| name | rd-tool-catalog |
| description | Choose direct standalone V3 CLI tools when a high-level skill boundary is insufficient. |
| argument-hint | <tool name> | --list | --category <orchestration|inspection|primitives> |
| allowed-tools | ["Read","Bash"] |
<codex_skill_adapter>
Invocation
- Claude Code:
rd-tool-catalog skill
- Codex:
$rd-tool-catalog
- Treat text after skill mention as arguments.
</codex_skill_adapter>
Decision-oriented skill for the standalone V3 CLI tool surface. Narrows a vague
request to a concrete tool choice or routes back to the correct high-level skill.
Inspect round truth artifacts such as the capability snapshot, dispatch handle, branch receipt, round summary, provenance, and recovery actions.
Maps to rd_agent.entry.tool_catalog and the rdagent-tool list / describe interface.
<execution_context>
@skills/rd-tool-catalog/references/tool-selection.md
@skills/_shared/references/tool-execution-context.md
</execution_context>
<routing_model>
- Stay in high-level skills unless they are insufficient
- If dropping lower, inspect catalog and narrow by category first
- Use recommended_entrypoint to decide rd-agent vs direct tool
- Do not force a direct tool choice when the caller really wants rd-agent or
a stage skill
</routing_model>
1. If caller has not established that a high-level skill is insufficient, route
back to rd-agent or the appropriate stage skill
2. Narrow by top-level category: orchestration, inspection, primitives
3. If primitives, narrow by subcategory: branch_lifecycle, branch_knowledge,
branch_selection, memory
4. Select the concrete tool
CLI commands:
uv run rdagent-tool list
uv run rdagent-tool describe <tool>
<success_contract>
Success means the skill narrows the request to a concrete tool choice or
explicitly routes the caller back to the correct high-level skill. The final
output identifies the selected category, any relevant subcategory, and the
recommended entrypoint.
</success_contract>