fix(agent): drop removed memory_tree modes from orchestrator prompt

[sanil-23]

Summary

• Remove the query_topic and query_global modes from the orchestrator's memory_tree documentation — both were deleted when the global/topic trees were removed, but two LLM-facing strings still advertised them.
• Add the modes the prompt was missing: walk, smart_walk, and ingest_document.
• Fix the memory_tree_search_entities tool description, which pointed the model at a nonexistent memory_tree_query_topic tool.

Problem

The memory_tree dispatcher (src/openhuman/memory/query/mod.rs) supports exactly: search_entities, query_source, drill_down, fetch_leaves, ingest_document, walk, smart_walk. The query_topic / query_global modes were removed with the global/topic trees (asserted gone in mod.rs tests), but two prompts the model actually reads were never updated:

• orchestrator/prompt.md still listed mode: "query_topic" and mode: "query_global" as valid, and the "Start cheap (query_* summaries)" line leaned on them. If the orchestrator followed its prompt it would emit a call the dispatcher rejects with unknown mode. The prompt also never mentioned the more capable walk / smart_walk retrieval paths, so the agent was blind to them.
• memory_tree_search_entities's tool description() told the model to call a memory_tree_query_topic tool that does not exist.

Solution

Align both LLM-facing strings with the dispatcher's real mode list — drop the two dead modes, document smart_walk (multi-strategy retrieval) and walk (agentic traversal) as the open-ended-query defaults, add ingest_document, and update search_entities to reference memory_tree's real retrieval modes.

Submission Checklist

• N/A: prompt/description string-only change — no executable behaviour to unit-test. The dispatcher already asserts the removed modes are gone (memory/query/mod.rs).
• N/A: behaviour-only change — no changed executable lines for diff-cover.
• Coverage matrix updated — N/A: behaviour-only change
• Affected feature IDs — N/A
• No new external network dependencies introduced
• Manual smoke checklist — N/A: does not touch release-cut surfaces
• Linked issue — N/A: no tracking issue

Impact

• Desktop/CLI agent runtime: the orchestrator and search_entities tool now describe only modes the tool actually accepts, eliminating a class of guaranteed-failing memory_tree calls and surfacing smart_walk / walk. No code-path or schema change.

Related

• Closes: N/A
• Follow-up PR(s)/TODOs: optional tidy-up of historical query_topic mentions in memory_tree/retrieval/{drill_down,types}.rs comments (non-LLM-facing).

---

AI Authored PR Metadata (required for Codex/Linear PRs)

Linear Issue

• Key: N/A
• URL: N/A

Commit & Branch

• Branch: fix/memory-tree-prompt-stale-modes

Summary by CodeRabbit

• Improvements
  • Expanded AI agent memory retrieval strategies with new search modes to improve context understanding.
  • Clarified tool usage instructions for better performance when searching by user name.

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR updates memory tree retrieval guidance across two files: the orchestrator prompt now describes expanded retrieval modes (smart_walk, walk, ingest_document) while removing obsolete modes, and the search entities tool description is aligned to reference the canonical id-based lookup pattern.

Changes

Memory Tree Retrieval Mode Updates

Layer / File(s)	Summary
Orchestrator prompt memory_tree modes and guidance src/openhuman/agent_registry/agents/orchestrator/prompt.md	Orchestrator prompt removes query_topic and query_global modes, retains query_source, and introduces smart_walk and walk as new retrieval strategies. ingest_document mode added to the option list. "Start cheap" guidance is repositioned to recommend query_* summaries before expensive operations.
Search entities tool description update src/openhuman/memory/query/search_entities.rs	MemoryTreeSearchEntitiesTool description updated to direct users to call this tool first when a user mentions someone by name, then perform memory_tree retrieval keyed on the canonical id.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Possibly related PRs

• tinyhumansai/openhuman#3077: Implements and wires up the smart_walk memory-tree retrieval mode that is now documented in the orchestrator prompt and tool descriptions.

Suggested labels

rust-core, memory

Poem

🐰 New modes for memory trees take flight,
Smart walks and ingest_document bright,
Orchestrator's wisdom set right,
Tools aligned in prompt delight!

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly describes the main change: removing stale memory_tree modes from the orchestrator prompt, which is the primary focus of the PR.
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_registry/agents/orchestrator/prompt.md (1)

187-187: 💤 Low value

Consider consistent spelling variant.

The document uses British English spelling ("synthesise" on line 11), but line 187 uses American spelling "synthesized". For consistency, consider "synthesised".

🤖 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_registry/agents/orchestrator/prompt.md` at line 187, The
phrase for the `mode: "walk"` description uses American spelling "synthesized"
which is inconsistent with the British variant "synthesise" used elsewhere;
update the text in the `mode: "walk"` line to use "synthesised" so the document
consistently uses British English (search for the string `mode: "walk"` and
replace "synthesized" with "synthesised").

🤖 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_registry/agents/orchestrator/prompt.md`:
- Line 187: The phrase for the `mode: "walk"` description uses American spelling
"synthesized" which is inconsistent with the British variant "synthesise" used
elsewhere; update the text in the `mode: "walk"` line to use "synthesised" so
the document consistently uses British English (search for the string `mode:
"walk"` and replace "synthesized" with "synthesised").

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 42aa277f-8a60-40c2-80a8-6b37350545c0

📥 Commits

Reviewing files that changed from the base of the PR and between 38b0883 and 3bc55b7.

📒 Files selected for processing (2)

• src/openhuman/agent_registry/agents/orchestrator/prompt.md
• src/openhuman/memory/query/search_entities.rs

[oxoxDev]

Strings-only fix aligning the orchestrator memory_tree prompt + search_entities tool desc to the real dispatcher. Verified the prompt now lists exactly the 7 live modes (search_entities/query_source/smart_walk/walk/drill_down/fetch_leaves/ingest_document) matching memory/query/mod.rs:143-149; the stale query_topic/query_global are dropped (the memory_tree_schema_mode_enum_has_all_modes test pins this), and search_entities no longer points at the nonexistent memory_tree_query_topic. No logic change.

Necessity: problem real (stale strings → unknown-mode routing errors) · right size (+6/-4) · right approach (align to dispatcher truth, CI-guarded) · Ship.

Approve with one nit (inline).