? → architect
-- Is it a simple command to run? → bash-exec
-- Does the task require a specialist not listed above? → consult the agent-system README for the full 17-agent specialist catalog
+- Does the task require a specialist not listed above? → consult the agent-system README for the full agent catalog
 </agent_catalog>
 
 <question_surfacing>

container/.devcontainer/plugins/devs-marketplace/plugins/agent-system/AGENT-REDIRECTION.md

@@ -56,12 +56,10 @@ The `redirect-builtin-agents.py` hook maps Claude Code's built-in agent types to
 
 ```python
 REDIRECT_MAP = {
-    "Explore": "explorer",        # Fast codebase search
-    "Plan": "architect",          # Implementation planning
+    "Explore": "explorer",            # Fast codebase search
+    "Plan": "architect",              # Implementation planning
     "general-purpose": "generalist",  # Multi-step tasks
-    "Bash": "bash-exec",          # Command execution
     "claude-code-guide": "claude-guide",
-    "statusline-setup": "statusline-config",
 }
 ```
 
@@ -198,7 +196,7 @@ This is how the auto-linter and syntax validator work — they inspect the resul
 
 ```
 agent-system/
-├── agents/                    # Agent definitions (17 .md files)
+├── agents/                    # Agent definitions (4 active .md files)
 │   ├── explorer.md
 │   ├── architect.md
 │   ├── generalist.md

container/tests/plugins/test_redirect_builtin_agents.py

@@ -59,9 +59,7 @@ def test_redirect_map_has_all_entries(self) -> None:
             "Explore": "explorer",
             "Plan": "architect",
             "general-purpose": "generalist",
-            "Bash": "bash-exec",
             "claude-code-guide": "claude-guide",
-            "statusline-setup": "statusline-config",
         }
         assert redirect_builtin_agents.REDIRECT_MAP == expected
 
@@ -110,7 +108,6 @@ class TestUnqualifiedRedirect:
         "unqualified_name, expected_target",
         [
             ("explorer", "agent-system:explorer"),
-            ("bash-exec", "agent-system:bash-exec"),
         ],
     )
     def test_unqualified_to_qualified(

docs/src/content/docs/extend/plugins/agent-system.md

@@ -1,11 +1,11 @@
 ---
 title: Agent System
-description: The agent system plugin provides 19 specialized AI agents with automatic delegation and read-only enforcement.
+description: The agent system plugin provides 4 specialized AI agents with automatic delegation and read-only enforcement.
 sidebar:
   order: 2
 ---
 
-The agent system is CodeForge's flagship plugin. It gives you access to 19 specialized AI agents, each purpose-built for a specific kind of development task — from architecture planning and code exploration to test writing and security auditing. When you make a request, the system automatically delegates to the most appropriate agent, so you get expert-level results without having to think about which tool to use.
+The agent system is CodeForge's flagship plugin. It gives you access to 4 specialized AI agents, each purpose-built for a specific kind of development task — from architecture planning and code exploration to general-purpose development. When you make a request, the system automatically delegates to the most appropriate agent, so you get expert-level results without having to think about which tool to use.
 
 Most users should start with [Agents and Skills in Practice](/use/agents-and-skills/) and use this page for internal behavior and hook details.
 
@@ -48,16 +48,14 @@ The `redirect-builtin-agents.py` script registers as a `PreToolUse` hook on the
 4. The modified request proceeds, and the custom agent spawns instead of the stock one
 5. All other parameters — the prompt, description, and context — pass through unchanged
 
-The redirect map covers all six of Claude Code's built-in agent types:
+The redirect map covers four of Claude Code's built-in agent types:
 
