Daily briefings
AI Briefing: Repository memory, chat integrations and review metrics
Copilot’s security-fix memory preview, richer Slack and Teams context, and the limits of a new review-timing report.
Security
Agentic autofix can reuse Copilot Memory
For customers who enable Copilot Memory, agentic autofix can consult existing repository memories when resolving security alerts and save a fix pattern for later use. GitHub says those memories can also inform other Copilot features. Both autofix and Memory remain public previews. A stored fix pattern is context for later work, not proof that a new alert has been resolved correctly.
Why it matters
Review fixes against the current code and alert. Reusing a pattern can save investigation time, but the surrounding dependencies and security conditions may have changed.
Source: GitHub · Source published:
Collaboration
Copilot expands context in Slack and Microsoft Teams
GitHub added support for more conversation context, including supported Slack files and message links, and Teams images, forwarded messages and thread history. It also describes model switching and better links between discussions and GitHub work. The integrations remain public previews for Business and Enterprise organizations, with gradual rollout and administrator-controlled access.
Why it matters
Before connecting a team conversation to coding work, verify which files and messages the integration can use. Traceable links help reviewers understand why a change was requested.
Source: GitHub · Source published:
Measurement
Copilot’s usage API separates three human review stages
Repository-level usage reports now provide median and 90th-percentile durations for waiting for the first review, moving through review, and waiting to merge after final review. This release times qualifying human reviews; bot reviews are excluded. GitHub also states that historical data is not backfilled and a day with no qualifying merged requests returns an empty array.
Why it matters
The report can locate a review bottleneck, but it is not a direct measure of AI review quality or productivity. Preserve the distinction between missing observations and measured zeros.
Source: GitHub · Source published: