agent-device-lab
Prepare and check trusted macOS hosts for physical Android and iPhone tests with agent-device, ADB, and Xcode, or set up iPhone…
Prepare a concise session handoff when the user asks to wrap up, capture continuation context, or use /done. Retain a Markdown file only when requested.
$ npx -y skills add vincentkoc/dotskills --skill session-done --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/session-doneContext preview
The summary Claude sees to decide when to auto-load this skill.
Prepare a concise session handoff when the user asks to wrap up, capture continuation context, or use /done. Retain a Markdown file only when requested.
name: session-done description: Prepare a concise session handoff when the user asks to wrap up, capture continuation context, or use /done. Retain a Markdown file only when requested. license: MIT metadata: source: "https://github.com/vincentkoc/dotskills"
Capture the outcome, evidence, unresolved work, and next action without making session completion a cleanup or memory-write command.
1. Use the current task's evidence. Record completed work, validation and its limits, blockers, and the next owner/action. Do not scan unrelated sessions or copy credentials, private logs, or raw transcripts. 2. Include known session, branch, commit/PR, and retained-checkout identities when relevant. Mark unknown facts as unknown. Completion does not terminate sessions, remove worktrees, release owners, or rerun cleanup. 3. Return the handoff in chat by default. For deterministic Markdown, run `python3 scripts/session-done --summary "..." --state "..." --follow-ups "..."`; it writes only stdout. 4. If the user explicitly requested retention, use one named destination with `--output <new-path>`. The parent must already exist; existing files and symlinks are refused. A PKM/Obsidian folder can be that destination, without a second copy or sync service. 5. Report the created path or exact failure. A failed write is not saved evidence. Preserve existing notes and recovery material; do not retry under new filenames automatically. 6. Update memory only when the user directly requests it, through the environment's existing memory workflow. This writer never appends to shared memory or edits agent-managed memory files. Report that separate action's outcome.
stateDiagram-v2
[*] --> GatherCurrentEvidence
GatherCurrentEvidence --> PrepareHandoff
PrepareHandoff --> DeliverInChat: retention not requested
PrepareHandoff --> PublishNewFile: retention explicitly requested
PublishNewFile --> ReportSavedPath: destination new and write succeeds
PublishNewFile --> ReportFailure: collision or write failure
DeliverInChat --> [*]
ReportSavedPath --> [*]
ReportFailure --> [*]We are moving from "prompt as text" to skill as runtime module. Each skill acts like a lightweight, containerized application for AI work: a stable interface, opinionated workflow, and bundled resources that can be installed, versioned, tested, and reused
Prepare and check trusted macOS hosts for physical Android and iPhone tests with agent-device, ADB, and Xcode, or set up iPhone…
Resolve canonical Git checkouts, index and verify codebase-memory-mcp graphs through the guarded CLI, open its browser viewer on…
Mine structured Codex /goal history locally or across a configured machine fleet, measure active goal time and resumed thread…
Recover Codex or Claude sessions from tmux panes, saved cockpit snapshots, session history, and recent agent logs without…
Interpret Crabpot and OpenClaw performance dashboard metrics, import-loop profiles, runtime profiles, and branch-to-branch perf…
Maintain and release the crawlkit Go library, preserving downstream compatibility for gitcrawl, slacrawl, discrawl, and notcrawl.