Developer Aide Suite¶
Kiket's Developer Aide Suite embeds Gemini 2.5 Flash-powered assistance directly into day-to-day engineering workflows. The suite currently ships four aides that can be invoked from the issue sidebar, project command palette, and Workflow Intelligence automations:
Feature Overview¶
| Aide | Purpose | Output |
|---|---|---|
| AI Code Review | Analyse pull requests for risky changes, performance regressions, and policy violations. | Inline review comments with severity, rationale, and suggested fixes. |
| Bug Scout | Triage production defects using issue text, logs, and recent activity. | Root-cause hypothesis, reproducible steps, and mitigation checklist stored with the issue. |
| Test Generator | Draft acceptance and regression tests from requirements and code context. | Framework-ready test cases with steps, expected results, and priority. |
| Estimator | Produce starting story point estimates with supporting rationale. | Suggested points, confidence level, and comparable work references. |
Each aide records usage in audit logs, respects organization data-scope settings, and captures human feedback for continuous prompt tuning.
Getting Started¶
- Open any project issue and locate the Developer Aide Suite panel in the right sidebar.
- Choose Generate Bug Analysis, Draft Test Cases, or Suggest Story Points to trigger the relevant aide.
- Review the output card, react with feedback, and use the category-specific actions (publish comments, attach tests, apply estimates) to push updates into your workflow.
- For code review, launch the AI Code Review command from the project command palette (⌘/Ctrl + K) while viewing a pull request. Comments remain in Kiket until a reviewer publishes them, at which point we post to GitHub.
Tip: Teams can run aides multiple times as context evolves; newest results appear at the top of the panel while older cards remain for reference.
Feedback Loop & Audit Trail¶
- Every aide invocation creates a
DeveloperAideOutputrecord linked to the issue or pull request and logs the call inAIAuditLog. - Users can submit quick feedback (thumbs up/down) which is stored in
AIInsightFeedbackand surfaced to evaluation dashboards. - Outputs include metadata such as model confidence, suggested severity, and contributing signals so reviewers can make fast follow-up decisions.
- Confidence below manifest thresholds automatically opens a
WorkflowApprovalassigned to project leads/managers so teams can gate AI suggestions.
Panel Lifecycle & Auto-Dismiss¶
- The aide panel keeps the three most recent outputs per category. When a fresh result is generated we automatically dismiss older cards for the same issue or pull request and mark them as superseded so the panel stays focused.
- Dismissed or published cards disappear from the sidebar after the next page refresh; teams can always retrieve the full history from Analytics → AI Usage → Developer Aide.
- If every card is cleared (published or dismissed) the sidebar collapses to a compact “Run an aide” prompt. It automatically expands again the next time a new result is delivered.
Category Actions¶
- Code Review – Inspect inline comments, then publish them to GitHub directly from Kiket or dismiss suggestions with a short rationale.
- Bug Scout – High severity findings notify assignees, project leads, and managers via in-app alerts for quick triage. When the aide is confident it also posts a structured Markdown comment with the root cause summary, reproduction steps, recommended actions, and related commits so teams don’t have to copy/paste the analysis.
- Test Generation – Attach curated scenarios to the issue as Markdown comments or download the generated patch stub.
- Estimation – Apply the AI suggestion (or a custom value) to the issue’s story points, and we log the decision for analytics.
Guardrails & Approvals¶
- Code review comments default to requires-review status when the agent reports confidence below 0.75. Approvals for auto-posting can be configured through Workflow Approvals (coming soon).
- Issue-based aides never mutate records automatically; teams decide when to apply AI suggestions.
- Gemini input payloads are filtered by the organization’s AI data scope (issues, telemetry, repositories, documents). Disable repository access to prevent code content from flowing to the model.
- Organizations on Professional plans and above can enable the Developer Aide Suite feature gate in Settings → AI Features. Other tiers see an upgrade prompt in the aide panel.
Extending the Suite¶
- Agent manifests live in
config/kiket/agents/*.ymland can be overridden in definition repositories. - Prompt templates under
config/kiket/prompts/*.ymlcontain the ERB bodies and output schemas. - Use the
developer_aidecommand palette actions as building blocks inside KiketScript automations to trigger aides from workflows, approvals, or extension hooks.
For manifest and prompt guidance see AI Agents & Prompts. Reach out to the Product Intelligence squad for early access to the telemetry and approval roadmap.