Skip to content

Sentinel — Risks

Wedge guardrail: Sentinel turns pull requests into merge-ready verdicts. First wedge: CLI review on exact HEAD → SHIP / DO NOT SHIP verdict with cited evidence → GitHub App.

  1. False confidence (existential). A SHIP on a diff with a missed vulnerability destroys the brand permanently. Mitigation: severity-gated language (never “safe”, only “no findings under rule pack vX”), conservative default to DO_NOT_SHIP on uncertainty, public miss log.
  2. GitHub ships verdicts into Copilot review (6-12 mo window). Compresses independence moat to regulated/self-host buyers. Mitigation: CLI-first portability; revisit competitor-analysis quarterly.
  3. Model cost per review. Deep reviews on large PRs burn tokens; per-PR cost must stay under ~$0.50 at team-tier pricing or unit economics invert. Mitigation: diff-size budgets, incremental re-review (only new hunks), rule pre-filter before model calls.
  4. Noise intolerance. One noisy week (duplicates, nits) and engineers mute the App forever. Mitigation: resolution memory from day 1, style findings off by default, silence-as-feature.
  5. Single-platform dependence. GitHub App distribution = GitHub policy risk. Mitigation: CLI works on any git remote; GitLab/Bitbucket adapters post-MVP, not before.
  6. Founder bottleneck. Reviews, rule packs, and verdict quality all route through Jonas initially. Mitigation: rule packs are data (delegable), verdict quality is measured (miss log), hiring trigger: 10 design partners.

Source: Aftergraph/sentinel@beececfb4885d8055cc6694e8979b47759a212a4 docs/risks.md.md — verbatim, pinned, no reinterpretation.

Was this page helpful?