[Bug] CLI session output references non-existent ./genie command

[namastex888]

Bug Description

Session 4d4c76a7-e58a-487a-b66f-7ff408dafb37 (prompt neuron) appeared active in mcp__genie__list_sessions but mcp__genie__resume returned "No run found" error. Session state becomes inaccessible after extended runtime (~45 minutes).

Steps to Reproduce

1. Start long-running agent session: mcp__genie__run with agent="orchestrator"
2. Note session ID (e.g., 4d4c76a7-e58a-487a-b66f-7ff408dafb37)
3. Wait ~45 minutes (session runs in background)
4. Resume same session: mcp__genie__resume with sessionId="4d4c76a7-e58a-487a-b66f-7ff408dafb37"
5. Observe error: "No run found for session 4d4c76a7-e58a-487a-b66f-7ff408dafb37"

Expected Behavior

• mcp__genie__view should show session state
• mcp__genie__resume should continue the session with full context preserved
• Session should remain queryable and resumable indefinitely
• Session lifecycle: start → (optional pause) → resume → complete

Actual Behavior

• Session appears in mcp__genie__list_sessions with status: running
• mcp__genie__view returns "No run found"
• mcp__genie__resume returns "No run found"
• Session context lost entirely after timeout
• State sync broken between list/view/resume operations

Environment

• OS: Linux (WSL2)
• Node: v22.0.0
• Package version: 2.4.0-rc.8
• MCP: Evolution API

Error Logs / Screenshots

Session 4d4c76a7-e58a-487a-b66f-7ff408dafb37
Error: "No run found for session 4d4c76a7-e58a-487a-b66f-7ff408dafb37"
Occurred: 2025-10-17 17:20 UTC
Duration before failure: ~45 minutes

Possible Solution

• Session storage backend should not drop data after timeout
• State sync mismatch between list/view/resume operations needs investigation
• Sessions should not be auto-expired by garbage collection
• Verify MCP backends are in sync (list hitting same data source as view/resume)

Affected Areas

• MCP Server
• Agents

[namastex888]

Problem Summary

Post v2.4.0, MCP session output still suggests ./genie resume/view/list commands that don't exist. The CLI was refactored to use MCP-only, but session response templates weren't updated.

Reproduction Steps

1. Start any Genie agent session via mcp__genie__run
2. Check session output for resume/view/list instructions
3. Try executing ./genie resume <session-id>
4. Result: 'command not found' error

Expected Behavior

Session output should reference correct MCP tools:

• mcp__genie__resume with sessionId="<session-id>"
• mcp__genie__view with sessionId="<session-id>"
• mcp__genie__list_sessions
• Or npx automagik-genie for CLI operations

Actual Behavior

Output shows legacy commands like:

• ./genie resume <session-id>
• ./genie view <session-id>
• ./genie list
• ./genie stop <session-id>

Root Cause

Architecture shift (v2.4.0) moved to MCP-only for agent operations, but CLI/MCP boundary wasn't fully updated in session response templates.

Impact

• ❌ Blocks users from resuming sessions
• ❌ Causes 'command not found' errors
• ❌ Undermines confidence in CLI

Related Issues

• [Bug] MCP session disappears after resume - "No run found" #66 ([Bug] MCP session disappears after resume - 'No run found')

Validation

When fixed, verify:

1. Start new agent session
2. Inspect output for correct MCP tool references
3. Verify no ./genie commands appear in output
4. Test resume with provided MCP command works

[namastex888]

Duplicate of #89 (consolidated issue). Root cause: git neuron self-delegation loop during issue creation.