feat(engine): add mission_get action for retrieving mission results

[serrrfirat]

Summary

• Add mission_get action so the LLM can retrieve mission results, status, approach history, and recent thread outputs
• Map routine_history → mission_get so the v1 alias also works in engine v2
• Root cause: when users asked "what is the result of the research?", the agent fell back to list_jobs because no mission results tool existed — missions spawn threads (not jobs), so list_jobs always returned empty

Changes

File	Change
src/bridge/router.rs	Register mission_get ActionDef in capability registry
src/bridge/effect_adapter.rs	Handler: loads mission + last 5 threads + extracts assistant responses; alias: routine_history → mission_get; regression test
crates/ironclaw_engine/src/runtime/mission.rs	Add pub fn store() accessor on MissionManager

Test plan

• cargo fmt --check — clean
• cargo clippy --all --all-features — zero warnings
• cargo test -- routine_history_maps — alias test passes
• Manual: create a mission, fire it, then ask "what is the result?" — should call mission_get instead of list_jobs

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces the mission_get action, enabling the retrieval of detailed status, approach history, and recent thread results for missions and routines. The implementation includes a new store() accessor in MissionManager, the mission_get handler in EffectBridgeAdapter, and a mapping for the legacy routine_history action. Review feedback highlights a security vulnerability where the mission_get handler lacks an ownership check, potentially allowing unauthorized access to mission data. Furthermore, it is suggested to truncate the approach_history in the response to prevent exceeding LLM context window limits.

[gemini-code-assist]

The mission_get handler lacks an ownership check, allowing any user to retrieve details of any mission by its UUID. This is a security regression compared to other mission actions like mission_fire or mission_pause. An explicit check against context.user_id should be added.

References

1. Tools that interact with user-owned resources must verify that the authenticated user ID in the tool context matches the resource owner's ID before performing any read or write operations to prevent unauthorized cross-user access.

[gemini-code-assist]

approach_history can grow significantly over time as it stores the full response of every mission run. Returning the entire history in mission_get could eventually exceed the LLM's context window or the safety layer's output limits. Consider returning only the most recent entries (e.g., the last 10).

Suggested change

	"approach_history": mission.approach_history,
	"approach_history": mission.approach_history.iter().rev().take(10).rev().cloned().collect::<Vec<_>>(),

References

1. Always truncate tool output for previews or status updates to a reasonable maximum length to prevent excessive memory/bandwidth usage and reduce the risk of leaking sensitive information.

[serrrfirat]

Addressed review feedback:

1. Security (high) — Added ownership check: mission.user_id != context.user_id now returns an error, preventing cross-user IDOR. Matches the guards on mission_fire/mission_pause/mission_resume.

2. Medium — Capped approach_history to last 10 entries via .iter().rev().take(10).rev() to prevent context window overflow.

Both fixes in commit 021d5c6.

[henrypark133]

Review: ownership fix and response shaping look sound

I re-checked the current head in src/bridge/effect_adapter.rs:344-382. The earlier ownership concern is addressed: mission_get now rejects missions whose mission.user_id does not match the caller, and the response also caps approach_history to the most recent 10 entries instead of returning the full unbounded list.

Prior feedback status: the previously raised ownership issue appears addressed in the current head.

Residual risk: I did not wait for the full local compile to finish for a targeted alias test in this pass, so approval here is based on code inspection rather than completed local test execution.