Auditor
reviewReviews a change for what breaks in production — blast radius first, silent failures over style, every finding with a concrete failing scenario.
View and download← Back to Sentinel
Sentinel
Sentinel Agents
Ready-made agents you can download and import into your own deployment. Each one is a persona and the skills it works from — nothing else.
What you are downloading
An agent package contains a profile — its instructions, which model to prefer, which built-in tools it expects — plus any skills it was written against. Some agents also ship their own tools, which are real Python. When one does, this page lists every tool by name, and the site refuses to serve a package carrying code it did not list. What you see on an agent's page is what is in the file.
Shipped tools arrive untrusted: every call pauses for your approval until you have read the code and marked it trusted, and that is true even in a chat set to approve tools automatically. MCP servers arrive disabled. No credentials or API keys travel in a package at all.
Built-in tools are separate again: the ones an agent lists come with it, because they are Sentinel's own rather than anything the package ships. Each is governed by your deployment's approval policy and the per-run sandbox — on a default install run_command asks before it runs — but the list travels with the profile, so read it before you use an agent.
Cartographer
engineeringExplains an unfamiliar codebase to someone who has to change it tomorrow — entry points first, one real trace, and where the surprises are.
View and downloadScribe
writingTurns a commit range into release notes people actually read — leading with what a reader can now do, not what was refactored.
View and downloadSurveyor
engineeringAnswers questions about a repository by measuring it — and ships the tool that does the measuring.
View and downloadTriage
supportTurns a vague bug report into one somebody can reproduce — asking only for the facts that are actually missing.
View and download