[codex] Fix v2 tool_info action inventory lookup

[henrypark133]

Summary

Fixes v2 tool_info discovery so it resolves against the current ActionInventory surface instead of falling back to the global ToolRegistry. This lets engine-native actions like mission_create return schema details even though they are not registered builtin tools, and prevents tool_info from explaining globally registered tools that are not callable/discoverable in the current v2 context.

Also removes the hardcoded mission_* Python signatures from the CodeAct preamble so compact tools are discovered through the runtime action surface and tool_info(..., detail="schema").

Root cause

The model could follow the prompt and call tool_info(name="mission_create", detail="schema"), but the executed path could still hit registry-backed lookup. Since mission_create is engine-native and lives in ActionInventory, not ToolRegistry, that produced No tool named 'mission_create' is registered.

Validation

• cargo fmt
• cargo test tool_info --lib
• cargo test -p ironclaw_engine tool_info --lib
• cargo test -p ironclaw_engine prompt_renders_compact_enabled_tools_once_with_schema_instruction --lib
• git diff --check

Notes

This PR intentionally keeps mission_create outside ToolRegistry; the v2 source of truth is the current action inventory.

[gemini-code-assist]

Code Review

This pull request refactors the agent's tool discovery mechanism by removing hardcoded tool signatures from the system prompt and introducing a dynamic schema retrieval system via tool_info(name="<tool>", detail="schema"). The changes include updates to the codeact_preamble.md, enhancements to the tool_info implementation in both scripting and structured executors to support schema lookups, and improved error handling in the bridge to handle missing action snapshots. I have no feedback to provide as the review comments did not point out actionable issues in the current code changes.

[Copilot]

Pull request overview

Fixes v2 tool_info discovery to resolve against the current per-context action surface (ActionInventory / action snapshots) instead of falling back to the global ToolRegistry, and updates CodeAct prompt content to rely on runtime schema discovery rather than hardcoded mission signatures.

Changes:

• Update EffectBridgeAdapter’s tool_info execution path to prefer available_action_inventory_snapshot, then available_actions_snapshot, and error when neither is present (no ToolRegistry fallback).
• Add/adjust tests across bridge + executors to cover schema lookup for engine-native / non-registry actions and prevent registry leakage.
• Remove hardcoded mission tool signatures from the CodeAct preamble and tighten prompt assertions accordingly.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/bridge/effect_adapter.rs	Switch tool_info snapshot lookup to ActionInventory-first, add regression tests to prevent ToolRegistry fallback.
src/bridge/action_discovery.rs	Add test helpers and new unit tests validating detail="schema" / include_schema behavior and schema overrides.
crates/ironclaw_engine/src/executor/structured.rs	Update test stubs/inventory to support tool_info schema resolution; add structured execution regression test.
crates/ironclaw_engine/src/executor/scripting.rs	Update scripting test stub + test to validate schema resolution via propagated ActionInventory snapshot.
crates/ironclaw_engine/src/executor/prompt.rs	Assert prompt no longer embeds hardcoded mission Python signatures.
crates/ironclaw_engine/prompts/codeact_preamble.md	Remove hardcoded mission_* signatures and instruct dynamic schema discovery via tool_info(..., detail="schema").

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[serrrfirat]

Reviewed changed files in full and ran targeted tool_info test coverage locally. No blocking findings.