Skip to content

Sentinel — Decisions

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.

# Decision Rationale Needs owner approval to reverse
1 Name: Sentinel by Aftergraph Grows out of the existing verifier role; guard-not-chatter semantics Yes (brand)
2 Wedge order: CLI/skill → GitHub App → hosted team tier Distribution where incumbents cannot follow; dogfood from sprint 1 Yes (strategy)
3 Verdict bound to exact HEAD; base move = auto-stale The entire moat; without it we are another comment bot Yes (product contract — never silently)
4 No auto-fix, no auto-approve in MVP Unverified actions destroy trust faster than missing features Yes
5 Free for public repos, paid team tier later OSS is distribution, not revenue Yes (pricing)
6 Interviews/pilot optional later, not next Prototype-first; validation via desk research + 5-PR prototype runs Jonas only
7 Docs before code; no scaffolding in this repo Strategy must survive contact with prototype results before LOC Jonas only
8 Vendor-doc claims marked [VERIFY], unpublished until confirmed One caught fabrication discounts the whole comparison No (process rule, permanent)
9 v0 verdict path fully deterministic, no LLM as judge “Verified” and probabilistic output cannot share a sentence; LLM allowed in v1 as explanation layer only Yes (product contract)

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

Was this page helpful?