A changelog that doesn't start from zero
CI and automationTeam generating release notes with an agent
Each release, the agent summarised every commit since the last tag, including ones it had already described in the previous draft.
How do I make my release notes agent summarise only the commits since the last release?
Save the last release's summary with its tag in ArcRoom. The next run fetches it with find_task and summarises only the new commits, and wording decisions, like what counts as breaking, come back as decisions. This needs the previous run to have saved its result.
Also asked as: Our release notes agent re-summarises commits it already covered. How do we stop that? · How do I reuse the previous release notes draft in an automated changelog? · Can an AI agent remember which commits it already described?
The last release's summary is saved with its tag.
- Reused answer
- Decision
find_task → reuse: "release notes v0.0.3" · summarise only commits after v0.0.3
Step by step
- The last release's summary is saved with its tag.
- The next run fetches it and summarises only the new commits.
- Wording decisions (what counts as 'breaking') come back as decisions.
What the agent sees
find_task → reuse: "release notes v0.0.3" · summarise only commits after v0.0.3Compared with today's tools
git log since the last tag in the prompt
The workflow passes git log v0.0.3..HEAD and the existing CHANGELOG.md to the agent, so it only sees new commits.
- Setup
- 20 min
- Ongoing
- None once scripted
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 2 min (npx arcroom init)
- Ongoing
- None
This works just as well for the commit range. ArcRoom adds the saved wording decisions, such as what counts as breaking, without restating them in the prompt.
Where it stops
Needs the previous run to have saved its result.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.