Everything we know
about running agents
against a real backlog.
Guides for the setup, patterns from real pilots, and the engineering behind the run — plus the templates we hand every new team.
Featured guide
Writing acceptance criteria an agent can actually implement
The difference between a ticket a run nails on the first pass and one it flails on is almost always the acceptance criteria. Here is the shape that works, with examples.
14 min · updated July 2026
START HERE
015 min
Connect your credentials
Tracker and repository credentials, tested as you save them. Agent infrastructure needs no separate key. Optional: Confluence or Notion credentials for Narratia runbook push.
022 min
Bind your first project
One tracker project to one repository, validated live.
031 min
Sync and read the board
The hierarchy arrives; check the parents look right.
044 min
Run one small item
Pick something mechanical for the first run, not the payments rewrite.
05Your call
Review the pull request
Normal review, normal CI. Nothing merges itself.
062 min
Run Aegis after commit
Click Run Aegis on the committed run. Review findings, push Medium/Low to your tracker, or hit Remediate Now for High findings — the loop closes in the same session.
Everything
GUIDE9 min
Connecting Jira without over-scoping the token
The three Jira permissions Blue Mantis needs, and the four it will never ask for.
Read →
GUIDE12 min
From epic to eight children in one breakdown
How to review an AI breakdown quickly: what to accept, what to rewrite, what to delete outright.
Read →
PATTERN6 min
Refinement prompts that survive code review
Short, repository-specific instructions beat long style essays. Nine examples with their diffs.
Read →
PATTERN7 min
When to switch auto-commit on
A rule of thumb: auto-commit for mechanical work, review-first for anything touching money or auth.
Read →
ENGINEERING11 min
How Synthesia ranks two answers
How Synthesia scores Raptia and Fovea on stack fit, blast radius, and AC coverage — and why the second answer sometimes wins.
Read →
ENGINEERING8 min
Scheduling, dispatch and the five-minute sweep
What happens between pressing Schedule and finding a pull request the next morning.
Read →
TEMPLATEDownload
Acceptance-criteria template for agent runs
A Given/When/Then skeleton that maps cleanly onto generated tests.
Get it →
TEMPLATEDownload
Pilot checklist for the first two weeks
What to instrument, which items to point it at, and how to tell whether it is working.
Get it →
POSTMORTEM10 min
Runs that failed, and why
Expired tokens, ambiguous tickets, a rebase that ate a commit. What we changed after each.
Read →
GUIDE8 min
Setting up Aegis stop gates on GitHub
One branch protection rule on main. How to configure it, what Aegis posts, and what happens when a High finding lands.
Read →
PATTERN6 min
Remediate Now: closing the security loop
From Aegis finding to remediation run in one click. When to use it, when to push to the tracker instead, and how the refinement prompt is pre-filled.
Read →
GUIDE7 min
Generating runbooks with Narratia
What goes into a Narratia runbook, how to configure the Confluence and Notion push, and why Markdown to the PR branch is the best starting point.
Read →
Changelog
2026-08-07
Aegis remediation loop — Remediate Now
Security findings can now be pushed to your tracker or fixed immediately with Remediate Now — one click creates the ticket, syncs the board, and starts a new run.
2026-08-05
Narratia runbook generation
After every committed run, Narratia generates an operational runbook and pushes it to Confluence, Notion, or the PR branch as Markdown.
2026-08-03
Aegis security agent with stop gate
Aegis scans committed code for OWASP Top 10 vulnerabilities. High findings block the PR via a GitHub status check until resolved.
2026-08-01
Stack-aware generation for all major frameworks
Raptia and Fovea now detect your stack — React, Node, .NET, Java Spring Boot, Python, Go — and write idiomatic code for each without being told.
2026-07-24
Test scripts stack on the open PR
The generated script now commits onto the branch head instead of the default branch.
2026-07-11
Scheduled runs email on failure
Failures carry the reason and a link straight to the run.
2026-06-29
Epic breakdown is editable before it saves
Proposals arrive as rows you can rewrite, retype or delete.
2026-06-15
Duplicate project bindings return a friendly 409
With a link to the project that already holds the binding.
Delivery notes, monthly
One email a month: what shipped, what we learned from the runs that failed, and the patterns worth stealing.
No product announcements dressed up as research. Unsubscribe in one click.