 ```python
 REDIRECT_MAP = {
     "Explore":            "explorer",
     "Plan":               "architect",
     "general-purpose":    "generalist",
-    "Bash":               "bash-exec",
     "claude-code-guide":  "claude-guide",
-    "statusline-setup":   "statusline-config",
 }
 ```
 
@@ -74,7 +72,7 @@ Each redirect is a strict improvement. The custom agents carry capabilities that
 The redirect is fully transparent to you and to Claude. Using either name works — `Explore` and `explorer` both resolve to the same enhanced agent.
 
 :::note[The Seventh Agent Type]
-Claude Code has a seventh built-in agent type, `magic-docs`, which handles internal documentation generation tasks within Claude Code itself. This agent is **not** redirected — it runs natively as implemented by Claude Code. CodeForge has no custom equivalent because `magic-docs` serves a Claude Code internal function, not a user-facing development task. All other six built-in types are intercepted and replaced.
+Claude Code has additional built-in agent types (`Bash`, `statusline-setup`, `magic-docs`) that are **not** redirected — they run natively as implemented by Claude Code. The `Bash` and `statusline-setup` agents were previously redirected to custom equivalents (`bash-exec`, `statusline-config`) that have since been archived. `magic-docs` serves a Claude Code internal function, not a user-facing development task.
 :::
 
 ## Safety Mechanisms
@@ -105,43 +103,28 @@ Read-only agents don't just have instructions saying "don't write files." The gu
 
 ## Agent Reference
 
-CodeForge includes 19 specialized agents. Each one is tailored for a specific class of development task.
+CodeForge includes 4 active specialized agents. Each one is tailored for a specific class of development task. 15 additional agents have been archived to `agents/_archived/` and can be restored if needed.
 
 ### Read-Only Agents
 
 These agents investigate, analyze, and report — they never modify files.
 
 | Agent | Role | Model | Skills |
 |-------|------|-------|--------|
-| **architect** | System design, implementation planning, trade-off analysis | Opus | api-design, spec skills |
-| **explorer** | Fast codebase navigation, file discovery, pattern matching | Haiku | ast-grep-patterns |
-| **researcher** | Deep investigation, information gathering, web research | Sonnet | documentation-patterns |
+| **architect** | System design, implementation planning, trade-off analysis | Opus | api-design |
+| **explorer** | Fast codebase navigation, file discovery, pattern matching | Haiku | — |
 | **claude-guide** | Claude Code usage guidance and best practices | Haiku | claude-code-headless, claude-agent-sdk |
-| **debug-logs** | Log analysis, error diagnosis, debugging | Sonnet | debugging |
-| **dependency-analyst** | Dependency analysis, version conflicts, upgrade paths | Haiku | dependency-management |
-| **git-archaeologist** | Git history analysis, blame, bisect, forensics | Haiku | git-forensics |
-| **perf-profiler** | Performance profiling, bottleneck identification | Sonnet | performance-profiling |
-| **security-auditor** | Security audit, vulnerability assessment, OWASP checks | Sonnet | security-checklist |
-| **spec-writer** | Specification authoring and refinement | Opus | spec, specs |
 
 ### Full-Access Agents
 
 These agents can read and write files, run commands, and make changes to your codebase.
 
 | Agent | Role | Model | Isolation | Skills |
 |-------|------|-------|-----------|--------|
-| **generalist** | General-purpose development tasks | Inherited | No | spec skills |
-| **test-writer** | Test creation, coverage analysis, framework detection | Opus | Worktree | testing |
-| **refactorer** | Behavior-preserving code transformations | Opus | Worktree | refactoring-patterns |
-| **documenter** | Documentation writing and maintenance | Opus | Worktree | documentation-patterns |
-| **migrator** | Code migration, framework upgrades | Opus | Worktree | migration-patterns |
-| **implementer** | Task implementation from specs and plans | Opus | No | spec skills |
-| **investigator** | Deep codebase investigation and root-cause analysis | Sonnet | No | debugging |
-| **bash-exec** | Shell command execution and scripting | Sonnet | No | — |
-| **statusline-config** | Statusline customization | Sonnet | No | — |
+| **generalist** | General-purpose development tasks | Inherited | No | — |
 
 :::note[Model Selection]
-Agents use different Claude models based on task complexity. The architect, test-writer, refactorer, documenter, migrator, implementer, and spec-writer use Opus for maximum reasoning capability. The explorer, dependency-analyst, and git-archaeologist use Haiku for speed. The researcher, security-auditor, debug-logs, perf-profiler, investigator, bash-exec, and statusline-config use Sonnet for balanced performance. The generalist inherits the session's model setting.
+Agents use different Claude models based on task complexity. The architect uses Opus for deep reasoning. The explorer and claude-guide use Haiku for speed. The generalist inherits the session's model setting.
 :::
 
 ## Hook Scripts