Resume Last (golast) recovers the previous session's final actionable request or conversation context for the current workspace and presents it so the agent or user can continue seamlessly. It runs a bundled script to locate the most relevant last user request or assistant action and returns structured fields such as resolved_request, literal_last_user_message, supports, and ambiguity hints.
Use this skill only at the start of a fresh session when the user's first substantive message asks to continue the previous session, or when recovering after a crash or workspace compaction. It should NOT auto-trigger later in an already active session; instead the user must explicitly invoke golast. Typical triggers: the user types golast, pastes the skill link, or asks to recover the previous session.
scripts/resume_last_request.py (recommended invocation and JSON output).--cwd <current-workspace> --format json and read the output fields such as resolved_request, last_conversation_content, and needs_more_context. If needs_more_context is true, inspect supporting_context before continuing.Agents that can run Python helper scripts and use returned JSON outputs (Codex, Copilot-style agents that allow local script invocation, or OpenClaw/ClawFlow runtimes). The skill is primarily script-driven and designed for local agent runtimes that can access workspace files and session transcripts.
Session recovery skill for Codex/OpenAI CLI that parses the local state DB to find the last actionable request from a previous workspace thread. The script is well-written with comprehensive ambiguity resolution for both English and Chinese messages, agreement detection, and supplement merging. Script failed only because the Codex state database doesn't exist on this machine — expected in an audit environment, not a code defect.
No security concerns whatsoever. Script is read-only and operates entirely on local data. Very thorough ambiguity resolution logic covering low-signal messages, agreements, supplements, and reference resolution. Single-file script is ~500 lines — could be split into modules but perfectly functional as-is.