feat(agents): assign #1208 coding-harness tools to relevant built-in agents

[senamakel]

Summary

• feat(tools): coding-harness baseline primitives (#1205) #1208 added 9 coding-harness primitives (grep, glob, list, edit, apply_patch, todowrite, plan_exit, web_fetch, lsp) to the global all_tools registry, but did not wire them into any agent's [tools] named allowlist.
• Result: only wildcard = {} agents (e.g. tools_agent) could actually see them. The agents that would benefit most — code_executor, planner, researcher, orchestrator — were silently still on the legacy surface.
• This PR distributes the new tools per-agent based on each agent's role and sandbox_mode.

Distribution

Agent	Sandbox	Added
code_executor	sandboxed	full set: grep, glob, list, edit, apply_patch, todowrite, plan_exit, web_fetch, lsp
planner	read_only	read-only nav (grep, glob, list) + todowrite + plan_exit + web_fetch
researcher	none	read-only nav (grep, glob, list) + web_fetch (the simple URL-GET sibling of http_request, same allowed_domains gate)
orchestrator	n/a	todowrite + plan_exit only — the orchestrator coordinates, it doesn't edit files

planner deliberately does NOT receive edit / apply_patch / lsp — sandbox_mode = "read_only" forbids workspace mutations and downstream agents do the writing.

Other agents (archivist, critic, summarizer, help, welcome, integrations_agent, tool_maker, trigger_triage, trigger_reactor, morning_briefing) are intentionally untouched — their narrow named allowlists reflect domain-specific surfaces (memory ops, composio meta-tools, gitbooks, onboarding) that the new coding primitives don't fit.

lsp is capability-gated by OPENHUMAN_LSP_ENABLED; listing it in code_executor's allowlist is harmless when the gate is off (the tool is simply not registered into the runtime).

Test plan

• cargo test --lib agent::agents::loader — 20/20 pass (covers code_executor_*, planner_is_read_only_with_composio_meta_tools, researcher_has_curl_for_artifact_downloads, orchestrator_has_reasoning_hint_and_named_tools, all_builtins_parse)
• cargo check --manifest-path Cargo.toml clean
• Manual: spin up a coding session, confirm code_executor actually sees grep / edit / apply_patch in its tool catalog

Impact

• TOML-only change to 4 agent definitions. No code, no schema, no migration.
• Backwards compatible: pure additions to existing named = [...] allowlists.
• Branch was pushed with --no-verify because the repo's pre-push hook fails on pre-existing app/ ESLint warnings (mascot / commands code) and a Tailwind class lint check — none of which this PR touches. Per CLAUDE.md (If a pre-push hook fails on something unrelated to your changes ... push with --no-verify and call it out in the PR body.)

Related

• Follows up feat(tools): coding-harness baseline primitives (#1205) #1208

Summary by CodeRabbit

New Features

• Enhanced Agent Capabilities
  • Expanded toolsets across multiple system agents with new navigation, search, and analysis tools
  • Code executor gains enhanced editing and task coordination support
  • Planner and researcher agents equipped with improved search and research primitives
  • System-wide improvements to agent coordination and workflow management

[coderabbitai]

📝 Walkthrough

Walkthrough

Four agent TOML configuration files are updated to expand their permitted tool allowlists. The code executor gains navigation and editing tools plus task flow helpers; the orchestrator adds todo and plan exit handlers; the planner gains read-only navigation and plan output tools; the researcher adds web fetch and read-only search capabilities.

Changes

Agent Tool Allowlist Expansion

Layer / File(s)

Summary

Tool Registry Configuration src/openhuman/agent/agents/code_executor/agent.toml, src/openhuman/agent/agents/orchestrator/agent.toml, src/openhuman/agent/agents/planner/agent.toml, src/openhuman/agent/agents/researcher/agent.toml

Each agent's [tools].named list is expanded with tools aligned to its role: code executor receives full capabilities (grep, glob, list, edit, apply_patch, todowrite, plan_exit, web_fetch, lsp); orchestrator receives plan handoff tools (todowrite, plan_exit); planner receives read-only navigation and plan output (grep, glob, list, todowrite, plan_exit, web_fetch); researcher receives web and search capabilities (web_fetch, grep, glob, list).

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• tinyhumansai/openhuman#500: Directly modifies the same four agent TOML configuration files to expand tool allowlists across code executor, orchestrator, planner, and researcher agents.

Poem

🐰 A rabbit hops with toolbox in hand,
Expanding capabilities across the land!
Executors execute, planners now plan,
Researchers research with web-fetch so grand. ✨
Each agent equipped for their destined role!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly and clearly describes the main change: assigning coding-harness tools to agent configurations, which aligns with the PR's core objective of wiring nine primitives into per-agent tool allowlists.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

src/openhuman/agent/agents/code_executor/agent.toml (1)

26-43: ⚡ Quick win

Add a regression assertion for this exact tool surface.

Given Line 26 through Line 43 materially expands executor capabilities, it’s worth adding/expanding a loader test that asserts expected tools for code_executor (and complementary exclusions for planner) so future edits don’t silently drift.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/openhuman/agent/agents/code_executor/agent.toml` around lines 26 - 43,
Add a regression unit test that asserts the exact tool surface loaded for the
code_executor agent (the "named" list that currently contains
"shell","file_read","file_write","git_operations","node_exec","npm_exec","curl","grep","glob","list","edit","apply_patch","todowrite","plan_exit","web_fetch","lsp")
and also asserts complementary exclusions for the planner agent so changes are
caught. Implement this in the existing loader/agent tests (e.g., add a test
named test_code_executor_tool_surface_regression or augment test_load_agents) by
loading the agent configuration, reading the tool names for the agent identified
as code_executor, and comparing against the exact expected list (order-sensitive
or order-insensitive per your test conventions); likewise assert planner’s tool
list does not include these executor-only tools. Ensure the test fails on any
divergence so future edits must update the asserted list.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@src/openhuman/agent/agents/code_executor/agent.toml`:
- Around line 26-43: Add a regression unit test that asserts the exact tool
surface loaded for the code_executor agent (the "named" list that currently
contains
"shell","file_read","file_write","git_operations","node_exec","npm_exec","curl","grep","glob","list","edit","apply_patch","todowrite","plan_exit","web_fetch","lsp")
and also asserts complementary exclusions for the planner agent so changes are
caught. Implement this in the existing loader/agent tests (e.g., add a test
named test_code_executor_tool_surface_regression or augment test_load_agents) by
loading the agent configuration, reading the tool names for the agent identified
as code_executor, and comparing against the exact expected list (order-sensitive
or order-insensitive per your test conventions); likewise assert planner’s tool
list does not include these executor-only tools. Ensure the test fails on any
divergence so future edits must update the asserted list.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 4fe2044c-a275-4fe5-8081-ffaee1ae6b0d

📥 Commits

Reviewing files that changed from the base of the PR and between e945390 and 5da460f.

📒 Files selected for processing (4)

• src/openhuman/agent/agents/code_executor/agent.toml
• src/openhuman/agent/agents/orchestrator/agent.toml
• src/openhuman/agent/agents/planner/agent.toml
• src/openhuman/agent/agents/researcher/agent.toml