<< All versions

Skill v1.0.0

currentTrusted Publisher100/100
microsoft/agents/sync-teams-dotnet-samples-to-agents-sdk
──Details
PublishedOctober 2, 2026 at 01:25 PM
Content Hashsha256:686a7d71c6561ace...
Git SHA
──Files
Files (1 file, 2.8 KB)
SKILL.md2.8 KBactive
SKILL.md · 43 lines · 2.8 KB

version: "1.0.0" name: sync-teams-dotnet-samples-to-agents-sdk description: CI instructions for a bounded Teams SDK .NET to Agents SDK sample synchronization.


Teams sample synchronization

This CI caller uses one persistent implementation session. Preserve the source behavior with the smallest necessary changes inside the selected sample. Invoke the public migration skill before editing code and the manifest skill when updating the Teams app package. The sync context records minimumAgentsSdkVersion as a floor and selectedAgentsSdkVersion as this run's resolved release line. Plan and implement with the selected line; do not copy the minimum into project references. Keep the three required Agents SDK packages on one major.minor line, allowing their published patch versions to differ.

The first turn is read-only planning. Before finalizing the plan, reconcile every user-facing capability across source handlers, README/help text, the destination implementation, and the actual destination manifest JSON. For commands, record the exact title, slash or mention surface, handler evidence, and whether it exists under bots[].commandLists. Treat documentation as a requirement to verify, not proof of the manifest's current contents. A claim that no manifest change is needed must cite the relevant JSON path and current value.

Return a concise Markdown specification with an item for each relevant source behavior: upstream path and stable symbol, destination change, Agents-specific behavior to preserve, manifest impact, and validation expectation. Include every mismatch found during capability reconciliation. The coordinator freezes that plan before it grants write access.

After write access is granted, implement the frozen plan without editing it. The only CI tool is validate_sample; use code during implementation, manifest when editing the package, and all before completing the task. A validation result is authoritative only for the current candidate. Add meaningful selected-sample tests when the behavior can be tested locally.

Before full validation, re-open the final implementation, README/help text, and manifest. Compare them against the capability inventory by exact command title and surface; schema validation alone does not prove capability coverage. Finish by returning a Markdown self-audit that accounts for every frozen-plan item and names the validation run. Start it with ## Self-audit, then write Outcome: changed or Outcome: no changes required on its own line. A no-changes outcome requires evidence that every plan item was verified and that no selected-sample file needs modification. If a required behavior cannot be implemented, state the actual blocker. Do not claim validation passed when it did not. The coordinator may request one repair for deterministic validation failures; keep the frozen plan unchanged during that repair.

All versions