diff --git a/src/tools.zig b/src/tools.zig
index 818a005..2d38b97 100644
--- a/src/tools.zig
+++ b/src/tools.zig
@@ -218,11 +218,11 @@ pub const tools_list =
     \\{"name":"run_reviewer","description":"Invoke the Codex reviewer subagent on the current branch. Checks errdefer gaps, RwLock ordering, Zig 0.15.x API misuse, and missing test coverage. Returns the agent's full findings.","inputSchema":{"type":"object","properties":{"prompt":{"type":"string","description":"Override the default review prompt"},"timeout_seconds":{"type":"integer","description":"Maximum time for agent execution (default 300, max 600)"}},"required":[]}},
     \\{"name":"run_explorer","description":"Invoke the Codex explorer subagent to trace execution paths through the codebase. Read-only — maps affected code paths and gathers evidence without proposing fixes.","inputSchema":{"type":"object","properties":{"prompt":{"type":"string","description":"What to explore, e.g. 'trace how get_next_task flows through gh.zig'"},"timeout_seconds":{"type":"integer","description":"Maximum time for agent execution (default 300, max 600)"}},"required":["prompt"]}},
     \\{"name":"run_zig_infra","description":"Invoke the Codex zig_infra subagent to review build.zig module graph, named @import wiring, and test step coverage.","inputSchema":{"type":"object","properties":{"prompt":{"type":"string","description":"Override the default build wiring check prompt"}},"required":[]}},
-    \\{\"name\":\"run_swarm\",\"description\":\"Spawn a self-organizing swarm of parallel sub-agents to tackle a task. Provider-agnostic: resolves the best backend (Claude/Codex) based on the model/mode you specify. An orchestrator decomposes the task into sub-tasks, up to max_agents run concurrently via Zig threads, and a synthesis agent combines their outputs. Set writable=true to allow agents to edit files (for bug fixes, refactors). Best for broad research, multi-file analysis, multi-angle reviews, or parallel bug fixing.\",\"inputSchema\":{\"type\":\"object\",\"properties\":{\"prompt\":{\"type\":\"string\",\"description\":\"The high-level task for the swarm to solve\"},\"title\":{\"type\":\"string\",\"description\":\"Short human-readable label shown during execution\"},\"max_agents\":{\"type\":\"integer\",\"description\":\"Maximum parallel sub-agents (default 5, hard cap 100)\"},\"writable\":{\"type\":\"boolean\",\"description\":\"Allow agents to edit files and run shell commands (default false = read-only analysis)\"},\"model\":{\"type\":\"string\",\"description\":\"Model alias or full ID for all swarm agents (default: auto-resolved per role). Use \\\\\\\"opus\\\\\\\" for hardest tasks, \\\\\\\"haiku\\\\\\\" for fast/cheap parallel work.\"},\"mode\":{\"type\":\"string\",\"enum\":[\"smart\",\"rush\",\"deep\",\"free\"],\"description\":\"Agent mode applied to workers and synthesis agent (default: smart). Orchestrator uses rush unless overridden.\"},\"per_agent_model\":{\"type\":\"object\",\"description\":\"Per-role model overrides as a JSON object mapping role names to model aliases/IDs. Example: {\\\\\"finder\\\\\":\\\\\"haiku\\\\\",\\\\\"fixer\\\\\":\\\\\"opus\\\\\"}. Takes precedence over the top-level model param.\"},\"telemetry_out\":{\"type\":\"string\",\"description\":\"Optional file path to write telemetry JSON (cost, tokens, wall time, parallelism)\"}},\"required\":[\"prompt\"]}},
-    \\{\"name\":\"run_agents\",\"description\":\"Run multiple agents in parallel within a single tool call. Each element of the agents array is a run_agent spec (same fields as run_agent). All agents execute concurrently via Zig threads; results are collected and returned as a JSON array once every agent completes. Use this instead of multiple sequential run_agent calls when the tasks are independent.\",\"inputSchema\":{\"type\":\"object\",\"properties\":{\"agents\":{\"type\":\"array\",\"description\":\"Array of agent specs to run in parallel\",\"items\":{\"type\":\"object\",\"properties\":{\"prompt\":{\"type\":\"string\",\"description\":\"The task or question for this agent\"},\"model\":{\"type\":\"string\",\"description\":\"Model alias or full ID (default: auto-resolved)\"},\"role\":{\"type\":\"string\",\"description\":\"Agent role: finder, reviewer, fixer, explorer, architect, orchestrator, synthesizer, monitor\"},\"mode\":{\"type\":\"string\",\"enum\":[\"smart\",\"rush\",\"deep\",\"free\"]},\"writable\":{\"type\":\"boolean\"},\"allowed_tools\":{\"type\":\"string\"},\"permission_mode\":{\"type\":\"string\",\"enum\":[\"default\",\"acceptEdits\",\"bypassPermissions\"]},\"cwd\":{\"type\":\"string\"}},\"required\":[\"prompt\"]}}},\"required\":[\"agents\"]}},
