Failure PatternDecision layer
The Vigil Timeout Trap: Why Agencies Fail With Vigil on Long Agent Runs
Symptom: A client's overnight Claude Code refactor run dies at 2am and the Mac is asleep when the developer checks it at 9am, even though Vigil is installed and the menu bar icon is present. Root cause: Vigil releases the sleep hold when an agent finishes, fails, or waits too long, so the configurable grace period must be tuned to the client's actual input-wait window. Agencies that leave the default in place on agents that pause for human approval mid-run get a released hold and a dead session.
By InnovaAI ResearchPublished
How do you recognize it?
- •A client's overnight Claude Code refactor run dies at 2am and the Mac is asleep when the developer checks it at 9am, even though Vigil is installed and the menu bar icon is present.
- •The menu bar dashboard shows a session stuck in 'waiting' for hours while the agent has actually exited, so the sleep hold never releases and the Mac stays awake all weekend.
- •Two developers on the same client team both activate the same license key and one Mac silently loses its hold mid-session.
- •A Windows-based contractor on the client's team cannot reproduce the setup, so the agency's documented Vigil workflow only covers part of the delivery team.
Why does it happen?
- •Vigil releases the sleep hold when an agent finishes, fails, or waits too long, so the configurable grace period must be tuned to the client's actual input-wait window. Agencies that leave the default in place on agents that pause for human approval mid-run get a released hold and a dead session.
- •Lifecycle hooks are per-agent. Installing Vigil for Claude Code does nothing for a developer who switches to Cursor or Gemini CLI that afternoon, and the menu bar gives no warning that the active agent has no hook wired up.
- •One license key activates two Macs. Agencies that standardize on a single key across a client team of five hit activation limits and lose track of which machines are actually covered.
- •Vigil runs on macOS 14 and later only. Any client delivery team with Windows or Linux developers cannot be covered by the same managed deployment, which breaks a uniform agency rollout.
How do you fix it?
- •Open the Vigil menu bar dashboard, select the affected session, and raise the grace period so the hold survives the client's longest expected human-approval pause before releasing.
- •Run through the client's agent list (Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, pi, OpenCode) and install the lifecycle hook for every agent in active use, not just the one the lead developer prefers.
- •Audit license activation per client team and buy additional $3.99 keys so no more than two Macs share a key, then record which machines are covered in the client handoff doc.
- •Confirm each client developer's macOS version is 14 or later before promising coverage, and document the Windows and Linux developers as out of scope in the retainer agreement.
More on Vigil
- StrategyWhy Vigil Turns Unattended Agent Runs Into Billable Agency Delivery
- ConceptVigil Sleep-Hold Window
- Evaluation RuleWhen to Adopt Vigil: Only If Your Client's Agent Runs Are Unattended and Mac-Based
- Decision FrameworkVigil: Buy vs Skip (Mac Agent Teams Running Unattended Sessions)
- Implementation BlueprintVigil Client Agent Uptime Deployment (5-7 days)
- Operating ProcedureVigil Agent Hook Deployment and Sleep Hold Configuration (Onboarding)