Agents Now Get Clear Instructions For Each Run
Every agent run now receives specific instructions for what that run is for, fixing reviews, rework and retries that used to start with too little context.
Summary
An agent run needs to know more than just its task — it needs to know what this specific run is for, whether that's reviewing a pull request, reworking a rejected change, or retrying after a failure. A gap in how those run-specific instructions reached the agent meant some runs, especially reviews, were starting with far too little context to do their job. That's fixed: every run type now gets the detail it actually needs.
What You Gain
Previously, a run's specific purpose and context weren't reliably reaching the agent doing the work. The most visible effect was reviews: a reviewing agent could be started with essentially nothing but a task reference, which is part of why reviews so often ended without a real verdict. Rework runs, automatic retries, and coordination steps on larger multi-part tasks had similar gaps, each missing a piece of context that would have helped them succeed the first time.
Now each run type gets exactly what it needs: a review gets the pull request and commit to examine and how to submit its verdict; a rework run gets the previous reviewer's findings; a retry gets a diagnosis of what went wrong last time. Chat and Slack-driven agent work now also reliably receives the message or step that started it.
How It Works
- Reviews receive the pull request, the exact commit, and how to submit a verdict.
- Rework runs receive the previous review's findings, so the agent knows what to fix.
- Retries receive a diagnosis of the failed attempt rather than starting cold.
- Multi-part tasks get the right coordination context at each stage.
- Chat and Slack-driven work now reliably reaches the agent with the message that triggered it.
- Agents are now told how to flag a question that needs a person's attention.
- "View prompt" on a task now shows exactly what the agent was told.
Where to Find It
This applies automatically to every run across every project. Keeping the desktop app and CLI current ensures every run type gets the full benefit.
See what else shipped that month on the Effigate Command Center changelog, or try Effigate Command Center with your own team.