-    \\{\"name\":\"review_fix_loop\",\"description\":\"Iterative review-fix-review loop. Runs a read-only reviewer to find issues, then a writable agent to fix them, then re-reviews. Repeats until the reviewer reports no remaining issues or max_iterations is reached. Returns a JSON object with iteration history and convergence status.\",\"inputSchema\":{\"type\":\"object\",\"properties\":{\"prompt\":{\"type\":\"string\",\"description\":\"Override the default review criteria\"},\"max_iterations\":{\"type\":\"integer\",\"description\":\"Maximum review-fix cycles (default 3, max 5)\"}},\"required\":[]}},
-    \\{\"name\":\"run_agent\",\"description\":\"Run a single agent turn. Provider-agnostic: resolves the best backend (Claude/Codex) based on mode, role, and available providers. The primitive layer — use run_task for smart multi-step execution.\",\"inputSchema\":{\"type\":\"object\",\"properties\":{\"prompt\":{\"type\":\"string\",\"description\":\"The task or question for the agent\"},\"model\":{\"type\":\"string\",\"description\":\"Model alias or full ID (default: claude-sonnet-4-6). Use \\\"opus\\\" for hardest tasks, \\\"haiku\\\" for fast/cheap.\"},\"role\":{\"type\":\"string\",\"description\":\"Agent role: finder, reviewer, fixer, explorer, architect, orchestrator, synthesizer, monitor\"},\"mode\":{\"type\":\"string\",\"enum\":[\"smart\",\"rush\",\"deep\",\"free\"],\"description\":\"Agent mode: smart (Sonnet), rush (Haiku), deep (Opus), free (Haiku)\"},\"allowed_tools\":{\"type\":\"string\",\"description\":\"Comma-separated tool allowlist, e.g. \\\"Bash,Read,Edit\\\". Omit to allow all tools.\"},\"permission_mode\":{\"type\":\"string\",\"enum\":[\"default\",\"acceptEdits\",\"bypassPermissions\"],\"description\":\"Permission mode for file and shell operations\"},\"writable\":{\"type\":\"boolean\",\"description\":\"Allow file writes (maps to bypassPermissions when permission_mode is unset)\"},\"cwd\":{\"type\":\"string\",\"description\":\"Working directory override (default: current repo path)\"}},\"required\":[\"prompt\"]}},
-    \\{\"name\":\"run_task\",\"description\":\"Smart executor: analyzes a task, picks the right strategy and agents, runs them with appropriate roles and models. Use this instead of run_agent for multi-step tasks. Supports chain presets (finder_fixer, reviewer_fixer, explore_report, architect_build) or auto-selection.\",\"inputSchema\":{\"type\":\"object\",\"properties\":{\"task\":{\"type\":\"string\",\"description\":\"Task description — what needs to be done\"},\"preset\":{\"type\":\"string\",\"enum\":[\"finder_fixer\",\"reviewer_fixer\",\"explore_report\",\"architect_build\",\"custom\"],\"description\":\"Chain preset (default: auto-select based on task)\"},\"mode\":{\"type\":\"string\",\"enum\":[\"smart\",\"rush\",\"deep\",\"free\"],\"description\":\"Agent mode for all agents in the chain\"},\"max_agents\":{\"type\":\"integer\",\"description\":\"Max agents to spawn (default: preset-determined)\"},\"writable\":{\"type\":\"boolean\",\"description\":\"Override write access (default: role-determined)\"},\"permission_mode\":{\"type\":\"string\",\"enum\":[\"default\",\"acceptEdits\",\"bypassPermissions\"],\"description\":\"Permission mode for file and shell operations\"},\"timeout_seconds\":{\"type\":\"integer\",\"description\":\"Maximum total time for the full chain (default 300, max 600)\"},\"model\":{\"type\":\"string\",\"description\":\"Model alias or full ID for all chain agents (default: auto-resolved per role). Use \\\\\"opus\\\\\" for hardest tasks, \\\\\"haiku\\\\\" for fast/cheap.\"}},\"required\":[\"task\"]}}
+    \\{"name":"run_swarm","description":"Spawn a self-organizing swarm of parallel sub-agents to tackle a task. Provider-agnostic: resolves the best backend (Claude/Codex) based on the model/mode you specify. An orchestrator decomposes the task into sub-tasks, up to max_agents run concurrently via Zig threads, and a synthesis agent combines their outputs. Set writable=true to allow agents to edit files (for bug fixes, refactors). Best for broad research, multi-file analysis, multi-angle reviews, or parallel bug fixing.","inputSchema":{"type":"object","properties":{"prompt":{"type":"string","description":"The high-level task for the swarm to solve"},"title":{"type":"string","description":"Short human-readable label shown during execution"},"max_agents":{"type":"integer","description":"Maximum parallel sub-agents (default 5, hard cap 100)"},"writable":{"type":"boolean","description":"Allow agents to edit files and run shell commands (default false = read-only analysis)"},"model":{"type":"string","description":"Model alias or full ID for all swarm agents (default: auto-resolved per role). Use \"opus\" for hardest tasks, \"haiku\" for fast/cheap parallel work."},"mode":{"type":"string","enum":["smart","rush","deep","free"],"description":"Agent mode applied to workers and synthesis agent (default: smart). Orchestrator uses rush unless overridden."},"per_agent_model":{"type":"object","description":"Per-role model overrides as a JSON object mapping role names to model aliases/IDs. Example: {\"finder\":\"haiku\",\"fixer\":\"opus\"}. Takes precedence over the top-level model param."},"telemetry_out":{"type":"string","description":"Optional file path to write telemetry JSON (cost, tokens, wall time, parallelism)"}},"required":["prompt"]}},
