Agent Library · By platform · Review · Release · Docs
C
The first pass nobody has time for.
Code review in a small engineering team is either fast or thorough, rarely both. An agent that carries the mechanical pass — conventions, obvious omissions, missing tests, changelog impact — leaves human review for the part that needs judgement.
Review is the bottleneck
PRs wait on the two people who can review them, who are also the two people building.
Release notes are written last, badly
Under time pressure, by someone reconstructing a fortnight of commits.
Docs drift from code
The README describes an interface that changed three releases ago.
The agents
5 agents. Each one earns its autonomy.
Agent
What it does
Autonomy
Human gate
First-Pass Reviewer
Checks a PR against your own documented conventions and flags omissions before a human opens it.
Notify
Merge approval always human
Release Notes Drafter
Assembled from merged work in language a customer can read.
Review
—
Dependency Watch
Flags outdated and vulnerable dependencies with the actual blast radius, not a raw CVE list.
Flag only
—
Docs Drift Detector
Finds documentation that contradicts the code it describes.
Flag only
—
Changelog Curator
Keeps the changelog current and consistent as work merges.
Notify
—
Autonomy key — Auto: executes above 95% confidence. Notify: executes at 85–95% with notification and undo. Review: always queued for a human. Flag only: never acts, alerts a person.
What changes
Three shifts, not a feature list.
Human review starts from a cleaner PR
Mechanical issues are resolved before your senior engineer opens the diff.
Release notes stop being written from memory
They are assembled from what actually merged.
Drift gets flagged, not discovered
By the system, rather than by a confused new joiner.
The eval bench
What these agents have to pass before they touch real work.
- Review-flag precision measured against 100 historic PRs with known review outcomes
- Zero autonomous merges, approvals or branch operations
- Release note completeness checked against merged commits for the same period
Written against your data, not ours. The bench is built during Blueprint and runs on every deploy from then on.
Related
Where this connects.
See it on your GitHub.
A Blueprint maps which of these agents pays off first in your exact setup — and which to leave alone.
Apply for a Blueprint