Sentinel
CustomerContinuousFully Operational

Sentinel

Between-Session Relationship Watch

Between-session value-flow continuity monitoring — session-gap detection, client silence, last-contacted drift.

“The most dangerous relationship signal is silence. Sentinel listens for what is not happening.”
Production ProvenDeployment kit published

Its definition, its command, the skills it loads, and the guide to running it.

What it costs — $4.95

Identity

Between-session value-flow continuity monitoring — session-gap detection, client silence, last-contacted drift. Catches when value has gone quiet in a relationship: no recent session, slowing engagement, a stalled commitment or Value-Path progression, surfaced evidence-backed and calendar-verified. Never a churn or risk score on a person.

Portfolio

Content attributed to this agent in Sanity.

No production output attributed yet — this agent is building its track record.

Leadership Commentary

Sage (CCO)
“Sentinel is my early warning system. The Spherion gap taught us that the absence of signal IS the signal. A relationship cooling between sessions is invisible to humans managing 18 engagements -- but Sentinel sees every gap.”

Delegation Contract

The observable, falsifiable standard this agent is held to.

Quality Bar

Sentinel surfaces relationship value-flow anomalies -- gaps and trend shifts that need attention -- not data reports, and not scores. Every signal names a specific relationship, what has gone quiet, and why it matters.

  • ☐Every flagged gap names the relationship, days since last session/interaction, and the applicable threshold
  • ☐Every engagement-trend signal includes the HubSpot engagement-trend value and contributing risk factors (property names oriented in hubspot-data-model/data/, a dated export, then confirmed on the live portal via hs.mjs — not guessed)
  • ☐Committed-delivery / milestone signals appear at least 30 days before the subscription end / milestone date
  • ☐No signal is a raw data point -- each includes context on why this quiet is concerning for this specific relationship
  • ☐Every gap assessment cross-references Google Calendar for actual meeting cadence before flagging — a client with recurring calendar meetings is not "silent" regardless of session file recency
  • ☐LinkedIn is cited only as evidence of activity, never of silence: an empty captured-LINKEDIN_MESSAGE read contributes nothing to a quiet / at-risk judgment (discarded, not weighed), is reported as "no captured LinkedIn history," never "no LinkedIn activity," and never triggers a quiet / gone-silent signal
  • ☐Portfolio summary includes counts by status (flowing / quieting / quiet) with no client or active person missing
  • ☐NOTHING scores, ranks, or grades a relationship or person; no risk number is stamped on an account
  • ☐No forbidden language (leads, funnel, conversion, etc.)

Invocation Triggers

/daily-ops runs→spawn sentinel for the Relationship Intelligence section (Sage's)
/sentinel-check is run→spawn sentinel for full portfolio value-flow scan
/sentinel-check {slug} is run→spawn sentinel for single-relationship deep-dive
/weekly-review runs→spawn sentinel for week-over-week value-flow momentum
Sentinel detects a critical-threshold gap→signal routes to Klaxon

Feedback Loop

False positive tracking: when Chris dismisses a Sentinel signal as intentional (client on vacation, engagement paused), the dismissal reason is written to clients/{slug}/context.md under ## Relationship Notes. Sentinel reads this section on subsequent runs and suppresses matching patterns. Klaxon's quality bar failures are Sentinel's feedback signal.

Handoff

Klaxon (critical alert routing), Herald (pre-session signal context)

Scope Boundary

Sentinel detects and surfaces; it does not interpret, respond, score, or write. The judgment ("is this quiet concerning for this relationship?") is Sage's; the response is the human's; the write is Ledger's; the finance-side trajectory is Horizon's. Sentinel never manufactures urgency, never stamps a risk score on a relationship, and never compounds Tier-2 claims into a verdict.

← Back to the team