Version history
What changed in Skipper Agent.
New capabilities, reliability improvements, and important fixes in each
Skipper Agent release for classic Windows Outlook.
This release makes multi-message and voice-driven work safer, keeps Outlook
actions attached to the message that started them, and improves recovery when
an agent tool or local provider briefly fails.
- Safer simultaneous work: concurrent reply and forward requests stay isolated so one Outlook message cannot take over another message's draft, instructions, or conversation context.
- Message-aware conversation history: pane turns are scoped to the Outlook item that started them while still preserving useful global context across the conversation.
- Reliable message actions: attachment reading, reply and forward drafting, and contact, appointment, or task creation remain bound to the captured source message, including messages in shared, archive, and PST stores.
- Better dictated replies: explicitly dictated details remain authoritative, and Skipper keeps track of speech-transcribed text so it can interpret and polish it conservatively without dropping requested facts.
- More resilient agent runs: temporary tool-definition errors, malformed structured responses, attachment-name mistakes, and tool-loop exhaustion now produce useful recovery behavior instead of abruptly ending the task.
- Supported Codex CLI discovery: local Codex setup now prefers OpenAI's standalone Windows installation and reports the selected executable and version more clearly.
This release added a safe way to move profile and learning data between
Outlook installations, along with workflow, status, and local Codex
compatibility improvements.
- Profile and learnings transfer: export local profile guidance and learned preferences to a versioned JSON file, then review and import that information on another installation.
- Safer imports: imported content is validated and merged deliberately while preserving identity and excluding API keys and unrelated mailbox-memory data.
- Better Codex compatibility: model discovery and local execution recover from the service-tier configuration mismatch produced when an older standalone Codex CLI reads settings written by the unified ChatGPT app.
- Clearer task-pane activity: the pane activity lights remain active for the full chat run, and mailbox-learning rebuilds show the correct initial operation.
- More accurate reply categorization: the built-in received-message workflow now marks mail as needing a reply only when the user is actually participating and has a concrete response obligation.
This release focused on faster dictated replies and more reliable local Codex
reply handling.
- Faster dictated replies: when the user supplies the substantive reply wording, Skipper can create the native Outlook reply or forward without first running unnecessary grounding searches.
- More reliable local Codex replies: Skipper can recover the final assistant response from Codex CLI event output when the normal last-message file is missing, and existing-draft edits are less likely to be rejected because of a mislabeled response intent.