What reviewers see
The PR shows the result.
A diff. A commit message. Maybe a short description. Nothing about how the code came to exist.
Understanding agent-written code
Review agent-written code
with the context intact.
Agents are shipping more of the code in your repo — and the PR alone no longer explains how it got there. LaserOwl captures the working session behind a change: what was asked, how the agent worked, what it tried, what it rejected, and what shipped. So your team can understand the code, not just the diff.
What’s insideWhat was askedHow the agent workedWhat it rejectedWhat it shipped
LaserOwl/sessions/abigail
live
AB
add to the conversation…
Live session preview
The problem
The PR shows the result. Not the session that produced it.
When a human wrote the code, the PR and the author were the record. When an agent writes the code, most of the reasoning happens in a session that nobody keeps.
What gets dropped
Not the session that produced it.
The intent, tool calls, attempts and rejected paths lived in an agent session — and closed when the PR opened.
What it costs
Reviewers inherit code they can’t interrogate.
“Why this approach?” “What else did it try?” The answers aren’t in the diff, and the author didn’t make every call.
What a session captures
The context behind a change.
Six pieces, captured as the agent works. Together they turn the session into something a reviewer, debugger, or auditor can actually read.
Intent
What the author asked the agent to do.
Tool activity
Every tool call, file read, and edit.
Attempts & rejections
What the agent tried, and what it backed out of.
Commit linkage
Every commit the session produced — linked to the PR.
Discussion
Anchored comments from humans and agents.
Outcome
How the session closed — merged, reverted, abandoned.
Why teams care
Review. Debug. Audit.
The same session record solves three problems teams already have with agent-written code. Each one is a workflow that pays rent today — not a promise about the future.
01 · Review
Review with context
Open a PR and see the session that produced it — intent, tools, attempts, rejected paths. Reviewers stop guessing why the code looks the way it does.
02 · Debug
Debug with provenance
When an agent-written change breaks later, the trail back to the intent is still there. Root cause in minutes of reading, not an afternoon of archaeology.
03 · Audit
Keep an audit trail
A durable, queryable record of how each change was produced. Enough to answer an internal review board — or a customer asking how an AI-assisted change came to exist.
Shared context for future agents
DownstreamSessions are machine-readable, so the next agent can read what prior agents tried, rejected, and shipped. Secondary for now — a natural follow-on once the record exists.
An example
A review conversation, with the context intact.
A reviewer questions a rate-limit value on an agent-authored PR. Because the session is linked to the commit, the author, reviewer, and the agent itself can all speak to the same record.
Discussion·anchored to commit cfa9e2
3 repliesM
Maya
reviewer · 14:14Token bucket for /v1/context · commit cfa9e2
Are we sure 50 rps is the right ceiling? Our analytics showed 38 rps p95 last week but there are bursts.
C
Claude Code
agent · 14:15agentThe intent said “cap around 50 rps”, so I kept headroom above p95. I tried 40 first and backed out — not enough slack for the burst window. Happy to tighten to 45 if you want a follow-up session.
A
Abigail
author · 14:16Good — that’s the reasoning I needed to see. Leaving it at 50 and watching for a week. Merging.
How it fits
Alongside GitHub, not in place of it.
LaserOwl sits quietly next to the tools your team already uses. It captures the session behind each change and links it back to the commits and PRs in GitHub.
Your agent workflow stays the same.
Developers keep using the IDE, CLI, and agents they already work with. LaserOwl listens in — it doesn’t get in the way.
PRs stay in GitHub.
Code review still happens where it happens. Sessions link back to the PR, so reviewers can open the context without switching tools.
LaserOwl adds the missing layer.
Every commit links to the session that produced it. Every session links to the commits it shipped. Two-way, durable, searchable.
Session ⇄ commitTwo-way link
LaserOwl session
sx-9f2a · Rate-limit /v1/contextauthoring agent · claude-codeoutcome · merged
- intent · tools · rejections · commits · discussion
- the working record behind the change
GitHub pull request
PR #1208commit cfa9e2merged by abigail
- diff · CI · review · merge commit
- source of truth for code — unchanged by LaserOwl
Working with a few teams
If this problem is live for you, we’d like to talk.
Speaking with a small number of engineering teams already dealing with agent-authored code in real workflows. Validating where session-based context actually helps with review, debugging, and traceability.