AgentSkillsCN

start

快速启动会话——无需完整初始化,即可加载最近的上下文。

SKILL.md
--- frontmatter
name: start
description: Quick session start — load recent context without a full prime.
user-invocable: true

Session Start — StakTrakr

Lightweight context loader for the start of a development session. Gets you up to speed in 30 seconds, not 5 minutes.

Do NOT read or re-analyze CLAUDE.md — it's already loaded as system context.

Phase 1: Sync Check

Detect if a cloud session (Claude Code web, Codex, GitHub) pushed changes we don't have locally:

  1. git fetch origin — update remote tracking refs
  2. git branch --show-current — confirm working branch
  3. git rev-list HEAD..origin/<branch> --count — count commits we're behind

If behind by 1+ commits:

  • Show what's incoming: git log --oneline HEAD..origin/<branch>
  • Check for uncommitted local work: git status --short
  • If clean working tree: ask the user to confirm, then git pull origin <branch>
  • If dirty working tree: warn that local changes exist, suggest git stash then pull, or manual merge
  • After pulling, note which files changed: git diff --stat HEAD~N..HEAD (where N = commits pulled)

If up to date: Continue silently.

Phase 2: Recent Git Activity

Run these commands (parallel where possible):

  1. git log --oneline -8 — last 8 commits for recent momentum
  2. git status --short — uncommitted work from a prior session
  3. git diff --stat — size of any uncommitted changes

Phase 3: Linear Quick Check

Query the StakTrakr team (ID: f876864d-ff80-4231-ae6c-a8e5cb69aca4) for active work:

  1. List In Progress issues (these are what we're likely continuing)
  2. List Todo issues (these are queued in the active sprint)
  3. For each In Progress issue, read the full description to understand current state

If no In Progress issues, skip to Phase 4.

Phase 4: Last Handoff

Search Memento for the most recent handoff. Use keyword search first (reliable for structured entities), then fall back to semantic search:

  1. Primarymcp__memento__search_nodes with query: "STAKTRAKR handoff"
    • This finds entities matching our naming convention: Handoff: STAKTRAKR-*
    • Sort by most recent (check TIMESTAMP observation)
  2. Fallback — If no results, try mcp__memento__semantic_search with query: "staktrakr handoff development session", limit: 3, min_similarity: 0.3
  3. If a handoff entity is found, read it with mcp__memento__open_nodes for full context
  4. If no handoff found, that's fine — Phase 2 and 3 provide sufficient context

Key handoff fields to surface:

  • ACTIVE_SPRINT — current sprint project name and theme
  • SPRINT_ISSUES — issues in the active sprint
  • NEXT_STEPS — what to work on next
  • VERSION — current app version

Output

Produce a concise summary (not a formal report — just conversational):

code
Session Start — StakTrakr v[VERSION] on [BRANCH]

Recent: [1-2 sentence summary of last few commits]
Sprint: [SPRINT-NAME] — [issue list] or "No active sprint"
In Progress: [STAK-XX: title, STAK-YY: title] or "None"
Uncommitted: [X files changed] or "Clean"
Last handoff: [1 sentence from NEXT_STEPS] or "No handoff found"

Ready to work. What's next?

Keep it brief. The goal is awareness, not a research paper.