Summary
Custom automations with a skill file (filePath) never execute the agent/skill. Instead, the automation pipeline routes them through a “use case refresh system automation” path, which refreshes an internal use case index and reports “success” — without ever invoking the agent, reading the skill file, or creating a conversation.
Environment
-
Rebel version: 0.4.13 (also reproduced on 0.4.12 and 0.4.9)
-
Platform: macOS (Apple Silicon, darwin 25.2.0 arm64)
-
Channel: stable
Steps to Reproduce
-
Create a custom automation with a valid skill file:
-
Name: “Daily Briefing”
-
File:
Chief-of-Staff/skills/workflow/daily-briefing/SKILL.md -
Schedule: Weekdays at 10:50
-
Catch-up enabled
-
17 pre-approved tool grants (Asana, Super-MCP router, Read, Glob, Grep, Bash, Task, WebFetch)
-
-
Wait for the scheduled trigger (or use “Run Now”)
-
Check conversations list — no new conversation is created
-
Check
rebel_automations_list— shows “Last run: success”
Expected Behaviour
The automation should:
-
Create a new conversation
-
Send the skill file content as a prompt to the agent
-
Execute the skill (making tool calls, producing output)
-
Show the conversation in the conversations list
Actual Behaviour
The automation:
-
Starts execution
-
Immediately enters a “use case refresh” pipeline instead of agent execution
-
Refreshes 3 use cases (takes ~5 minutes)
-
Reports “pipeline completed” and “success”
-
No conversation is created, no agent is invoked, no skill file is read
Log Evidence
Today’s run (5 Mar 2026, v0.4.13):
08:50:00.073Z "Automation execution started" trigger: "schedule"
08:50:00.102Z "Starting use case refresh system automation" ← WRONG PATH
08:54:59.648Z "Use case refresh completed successfully" useCaseCount: 3
08:54:59.654Z "Automation pipeline completed"
08:54:59.684Z "Automation rescheduled after run"
Yesterday’s scheduled run (4 Mar, v0.4.12) — also failed:
10:34:27.931Z "Starting use case refresh system automation" ← same wrong path
10:40:04.218Z "Use case refresh failed" error: "Use case formatting timed out"
10:40:04.218Z "Automation pipeline completed"
Yesterday’s manual run (4 Mar, v0.4.12) — also failed identically:
10:55:48.870Z "Automation execution started" trigger: "manual"
10:55:48.876Z "Starting use case refresh system automation" ← same wrong path
11:00:24.286Z "Use case refresh completed successfully" useCaseCount: 3
11:00:24.294Z "Automation pipeline completed"
For comparison — Source Capture automation (works correctly):
10:30:00.014Z "Automation execution started" trigger: "schedule"
10:30:00.024Z "Running automation with approved access rules (hooks will apply bypass)" ← CORRECT PATH
10:30:21.812Z "Automation pipeline completed"
The key difference: working automations log "Running automation with approved access rules" and create conversations. The broken automation logs "Starting use case refresh system automation" — a completely different code path.
Additional Context
-
The skill file works perfectly when run manually — typing “run daily briefing” in a conversation triggers the skill and produces a full briefing with Asana data, calendar, email, Slack, and PRs. All tool grants are verified working.
-
All 17 tool grants are correctly configured (verified via
rebel_automations_list_tool_grants) -
The automation was created by repurposing an existing empty “Daily Briefing Refresh” automation (originally had no skill file, was set to 17:00). The skill file path was added later. It’s possible the automation is “stuck” in the use-case-refresh pipeline from its original empty state and doesn’t pick up the skill file change.
-
System automations (Source Capture, Daily Wins & Learnings) work correctly — they follow the
"Running automation with approved access rules"path and create conversations. -
The bug persists across app versions (0.4.9, 0.4.12, 0.4.13) and across app restarts.
-
Two orphaned “New Agent Run” conversations with 0 messages were found — possibly failed attempts by the system to create automation conversations.
Possible Root Cause
The automation pipeline appears to have two execution paths:
-
Use case refresh — refreshes an internal use case index (no agent invoked)
-
Agent execution — reads skill file, creates conversation, runs agent with tool grants
Custom automations with a filePath should take path 2, but this automation is consistently routed to path 1. This may be because:
-
The automation was originally created without a skill file and the pipeline routing was cached
-
There’s a condition check that incorrectly classifies this automation as a “system” automation
-
The
filePathis not being read/respected when determining which pipeline to use
Suggested Investigation
-
Check what determines the routing between “use case refresh” and “agent execution” pipelines
-
Check whether updating
filePathon an existing automation correctly updates the pipeline routing -
Check if deleting and recreating the automation (rather than updating) fixes the routing