+    \\{"name":"run_agents","description":"Run multiple agents in parallel within a single tool call. Each element of the agents array is a run_agent spec (same fields as run_agent). All agents execute concurrently via Zig threads; results are collected and returned as a JSON array once every agent completes. Use this instead of multiple sequential run_agent calls when the tasks are independent.","inputSchema":{"type":"object","properties":{"agents":{"type":"array","description":"Array of agent specs to run in parallel","items":{"type":"object","properties":{"prompt":{"type":"string","description":"The task or question for this agent"},"model":{"type":"string","description":"Model alias or full ID (default: auto-resolved)"},"role":{"type":"string","description":"Agent role: finder, reviewer, fixer, explorer, architect, orchestrator, synthesizer, monitor"},"mode":{"type":"string","enum":["smart","rush","deep","free"]},"writable":{"type":"boolean"},"allowed_tools":{"type":"string"},"permission_mode":{"type":"string","enum":["default","acceptEdits","bypassPermissions"]},"cwd":{"type":"string"}},"required":["prompt"]}}},"required":["agents"]}},
+    \\{"name":"review_fix_loop","description":"Iterative review-fix-review loop. Runs a read-only reviewer to find issues, then a writable agent to fix them, then re-reviews. Repeats until the reviewer reports no remaining issues or max_iterations is reached. Returns a JSON object with iteration history and convergence status.","inputSchema":{"type":"object","properties":{"prompt":{"type":"string","description":"Override the default review criteria"},"max_iterations":{"type":"integer","description":"Maximum review-fix cycles (default 3, max 5)"}},"required":[]}},
+    \\{"name":"run_agent","description":"Run a single agent turn. Provider-agnostic: resolves the best backend (Claude/Codex) based on mode, role, and available providers. The primitive layer — use run_task for smart multi-step execution.","inputSchema":{"type":"object","properties":{"prompt":{"type":"string","description":"The task or question for the agent"},"model":{"type":"string","description":"Model alias or full ID (default: claude-sonnet-4-6). Use \"opus\" for hardest tasks, \"haiku\" for fast/cheap."},"role":{"type":"string","description":"Agent role: finder, reviewer, fixer, explorer, architect, orchestrator, synthesizer, monitor"},"mode":{"type":"string","enum":["smart","rush","deep","free"],"description":"Agent mode: smart (Sonnet), rush (Haiku), deep (Opus), free (Haiku)"},"allowed_tools":{"type":"string","description":"Comma-separated tool allowlist, e.g. \"Bash,Read,Edit\". Omit to allow all tools."},"permission_mode":{"type":"string","enum":["default","acceptEdits","bypassPermissions"],"description":"Permission mode for file and shell operations"},"writable":{"type":"boolean","description":"Allow file writes (maps to bypassPermissions when permission_mode is unset)"},"cwd":{"type":"string","description":"Working directory override (default: current repo path)"}},"required":["prompt"]}},
+    \\{"name":"run_task","description":"Smart executor: analyzes a task, picks the right strategy and agents, runs them with appropriate roles and models. Use this instead of run_agent for multi-step tasks. Supports chain presets (finder_fixer, reviewer_fixer, explore_report, architect_build) or auto-selection.","inputSchema":{"type":"object","properties":{"task":{"type":"string","description":"Task description — what needs to be done"},"preset":{"type":"string","enum":["finder_fixer","reviewer_fixer","explore_report","architect_build","custom"],"description":"Chain preset (default: auto-select based on task)"},"mode":{"type":"string","enum":["smart","rush","deep","free"],"description":"Agent mode for all agents in the chain"},"max_agents":{"type":"integer","description":"Max agents to spawn (default: preset-determined)"},"writable":{"type":"boolean","description":"Override write access (default: role-determined)"},"permission_mode":{"type":"string","enum":["default","acceptEdits","bypassPermissions"],"description":"Permission mode for file and shell operations"},"timeout_seconds":{"type":"integer","description":"Maximum total time for the full chain (default 300, max 600)"},"model":{"type":"string","description":"Model alias or full ID for all chain agents (default: auto-resolved per role). Use \"opus\" for hardest tasks, \"haiku\" for fast/cheap."}},"required":["task"]}}
     \\]}
 ;